Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
A multi-site neuroimaging approach to studying hippocampal damage in chronic stroke
(USC Thesis Other)
A multi-site neuroimaging approach to studying hippocampal damage in chronic stroke
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
Copyright 2021 Artemis Zavaliangos-Petropulu
A multi-site neuroimaging approach to studying hippocampal damage in chronic stroke
by
Artemis Zavaliangos-Petropulu
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the Requirements for the Degree of
DOCTOR OF PHILOSOPHY
(NEUROSCIENCE)
December 2021
ii
Acknowledgements
This project was funded by NIH grant U54 EB020403, NIH K01 HD091283, and
NIH R01 NS115845. I would like to thank the members of the ENIGMA Stroke Recovery
Working Group and all the participants and their families for their contributions to this
study.
I would like to express my most sincere gratitude to my friends, professors,
mentors, and colleagues for their advice, support and understanding throughout my
graduate studies. The saying “it takes a village” has never felt more true.
I would like to thank to my primary graduate mentors, Dr. Paul Thompson and Dr.
Sook-Lei Liew, for their guidance and support over the years. I am extremely grateful to
Dr. Paul Thompson, for giving me my first job after college and inspiring me to pursue a
PhD at USC. Thank you for guidance over the years and for introducing me to the world
of neuroimaging. It has been truly an honor to be mentored by one of neuroimaging’s
“greats”. I am also extremely grateful to Dr. Sook-Lei Liew, who continuously pushes me
to be a better researcher and has led by example for what it means to be a good, caring
mentor. Thank you for taking me along on the adventure of starting up ENIGMA Stroke
Recovery - it has been truly an incredible experience.
I would like to acknowledge the other members of my dissertation committee. Dr.
Nina Bradley, your kindness, enthusiasm, and approach to teaching is unparalleled. You
have sparked a passion for neuroanatomy that I will chase for the rest of my career. I am
so grateful to have you as a mentor. Also thank you to Dr. Schweighofer, for providing me
with guidance and for encouraging me to “think as a neuroscientist”. I would also like to
iii
thank my guidance committee members, Neda Jahanshad and Greg ver Steeg for their
guidance and ideas that greatly shaped this dissertation.
I would like to acknowledge my lab mates at the Neural Plasticity and Neural
Rehabilitation Laboratory for their friendship and support. I would especially like to thank
the other two lab members of the ENIGMA Stroke Recovery team, Bethany and Miranda.
Maintaining this project is not easy, but your enthusiasm, organization, willingness to
learn has not only made my dissertation work possible, but it has made it enjoyable.
None of this would be possible without the support of my lab mates over the years
at the Imaging Genetics Center, whose continued friendships and support reach far
beyond the workplace and have kept me sane: Chris, Sophia, Talia, Dan, Priya, Derrek,
Sarah, Meral, and Adam.
I offer special thanks to my parents Athina and Antonios, my brother Takis, my
grandparents Ioannis, Artemis, Maria, and Panagiotis, my cousin Christo, my extended
family in Greece and Boston, and Moose. Thank you for believing in me and always
pushing me to do my best. Σας αγαπώ απεριόριστα.
Finally, I offer special thanks to Chris for his endless support and encouragement.
Thank you for feeding me cheese during the highs and feeding me extra cheese during
the lows. This whole journey would not have been nearly as much fun without you.
iv
Table of Contents
Acknowledgements .......................................................................................................... ii
List of Figures ................................................................................................................. vii
List of Tables .................................................................................................................. viii
Chapter 1: Introduction.................................................................................................... 1
1.1 Introduction ............................................................................................................. 1
Organization of the Dissertation ................................................................................. 3
1.2 ENIGMA Stroke Recovery ...................................................................................... 6
1.2.1 ENIGMA Overview ............................................................................................. 6
1.2.2 ENIGMA Quality Control .................................................................................... 9
1.2.3 Lesion Masks ................................................................................................... 12
1.2.4 ENIGMA Stroke Subcortical Findings ............................................................. 12
1.3 Aim One Background ........................................................................................... 15
1.3.1 Overview .......................................................................................................... 15
1.3.2 FreeSurfer ........................................................................................................ 16
1.3.3 Hippodeep ........................................................................................................ 20
1.3.4 Aim One Purpose ............................................................................................. 20
1.3 Aim Two Background ........................................................................................... 21
1.3.1 Mechanisms of Stroke-related Damage .......................................................... 21
1.3.2 Post-stroke Secondary Degeneration of the Hippocampus ........................... 22
1.3.3 Post-stroke Hippocampal Studies ................................................................... 23
1.3.4 Associations Between Hippocampal Volume and Lesion Size. ..................... 24
1.3.5 Aim Two Purpose ............................................................................................. 24
1.4 Aim Three Background ........................................................................................ 25
1.4.1 Sensorimotor Impairment ................................................................................ 25
1.4.2 Role of the Hippocampus in Sensorimotor Impairment .................................. 26
1.4.3 Sex Differences ................................................................................................ 27
1.4.4 Aim Three Purpose .......................................................................................... 28
Chapter 2: Assessing automated hippocampal segmentation methods in a stroke
population........................................................................................................................ 30
2.1 Abstract.................................................................................................................. 30
2.2 Introduction ........................................................................................................... 31
2.3 Methods ................................................................................................................. 35
2.3.1 Data Acquisition ............................................................................................... 35
2.3.2. Hippocampal Segmentation Methods ............................................................ 35
2.3.3. Analyses .......................................................................................................... 39
2.4 Results ................................................................................................................... 43
2.4.1. Quality Control ................................................................................................ 43
2.4.2. Accuracy.......................................................................................................... 44
v
2.5 Discussion ............................................................................................................. 47
2.6 Limitations ............................................................................................................. 51
2.7 Conclusion ............................................................................................................ 52
Chapter 3: Associations between lesion size and post-stroke hippocampal
volume in patients with chronic stroke ....................................................................... 54
3.1 Abstract.................................................................................................................. 54
3.2 Introduction ........................................................................................................... 55
3.3 Methods ................................................................................................................. 59
3.3.1 ENIGMA Stroke Recovery Dataset ................................................................. 59
3.3.2 MRI Data Analysis ........................................................................................... 61
3.3.3 Manually Segmented Lesions ......................................................................... 62
3.3.4 Statistical Analysis ........................................................................................... 64
3.4 Results ................................................................................................................... 67
3.4.1 Lesion Effects on Hippocampal Volume ......................................................... 67
3.4.2 Hippocampal Volume in Participants with Stroke versus Healthy Controls ... 70
3.5 Discussion ............................................................................................................. 71
3.6 Limitations ............................................................................................................. 74
3.7 Conclusion ............................................................................................................ 75
Chapter 4: Associations between sensorimotor impairment and hippocampal
volume in chronic stroke survivors ............................................................................. 76
4.1 Abstract.................................................................................................................. 76
4.2. Introduction .......................................................................................................... 77
4.3 Methods ................................................................................................................. 81
4.3.1 ENIGMA Stroke Recovery Dataset ................................................................. 81
4.3.2 MRI Data Analysis ........................................................................................... 83
4.3.3 Manually Segmented Lesions ......................................................................... 83
4.3.4 Statistical Analysis ........................................................................................... 84
4.4 Results ................................................................................................................... 85
4.4.1 Hippocampal Volume and Sensorimotor Impairment ..................................... 85
4.4.2 Sex Effects on the Association between Hippocampal Volume and
Sensorimotor Impairment.......................................................................................... 91
4.5 Discussion ............................................................................................................. 92
4.6 Limitations and Future Directions ...................................................................... 94
4.7 Conclusion ............................................................................................................ 96
Chapter 5: Discussion ................................................................................................... 97
5.1 Summary of Key Findings ................................................................................... 97
5.2.1 Chapter 2 Summary - Aim One ....................................................................... 97
5.2.2 Chapter 3 Summary - Aim Two ....................................................................... 98
vi
5.2.3 Chapter 4 Summary - Aim Three .................................................................... 99
5.3 Implications and Significance ............................................................................. 99
5.3.1 Clinical Implications ......................................................................................... 99
5.3.2 Research Implications ................................................................................... 100
5.4 Future Directions ................................................................................................ 102
5.4.1 Hippocampal Subfields .................................................................................. 102
5.4.2 Hippocampal White Matter Pathways ........................................................... 103
5.4.3 Longitudinal Changes to Hippocampal Volume ............................................ 103
5.4.4 Hippocampal Damage and Post-Stroke Cognitive Impairment, Depression,
and Anxiety .............................................................................................................. 104
References .................................................................................................................... 106
Appendix A .................................................................................................................... 125
Appendix B .................................................................................................................... 127
vii
List of Figures
Figure 1.1 ENIGMA Stroke Recovery Workflow ............................................................... 7
Figure 1.2 ENIGMA Stroke Recovery Map ....................................................................... 8
Figure 1.3 FreeSurfer Quality Control Protocol .............................................................. 11
Figure 1.4 Subcortical Findings ....................................................................................... 14
Figure 1.5 Desikan-Killiany Atlas .................................................................................... 17
Figure 1.6 Hippocampal Subfields Atlas ......................................................................... 18
Figure 1.7 Hydrocephalus Ex Vacuo .............................................................................. 19
Figure 1.8 Spreading Depression. .................................................................................. 23
Figure 2.1 Hippocampal Segmentation Quality Control ................................................. 40
Figure 2.2 Hippocampal Segmentation Quality Control Results .................................... 41
Figure 2.3 Automated Hippocampal Segmentations ...................................................... 45
Figure 2.4 Comparing Segmentation Accuracy .............................................................. 46
Figure 3.1 Lesion Density Maps for Primary Lesions ..................................................... 64
Figure 4.1.Sensorimotor Impairment and Hippocampal Volume ................................... 88
Figure 4.2. Sensorimotor Impairment*Sex Interaction Findings. ................................... 92
viii
List of Tables
Table 2.1 Accuracy Compared to Manual Segmentations ............................................. 47
Table 2.2 Accuracy Compared Across Automated Methods .......................................... 47
Table 2.3 Pros and Cons of Each Segmentation Method .............................................. 51
Table 3.1 Demographics for ENIGMA Stroke Recovery Working Group ....................... 60
Table 3.2 Demographics for Stroke vs Control Demographics ...................................... 61
Table 3.3 Lesion Size and Hippocampal Volume ........................................................... 68
Table 4.1 ENIGMA Stroke Recovery Demographics ...................................................... 82
Table 4.2.Sensorimotor Impairment and Hippocampal Volume. .................................... 87
Table 4.3 Sensorimotor Impairment*Sex and Hippocampal Volume. ............................ 89
Table 4.4 Sensorimotor Impairment and Hippocampal Volume and Lesion Size ......... 90
ix
Abstract
The decrease in stroke mortality has led to a growing population of people in need
of rehabilitation. To help clinicians and caregivers make informed decisions about
rehabilitation planning, there is a critical need to identify reliable biomarkers of stroke
outcomes that help make accurate predictions of a patient’s potential to recovery.
However, the generalizability of stroke recovery biomarkers must be tested in diverse
patient populations to improve the probability of successful use in the clinic. This has led
to an increasing interest in multi-site research consortia to obtain large, diverse patient
samples. One such consortium is ENIGMA Stroke Recovery. ENIGMA Stroke Recovery
is an international multi-site collaboration of stroke researchers dedicated to providing a
reliable infrastructure for the collection and analyses of large, diverse datasets of
poststroke brain magnetic resonance imaging (MRI) and behavioral measures that can
be used to identify robust, reproducible biomarkers of stroke outcomes. In this
dissertation, I discuss my contributions to ENIGMA Stroke Recovery and walk through
the curation of MRI data for the hippocampus- a brain region particularly vulnerable to
stroke related secondary degeneration and thought to be a promising biomarker of stroke
outcomes. The overarching goal of this work was to set the foundation for future large-
scale research of post-stroke hippocampus by using ENIGMA Stroke Recovery data to
1) identify a robust automated hippocampal segmentation method, 2) investigate the
association between lesion size and post-stroke hippocampal volume, and 3) explore the
association between post-stroke sensorimotor impairment and hippocampal damage.
1
Chapter 1: Introduction
1.1 Introduction
Stroke annually affects over 800,000 people in the United States alone and
continues to be a major worldwide public health concern (Virani et al., 2020). Advances
in medicine have improved the stroke survival rate to an estimated 80% (Gittler and Davis,
2018). However, this decrease in stroke mortality has led to a growing population of
people in need of rehabilitation (Johnson et al., 2019; Mozaffarian et al., 2016; Virani et
al., 2020). For clinicians, caregivers, and patients to make informed decisions about
planning rehabilitation treatment, accurate predictions of a patient’s potential for recovery
are necessary (Boyd et al., 2017). Therefore, there is a critical need to identify reliable
biomarkers of stroke outcomes (Stinear, 2017). Unfortunately, this is very difficult, as no
one stroke is alike. In fact, the term stroke only truly describes the cerebrovascular
ischemic incident, but does not detail the causal underlying medical condition, nor does
it justly describe the extent of the brain injury, the wide range of resulting symptoms, or a
patient’s response to rehabilitation (Boyd et al., 2017). The inherent heterogeneity of
stroke presents a significant challenge to stroke rehabilitation research (Stinear, Lang,
Zeiler, & Byblow, 2020) and likely contributes to the “translational roadblock” of stroke
research. Despite success of thousands of preclinical stroke recovery studies, very few
studies successfully translate to the clinic (Endres et al., 2008).
Testing the generalizability of stroke recovery biomarkers in diverse patient
populations is crucial to identifying effective stroke rehabilitation programs (Bernhardt et
al., 2017, 2019; Borschmann et al., 2018). This has incentivized large multicenter
2
worldwide consortia to obtain large samples of the diverse post-stroke populations and
test the reproducibility and generalizability of prior research. One example of these
consortia is ENIGMA Stroke Recovery. ENIGMA Stroke Recovery is an international
multi-site collaboration of stroke researchers focused on providing a reliable infrastructure
for the collection and analyses of large, diverse datasets of post-stroke brain magnetic
resonance imaging (MRI) and behavioral measures (Liew et al., 2020). Throughout my
graduate studies, I have contributed to building a centralized database of retrospective
neuroimaging stroke data for ENIGMA Stroke Recovery, providing a resource that acts
as a foundation for research in MRI derived biomarkers of stroke recovery at a large-
scale.
MRI can be used to quantify information about the lesion as well as provide insight
on the structure and function of non-lesioned regions (Boyd et al., 2017). The curation of
MRI derived biomarkers of stroke recovery is particularly challenging, given the significant
changes in spatial distribution of anatomical landmarks that occur after stroke. The more
commonly used automated segmentation methods used to measure regional brain
volumes in an MRI were not designed to accommodate significant brain injury (Irimia et
al., 2011), and have a tendency to underperform in cases of severe stroke pathology
(Liew et al., 2020; Zavaliangos ‐Petropulu et al., 2020). With the ENIGMA Stroke
Recovery team, I have developed a robust standardized protocol of assessing
segmentation quality specific to stroke pathology and have scrutinized the performance
of novel automated segmentation tools in an effort to have accurate measures of regional
brain volumes in our stroke recovery database. Additionally, despite efforts to develop
automated lesion segmentation methods, manual tracing remains the gold standard for
lesion segmentation (Ito et al., 2019). For the past two years, I have overseen the
3
ENIGMA Stroke Recovery efforts to manually segment stroke lesions on T1w MRI images
for the entire database.
Organization of the Dissertation
In this dissertation, I walk through the curation of MRI data for the hippocampus -
a region particularly susceptible to post-stroke secondary degeneration (Brodtmann et
al., 2020; Haque et al., 2019; Khlif et al., 2019a; Schaapsmeerders et al., 2015) - in
participants of the ENIGMA Stroke Recovery database with chronic stroke. Additionally,
we use the resulting large sample to investigate novel associations between hippocampal
volumes and stroke outcomes that may have been previously undetectable in
underpowered samples. Specifically, in this thesis, I aimed to:
1) Identify an automated hippocampal segmentation method that provides
accurate hippocampal segmentations within the context of stroke pathology.
We compared three publicly available automated hippocampal segmentation tools
- Hippodeep (Thyreau et al., 2018), FreeSurfer version 6.0 gross hippocampal
segmentation (Fischl et al., 2002a; Fischl, 2012), and FreeSurfer version 6.0 ‘sum
of subfields’ (Iglesias et al., 2015). We evaluated the performance of each method
in terms of quality control and accuracy when compared to manual segmentations.
In terms of accuracy, we found that all three automated segmentation methods
had good correlation with manual segmentations and no one method was
significantly more correlated than the others. However, we found that Hippodeep
was able to segment more hippocampi than the FreeSurfer tools and had the
lowest quality control fail rate. Overall, this study suggests that both Hippodeep
and FreeSurfer may be useful for hippocampal segmentation in stroke
4
rehabilitation research, but Hippodeep may be more robust to stroke lesion
anatomy.
2) Investigate associations between lesion size and hippocampal volume in
patients with chronic stroke. There is conflicting evidence regarding the impact
of stroke-related lesion size on hippocampal volume, with some studies reporting
no significant associations between lesion size and ipsilesional hippocampal
volume (Tang et al., 2012; Xie et al., 2011), and others reporting that larger lesion
size is significantly associated with smaller hippocampal volumes
(Schaapsmeerders et al., 2015). In this aim, we investigated whether lesion size
was independently associated with hippocampal volume, using a large sample of
brain MRI scans with manually segmented stroke lesions. Overall, we found that
larger lesion size is significantly associated with smaller ipsilesional but not
contralesional hippocampal volume, and that previous controversy in the literature
may have been underpowered.
3) Investigate the possible association between sensorimotor impairment and
hippocampal volume in patients with chronic stroke. The hippocampus is
particularly vulnerable to stroke pathology (Brodtmann et al., 2020; Haque et al.,
2019; Khlif et al., 2019a; Schaapsmeerders et al., 2015) and is involved in
sensorimotor circuits (Albouy et al., 2008; Baumgartner et al., 2018; Burman, 2019;
Jacobacci et al., 2020), but has not been widely studied within the context of stroke
sensorimotor impairment. In this aim, we test associations between sensorimotor
impairment and hippocampal volume in patients with chronic stroke. We performed
a follow up exploratory analysis to see if sex may moderate the relationship
5
between sensorimotor impairment and hippocampal damage, as greater stroke
severity (Pappa et al., 2012), more dementia-related hippocampal atrophy (Nebel
et al., 2018), and poorer stroke outcomes (Cordonnier et al., 2017; Dehlendorff et
al., 2015; Gittler and Davis, 2018) have been found in women. We found greater
sensorimotor impairment was significantly associated with smaller ipsilesional but
not contralesional hippocampal volume, independent of lesion size. We also found
preliminary evidence of a sensorimotor impairment by sex interaction for bilateral
hippocampal volumes, where women showed progressively smaller hippocampal
volumes with worsening sensorimotor impairment compared to men.
Overall, this work sets the foundation for future large-scale research of the post-
stroke hippocampus. This work is particularly relevant to the stroke rehabilitation field
given that a recent systematic review of research studies investigating biomarkers of
stroke recovery revealed that most studies are statistically underpowered (Kim and
Winstein, 2017). The work I present in this dissertation represents an international effort
to promote collaborative science with accessible and importantly, rigorously quality-
controlled data that can be used to study stroke rehabilitation and test the generalizability
of biomarkers of stroke outcomes. In the remainder of this chapter, I will provide an
overview of the ENIGMA Stroke Recovery efforts and present literature reviews that serve
as the background and rationale for each aim.
6
1.2 ENIGMA Stroke Recovery
1.2.1 ENIGMA Overview
The ENIGMA Stroke Recovery Working Group is just one of the many disease
working groups part of the ENIGMA Consortium. ENIGMA stands for Enhancing
Neuroimaging and Genetics through Meta-Analysis. The ENIGMA Consortium was
founded by Dr. Paul Thompson and colleagues in 2009 in an international collaborative
effort that brings together researchers in imaging genomics, neurology, and psychiatry to
understand brain structure and function based on multimodal neuroimaging and genetic
data across a range of patient populations (Thompson et al., 2020).
ENIGMA was conceived in an effort to address the crisis of reproducibility in
neuroimaging. The financial cost of collecting neuroimaging data makes it very difficult to
collect large samples of neuroimaging data at any one single institution, with studies often
sampling fewer than one hundred participants. The resulting small samples can lead to
underpowered research studies with findings that are not reproducible in independent
samples (Button et al., 2013; Munafò et al., 2017). The purpose of ENIGMA is to take
advantage of already collected data across multiple institutions to allow for larger and
more diverse samples when investigating structural brain alterations across disorders.
ENIGMA was initially designed for multi-site meta-analyses, with harmonized
preprocessing pipelines and statistical tools to allow researchers to preprocess data
locally and participate in some of the largest studies of psychiatric disorders such as major
depression (Schmaal et al., 2016), schizophrenia (Van Erp et al., 2016), and bipolar
disorder (Hibar et al., 2016), without needing to share their raw data.
7
Founded in 2015, ENIGMA Stroke Recovery has become one of the first ENIGMA
working groups to create a centralized database of raw neuroimaging data. ENIGMA
Stroke Recovery is dedicated to developing a reliable infrastructure for international data
collection and analysis of post-stroke MRI and behavioral data (Liew et al., 2020).
Throughout my PhD, I have helped develop the ENIGMA Stroke Recovery infrastructure
and processing pipelines (Figure 1.1) as well as curate incoming data for over 2,000 MRI
scans acquired across 40 research studies (Figure 1.2; Liew et al., 2020).
Figure 1.1 ENIGMA Stroke Recovery Workflow This diagram represents the
ENIGMA Stroke Recovery workflow, from data intake to data analysis (Liew et
al., 2020).
8
Briefly, incoming data is first visually inspected to ensure good MRI data quality.
The data is then reformatted to conform to the Brain Imaging Data Structure (BIDS), a
standardized naming convention for neuroimaging and behavioral data (Gorgolewski et
al., 2016). T1-weighted (T1w) MRI images are then run through FreeSurfer, a brain
imaging software developed to preprocess, segment, and label anatomical brain regions
(Fischl et al., 2002). FreeSurfer outputs volume and surface area estimates for both
regional cortical and subcortical areas. In the next subsections, I will discuss FreeSurfer
output quality control (1.2.2 ENIGMA Quality Control) and manual lesion segmentation
(1.2.3 Lesion Segmentation). Finally, the processed data is stored in an easily queryable
SQLite database.
Figure 1.2 ENIGMA Stroke Recovery Map The ENIGMA Stroke Recovery
consists of over 100 stroke rehabilitation researchers from 43 research centers
across 10 countries.
9
1.2.2 ENIGMA Quality Control
One of my first major contributions to ENIGMA Stroke Recovery was the quality
control protocol (QC) I developed for visually inspecting the quality of subcortical
segmentations in participants with stroke pathology. This protocol is based on an adaption
of the ENIGMA Cortical/Subcortical Quality Control protocol
(http://enigma.ini.usc.edu/protocols/imaging-protocols/). The initial ENIGMA protocol was
developed because, although automated segmentations perform relatively well in
neurotypical, healthy populations, we have found that an estimated 1-5% of
segmentations still fail to correctly capture the region of interest even without severe brain
pathology. In the ENIGMA Stroke Recovery stroke population, we have found that 10-
20% of segmentations fail to correctly capture the region of interest (Liew et al., 2020).
Stroke pathology can make it difficult for automated segmentation tools such as
FreeSurfer to correctly identify and measure brain regions. A lesion can make it
impossible to identify the boundaries of a brain structure that has been directly impacted,
therefore accurate measurements of brain regions can only be made for non-lesioned
areas. Additionally, lesions can introduce large alterations to the expected spatial
distribution of brain structures, presenting a significant challenge to FreeSurfer in
segmented regions adjacent to the lesion.
For the ENIGMA Stroke Recovery QC protocol, we adapted the existing ENIGMA
subcortical QC protocol to specifically account for segmentations that fail QC due to
stroke pathology. Expert raters manually inspect subcortical segmentations for every
participant. Screenshots of nine slices of the brain (three coronal, three axial, and three
sagittal) are generated with bilateral segmentations overlaid onto the T1w MRI (left
10
segmentation appears as transparent blue, right segmentation as transparent red) for
every subcortical region for each subject. These screenshots are compiled into eight
separate web-based php files (one for each subcortical region). These files are then used
to inspect segmentations for quality. Segmentations are marked as PASS, FAIL_FS when
FreeSurfer fails to output a segmentation entirely, FAIL_LESION_ADJACENT when the
segmentation of a region that is lesion adjacent does not pass quality control,
FAIL_LESION_DISTORTED when the lesion impacted the structure and resulted in a bad
segmentation, FAIL_LESION_ENCOMPASSED when the lesion impacted the structure
but did not distort the segmentation, or FAIL_OTHER for segmentation that over or
underestimated the region for a reason likely unrelated to stroke pathology (Figure 1.3).
The purpose of documenting the reason for why the segmentation failed is so that it may
be used for future studies of the lesioned areas.
11
Figure 1.3 FreeSurfer Quality Control Protocol Here, we show an example of the
platform used to perform quality control of subcortical segmentations measured with
FreeSurfer.
12
1.2.3 Lesion Masks
A lesion mask for every participant in the ENIGMA Stroke Recovery database is
manually drawn by expert raters on the T1w. Although there are automated and semi-
automated lesion segmentation tools in development for T1w MRI, manual segmentations
remain the gold standard (Ito et al., 2019). It is possible that the reason why automated
and semi-automated methods underperform is because they are trained and tested in
small datasets so they do not generalize. One of the major projects of ENIGMA Stroke
Recovery is to curate a large, public database of manually traced lesions that can be used
to train and test developing automated tools. The first release of this database was made
public in 2017, titled Anatomical Tracings of Lesions After Stroke (ATLAS; Liew et al.,
2017). ATLAS version 1.0 consists of 304 anonymized T1w MRI with manually traced
lesions. The second release of ATLAS (2.0) is upcoming.
We have a trained team of ENIGMA Stroke Recovery expert lesion tracers working
on manually segmenting lesions. Tracers identify brain lesions and manually draw lesion
masks on each individual brain in native space using ITK-Snap (Yushkevich et al., 2006).
Each lesion is checked for quality by at least two different tracers. An expert
neuroradiologist is also consulted to ensure lesion segmentation quality and to
differentiate stroke lesions from perivascular spaces. We aim to provide all ENIGMA
Stroke Recovery collaborators with manually traced lesions for their contributed data.
1.2.4 ENIGMA Stroke Subcortical Findings
Using the ENIGMA Stroke Recovery Pipeline (Liew et al., 2020), we published the
first ENIGMA Stroke Recovery analysis (currently in press), where we investigated
13
associations between non-lesioned subcortical volumes and post-stroke sensorimotor
outcomes. We found significant associations between post-stroke sensorimotor behavior
and smaller ipsilesional subcortical volumes that differed by time since stroke and
sensorimotor assessment (Figure 1.4; Liew et al., 2021).
14
Figure 1.4 Subcortical Findings Effect sizes for relationships between post-stroke
sensorimotor behavior and non-lesioned subcortical volumes are mapped on to a
template of subcortical volumes. Non-lesioned subcortical regions (1D, bottom right) that
relate to sensorimotor behavior from linear mixed-effects models of people with subacute
(1A, top left) and chronic (1B, bottom left) stroke. Non-lesioned subcortical volume
relationships with chronic sensorimotor impairment are shown in 1C (top right). Colors
represent the beta estimate (β) for sensorimotor behavior from each model. Warmer
colors represent stronger positive relationships (e.g., larger brain volumes relate to better
behavior), and cooler colors represent stronger negative relationships (e.g., larger brain
volumes relate to worse behavior) (Liew et al., 2021).
15
1.3 Aim One Background
1.3.1 Overview
Changes to the hippocampus are observed across a number of different
neurological and psychiatric disorders and in healthy ageing (Small et al., 2011). The
hippocampus is particularly vulnerable to post-stroke secondary degeneration, and
hippocampal volume is thought to be a promising biomarker of post-stroke cognitive
impairment (Schaapsmeerders et al., 2015). To measure hippocampal volume, accurate
segmentations are necessary. Currently, manual segmentations are arguably the gold
standard for analyzing hippocampal volume in MRI studies (Frisoni et al., 2015), but this
approach is extremely time consuming and not feasible for large datasets, such as the
ENIGMA Stroke Recovery database. Therefore, efforts to develop and test automated
hippocampal segmentation methods have been undertaken to provide a more efficient
way to study hippocampal volume on a large scale.
Recent studies by Khlif et al. (Khlif et al., 2019b, 2019a) tested the performance of
automated hippocampal segmentation methods in mild-moderate ischemic stroke
populations with lesions outside the hippocampus. They compared automated
hippocampal segmentation methods, including the gross hippocampal segmentation
available in FreeSurfer version 5.3, version 6.0, and the “sum of subfields” segmentation
available in FreeSurfer version 6.0. Khlif et al. (Khlif et al., 2019b, 2019a) found that the
FreeSurfer version 6.0 “sum of subfields” segmentation was among the most accurate
methods for estimating hippocampal volume in stroke populations.
16
1.3.2 FreeSurfer
FreeSurfer is an atlas-based software that employs a Bayesian statistical approach
to segment and label brain regions (Fischl, et al., 2002). It involves a series of data
preprocessing steps, such as intensity normalization, mapping of the input brain to a
probabilistic brain atlas, estimation of statistical distributions for the intensities of different
tissue classes, and labeling of cortical and subcortical structures based on known
information on the locations and adjacencies of specific brain substructures (Fischl et al.,
2002). FreeSurfer is widely used in the neuroimaging community because of its good
test–retest reliability across scanner manufacturers and field strengths (Han et al., 2006).
Additionally, FreeSurfer has an active and extensive archive of user questions for
troubleshooting and updated and improved versions are consistently being released.
One way to use FreeSurfer to estimate hippocampal volume is to use the output
from the Desikan-Killiany atlas (Desikan et al., 2006), which FreeSurfer uses to generate
cortical and subcortical segmentation (Figure 1.5). This method is the suggested method
for measuring cortical and subcortical volume by the ENIGMA Consortium
(http://enigma.ini.usc.edu/protocols/imaging-protocols/).
17
Figure 1.5 Desikan-Killiany Atlas FreeSurfer cortical and subcortical segmentations are
generated using the Desikan-Killiany atlas (https://surfer.nmr.mgh.harvard.edu/).
FreeSurfer version 6.0 can also output segmentations of 13 hippocampal subfields
using a refined probabilistic atlas (Iglesias et al., 2015). This atlas was built from a
combination of ultra-high resolution ex vivo and in vivo MRI scans to identify borders
between subfields of the hippocampus (Figure 1.6; Iglesias et al., 2015). The 13
outputted hippocampal subfields can be combined to create a segmentation of the entire
hippocampus, which is referred to as the “sum of subfields” segmentation.
18
Figure 1.6 Hippocampal Subfields Atlas Hippocampal subfields segmentations
generated using FreeSurfer version 6.0 Subfields can be combined to create a gross
hippocampal segmentation, referred to as “sum of subfields” segmentation (Iglesias et
al., 2015).
FreeSurfer was specifically designed to account for structural brain abnormalities
common to Alzheimer’s disease and aging (Fischl, et al., 2002), which share some
overlapping features with stroke populations (Mok et al., 2017; Yousufuddin and Young,
2019). Perhaps as a result, FreeSurfer has performed relatively well in stroke studies
such as Khlif, Egorova, et al., 2019 and Khlif, Werden, et al., 2019. However, FreeSurfer
labels and segments cortical and subcortical structures based on known information on
the locations and adjacencies of specific brain substructures from a probabilistic atlas
(Fischl, et al., 2002), which presents a significant challenge in stroke studies.
Hippocampal lesions are relatively uncommon in stroke (Chen et al., 2015; Szabo et al.,
2009), therefore, a lesion immediately disrupting hippocampal segmentation is not often
19
a concern. However, large brain lesions can introduce significant alterations to the
expected spatial distribution of brain structures. Following a large lesion, brain atrophy
occurs in the lesioned area and becomes filled with cerebrospinal fluid, leading in an
ipsilesional enlargement of the lateral ventricles in a process known as hydrocephalus ex
vacuo (Figure 1.7; Egorova, Gottlieb, Khlif, Spratt, & Brodtmann, 2019). The resulting
asymmetrical altered spatial distribution of brain structures can make it difficult for
Freesurfer to map the input brain to the probabilistic brain atlas.
Figure 1.7 Hydrocephalus Ex Vacuo A T1w coronal slice of a participant with chronic
stroke from the ENIGMA Stroke Recovery database is shown a large lesion. In the
lesioned hemisphere, hydrocephalus ex vacuo can be observed in the enlarged lateral
ventricle. The enlargement of the inferior horn of the lateral ventricle suggests secondary
degeneration of the hippocampus may have occurred. The resulting spatial shift has
caused FreeSurfer to under segment the hippocampus in the lesioned hemisphere
(labeled in red).
20
1.3.3 Hippodeep
Hippodeep, a new convolutional neural network-based (CNN) algorithm, has
recently emerged as a fast and robust hippocampal segmentation method (Thyreau et
al., 2018). Hippodeep does not warp individual images to an atlas; instead, it relies on a
hippocampal appearance model learned from existing FreeSurfer v5.3 labeled online data
sets as well as synthetic data. Nogovitsyn et al., 2019 compared hippocampal
segmentations by Hippodeep and FreeSurfer version 6.0 “sum of subfields”
segmentations in healthy aging populations and found that Hippodeep had better spatial
agreement with manual segmentations (Nogovitsyn et al., 2019). Prior to our study,
Hippodeep had not yet been evaluated in a stroke population.
1.3.4 Aim One Purpose
The purpose of aim one was to expand on previous findings (Khlif et al., 2018,
2019b; Nogovitsyn et al., 2019) and evaluate how Hippodeep compares to previously
tested methods for hippocampal segmentation in a stroke population in terms of (a)
quality control (QC) and (b) accuracy when compared to expert manual segmentations.
Overall, we hypothesized that Hippodeep's CNN-based method would perform
better on lesioned brain anatomy, resulting in fewer segmentation failures and
more accurate hippocampal segmentations than either FreeSurfer method.
21
1.3 Aim Two Background
1.3.1 Mechanisms of Stroke-related Damage
A stroke is a cerebrovascular incident that occurs because of disruption in blood
supply to a brain region. Reduced brain oxygen supply from an ischemic event causes
dramatically decreased adenosine triphosphate (ATP) production. This energy depletion
leads to an ionic imbalance resulting in an overload of intracellular calcium. Too much
intracellular calcium activates several death-signaling proteins such as calcium
dependent proteases, lipases, and DNAses that trigger in cell death in the ischemic core
(Khoshnam et al., 2017). However, the damage following a stroke reaches beyond tissue
damage within the ischemic core.
Remote structures synaptically connected to the site of stroke injury go through a
temporary period of depressed metabolism and blood flow as well (Chen et al., 2015).
This leads to a delayed neurodegeneration evident in the weeks to months following the
ischemic incident known as secondary degeneration (Zhang et al., 2012). The last 10
years of stroke rehabilitation research has ramped up efforts to improve our
understanding of the role of secondary degeneration in stroke recovery (Boyd et al.,
2017). Regions remote from the lesion have been associated with neurological deficits
and are thought to influence sensorimotor recovery (Liew et al., 2021; Zhang et al., 2012),
making them a prime target for stroke rehabilitation. Improving our understanding of
secondary degeneration is of great clinical interest, as the delayed onset of damage from
secondary degeneration provides a feasible window for clinical intervention (Chen et al.,
2015).
22
1.3.2 Post-stroke Secondary Degeneration of the Hippocampus
The hippocampus is notoriously vulnerable to post-stroke secondary degeneration
(Brodtmann et al., 2020; Haque et al., 2019; Khlif et al., 2019a; Schaapsmeerders et al.,
2015; Tang et al., 2012; Xie et al., 2011). Researchers have speculated about a number
of different mechanisms that may underlie secondary degeneration of the hippocampus.
Wang, Redecker, Bidmon, & Witte, 2004 proposes that symptoms of ischemic injury such
as brain edema, swelling, and increased intracerebral pressure may constrict the
hippocampus and limit its blood supply. Secondary degeneration could also be a result
of dysconnectivity, with glial scarring around the ischemic core acting as a mechanical
barrier (Xie et al., 2011). Hippocampal atrophy could also be caused by retrograde
degeneration from the destruction of connected pathways (Chen et al., 2015). One of the
more common theories is that secondary degeneration is a result of spreading depression
(SD), where neurotoxic signals from the ischemic core propagate to remote regions
through cell-to-cell communication (Schaapsmeerders et al., 2015; Tang et al., 2012;
Wang et al., 2004; Xie et al., 2011). In the acute phase of an ischemic incident, there is a
buildup of extracellular glutamate that may spread to the hippocampus through
neighboring gray matter regions. The hippocampus is filled with tightly packed, easily
excitable, glutamatergic neurons and a high density of NMDA receptors. Overexcitation
of the hippocampal glutamatergic network leads to hippocampal excitotoxicity, resulting
in hippocampal neuron apoptosis (Figure 1.8; Tang et al., 2012). The damaging effects
of SD are likely restricted to the lesioned hemisphere because SD waves do not
propagate easily through white matter (Chung et al., 2016), therefore the waves cannot
easily traverse to the contralesional hippocampus.
23
1.3.3 Post-stroke Hippocampal Studies
The significant damage of secondary degeneration on the spared post-stroke
hippocampus has been reported in animal models and in-vivo human models.
Immunohistochemistry in a study of rats with middle cerebral artery occlusion (MCAO)
revealed evidence of delayed pyramidal neuron loss in the ipsilesional hippocampus
(Wang et al., 2004). Xie et al., 2011 found smaller ipsilesional hippocampal volumes using
magnetic resonance imaging (MRI) and evidence of neuronal loss and glial cell activation
with magnetic resonance spectroscopy (MRS) in both rats with MCAO and patients with
chronic stroke. Accelerated ipsilesional hippocampal atrophy measured in humans with
in-vivo MRI compared to both the contralesional hippocampus and healthy controls has
Figure 1.8 Spreading Depression This figure from Tang et al., 2012 maps out the
pathophysiological mechanism of spreading depression after cortical and/or striatal
stroke that causes secondary degeneration of the hippocampus (and other regions).
24
been reported in a number of studies (Brodtmann et al., 2020; Haque et al., 2019; Khlif
et al., 2019a; Schaapsmeerders et al., 2015; Tang et al., 2012; Xie et al., 2011). In
summary, the effects of secondary degeneration on the hippocampus are most prominent
in the ipsilesional hemisphere in the weeks to months following a stroke.
1.3.4 Associations Between Hippocampal Volume and Lesion Size.
Although prior studies have found no significant associations between lesion size
and hippocampal volume (Tang et al., 2012; Xie et al., 2011), Schaapsmeerders et al.
(2015) found that larger lesion size was, in fact, significantly associated with smaller
ipsilesional hippocampal volumes. It is possible that the prior studies that found no
significant associations between lesion size and hippocampal volume may have been
underpowered given that they both had a sample size of less than 50 while
Schaapsmeerders et al., 2015 had a sample of 170 participants. The relationship between
lesion size and hippocampal volume may require a diverse, high-powered sample to
account for the heterogeneity of the lesion size and location to be detectable.
1.3.5 Aim Two Purpose
In this aim, we investigate the conflicting findings regarding associations between
lesion size and hippocampal volume. Although previous smaller studies have found no
significant associations between lesion size and hippocampal volume, a larger study with
170 participants found larger lesion size to be significantly associated with smaller
ipsilesional hippocampal volume. We hypothesized that large lesion size would be
significantly associated with smaller ipsilesional hippocampal volume. We use the
25
multi-site ENIGMA Stroke Recovery database to test for this association within each
research site individually and when pooling multi-site data. We also conduct a power
analysis to determine what is the minimum sample size necessary to detect associations
between lesion size and hippocampal volume.
1.4 Aim Three Background
1.4.1 Sensorimotor Impairment
Although there is a range of symptoms that can result from a stroke incident, the
most prevalent symptom is hemiparesis of upper extremities (Krakauer and Carmichael,
2017). An estimated two-thirds of stroke survivors suffer from persisting symptoms that
limit their ability to perform skilled hand movements necessary for daily functioning
(Edwards et al., 2019; Krakauer and Carmichael, 2017). Sensorimotor impairment refers
to a deficit in body structure or function such as decreased strength or sensation loss
(Krakauer & Carmichael, 2017; Stinear, Lang, Zeiler, & Byblow, 2020); in other words, a
physical impairment that may prevent a patient with stroke from successfully performing
a task. For example, a patient with a stroke who has difficulty grasping an object
(sensorimotor impairment) may struggle to raise a glass of water to their mouth.
Incomplete post-stroke sensorimotor recovery is considered a primary contributor to long-
term disability (Edwards et al., 2019). Persistent sensorimotor impairment in upper
extremities is associated with decreased independence during activities of daily living and
reduced quality of life in chronic stroke patients (Edwards et al., 2019; Krakauer and
Carmichael, 2017)
26
The Fugl-Meyer Assessment is considered the gold standard to assess
sensorimotor impairment of post-stroke hemiparesis (Hijikata et al., 2020). The Fugl-
Meyer Assessment was developed in the 1970’s to assess sensorimotor impairment after
stroke (Fugl-Meyer et al., 1975; Krakauer and Carmichael, 2017). The primary focus of
the Fugl-Meyer is to test a patient’s ability to isolate and control individual joint movements
(Krakauer and Carmichael, 2017). The Fugl-Meyer Assessment of Upper Extremities
(FMA-UE) consists of 33 items, including assessments of single-joint and multi-joint
movement, out-of-synergy movement, digit individuation, movement speed, dysmetria,
ataxia, and reflexes (Stinear et al., 2017). For each item, a clinician scores the task
performance as “0 - inability”, “1 - beginning ability”, or “2 - normal”. The scores of the
items are then summed to create a total score, ranging from 0 (severe impairment) to 66
(no impairment) (Hijikata et al., 2020). The FMA-UE has very high inter- and intra-rater
reliability and validity and remains the primary clinical batteries for sensorimotor
impairment assessment in both stroke research and clinical trials (Krakauer and
Carmichael, 2017).
1.4.2 Role of the Hippocampus in Sensorimotor Impairment
Most studies investigating stroke related secondary degeneration of the
hippocampus have focused on the implications of hippocampal damage on post-stroke
dementia. This is not surprising, as the hippocampus is best known for its role in learning
and memory (Small et al., 2011). However, the hippocampus is a densely connected
structure and is involved in circuitry beyond the limbic system. Although typically not
considered as a primary sensorimotor region, evidence suggests the hippocampus may
27
be involved in sensorimotor circuits. Structurally, the hippocampus is connected to
important sensorimotor areas such as the thalamus and basal ganglia through the spinal-
limbic pathway (Maller et al., 2019). There is good evidence that the hippocampus plays
a role in sensorimotor integration (Suzuki, 2007), sensorimotor learning (Albouy et al.,
2008; Burman, 2019; Jacobacci et al., 2020), and motor control (Burman, 2019).
Functional hippocampal connectivity in relation to sensorimotor behavior has also been
reported with the thalamus (Baumgartner et al., 2018), sensorimotor cortex (Burman,
2019), and the supplementary motor area (Mukamel et al., 2010). However, the
relationship between hippocampal integrity and post-stroke sensorimotor impairment is
unclear. Hippocampal damage due to secondary degeneration after stroke could weaken
sensorimotor circuits, leading to worsened chronic sensorimotor impairment.
Alternatively, greater sensorimotor impairment associated with degeneration of the
thalamus, basal ganglia, sensorimotor cortex, or supplementary motor area may lead to
downstream degeneration in the hippocampus.
1.4.3 Sex Differences
Data in dementia research (Nebel et al., 2018) and healthy aging (Wierenga et al.,
2020) suggests that hippocampal atrophy may accelerate in women over a certain age.
The hippocampus has an abundance of estrogen receptors and may be particularly
sensitive to the effects of estrogen (Bean et al., 2014). Bilateral increases in hippocampal
volume have been reported in women who receive estrogen supplements (Albert et al.,
2017). Furthermore, lower estrogen levels have also been associated with stroke severity
and mortality (Pappa et al., 2012) and postmenopausal women are considered to be at
28
higher risk of stroke (Cordonnier et al., 2017). Stroke-related outcomes including disability
and quality of life have been shown to be generally poorer in women than men
(Cordonnier et al., 2017; Dehlendorff et al., 2015; Gittler and Davis, 2018) although
conclusive sex differences have not been reported in terms of post-stroke sensorimotor
impairment (Hawe et al., 2020a). Since both stroke severity (Cordonnier et al., 2017;
Dehlendorff et al., 2015; Gittler and Davis, 2018; Pappa et al., 2012) and dementia-related
hippocampal atrophy (Nebel et al., 2018) are thought to be greater in women compared
to men, sex could moderate the relationship between sensorimotor impairment and
hippocampal volume. In particular, women may show a wider range of post-stroke
hippocampal volumes and sensorimotor impairment scores, perhaps leading to stronger
effect sizes in associations with women as compared to men.
1.4.4 Aim Three Purpose
In this aim, we investigated associations between hippocampal volume and post-
stroke sensorimotor impairment in participants with chronic stroke from the ENIGMA
Stroke Recovery database. Due to the primarily ipsilesional hippocampal damage
reported to prior stroke studies, we hypothesized that more severe post-stroke
sensorimotor impairment would be correlated with smaller ipsilesional
hippocampal volumes, independent of lesion size. In an exploratory analysis, we
tested to see if sex had a moderating effect on the relationship between sensorimotor
impairment and hippocampal volume. Due to more severe hippocampal vulnerability and
poorer stroke outcome trends in women, we hypothesized that women would have a
29
stronger relationship between greater sensorimotor impairment and smaller
hippocampal volume than men.
30
Chapter 2: Assessing automated hippocampal segmentation
methods in a stroke population
This section is adapted from:
Zavaliangos-Petropulu A, Tubi MA, Haddad E, Zhu A, Braskie MN, Jahanshad N,
Thompson PM, Liew S-L (2020) Testing a convolutional neural network-based
hippocampal segmentation method in a stroke population. Human Brain Mapping.
https://doi.org/10.1002/hbm.25210
2.1 Abstract
As stroke mortality rates decrease, there has been a surge of effort to study post-
stroke dementia to improve long-term quality of life for stroke survivors. Hippocampal
volume may be an important neuroimaging biomarker in post-stroke dementia, as it has
been associated with many other forms of dementia. Hippocampal volume may be an
important neuroimaging biomarker because it is vulnerable to post-stroke secondary
degeneration and hippocampal damage has been associated with post-stroke dementia.
However, studying hippocampal volume using MRI requires hippocampal segmentation.
Advances in automated segmentation methods have allowed for studying the
hippocampus on a large scale, which is important for robust results in the heterogeneous
stroke population. However, most of these automated methods use a single atlas-based
approach and may fail in the presence of severe structural abnormalities common in
stroke. Hippodeep, a new convolutional neural network-based hippocampal
31
segmentation method, does not rely solely on a single atlas-based approach and thus
may be better suited for stroke populations. Here, we compared quality control and the
accuracy of segmentations generated by Hippodeep and two well-accepted hippocampal
segmentation methods on stroke MRIs (FreeSurfer 6.0 whole hippocampus and
FreeSurfer 6.0 sum of hippocampal subfields). Quality control was performed using a
stringent protocol for visual inspection of the segmentations, and accuracy was measured
as volumetric correlation with manual segmentations. Hippodeep performed significantly
better than both FreeSurfer methods in terms of quality control. All three automated
segmentation methods had good correlation with manual segmentations and no one
method was significantly more correlated than the others. Overall, this study suggests
that both Hippodeep and FreeSurfer may be useful for hippocampal segmentation in
stroke rehabilitation research, but Hippodeep may be more robust to stroke lesion
anatomy.
2.2 Introduction
According to the World Health Organization, approximately 10.3 million people
experience a stroke each year worldwide (Feigin et al., 2017). Post-stroke dementia
(PSD), defined as any dementia occurring after stroke (including cognitive impairment,
Alzheimer's disease [AD], and vascular dementia) presents in roughly 30% of stroke
survivors (Mok et al., 2017). PSD is one of the leading causes of dependency in stroke
survivors (Leys et al., 2005) and is of growing concern for patients, families, and health-
care providers as stroke survival rates improve (Dichgans, 2019). Therefore, early
neuroimaging biomarkers that may contribute to PSD remain important to investigate.
32
The hippocampus may be an important biomarker for PSD. The hippocampus,
essential for memory function, is vulnerable to pathology and atrophy in multiple dementia
subtypes (Braak and Braak, 1991; Braskie and Thompson, 2014; Halliday, 2017),
including PSD (Gemmell et al., 2012; Gemmell et al., 2014). The hippocampus is usually
not directly impacted by an ischemic stroke lesion (Szabo et al., 2009). However,
emerging evidence suggests hippocampal damage occurs after a stroke as a result of
secondary degeneration (Xie et al., 2011). Specifically, ischemic stroke is associated with
reduced hippocampal volume, which is detectable in vivo by non-contrast MRI (Werden
et al., 2017).
Robustly studying patterns of hippocampal atrophy after stroke requires large
datasets, given the vast heterogeneity of stroke lesions in terms of lesion size, location,
and presentation. This has incentivized large multi-center worldwide consortia to obtain
large samples of post-stroke MRI to evaluate robust hippocampal patterns. Consortia
around the world - such as the Cognition and Neocortical Volume After Stroke Consortium
(CANVAS; Brodtmann et al., 2014) and the Stroke and Cognition Consortium
(STROKOG; Sachdev et al., 2017) - have made significant efforts to study the role of
hippocampal volumes in the context of overall stroke recovery on a large scale. The
Enhancing Neuroimaging through Meta-Analysis (ENIGMA) Stroke Recovery working
group (Liew et al., 2020) is also interested in studying the post-stroke hippocampus in the
context of sensorimotor recovery. Currently, manual segmentations are arguably the gold
standard for analyzing hippocampal volume in MRI studies (Frisoni et al., 2015), but this
approach is extremely time consuming and not feasible for large datasets such as these.
Therefore, efforts to develop and test automated hippocampal segmentation methods
33
have been undertaken to provide a more efficient way to study hippocampal volume on a
large scale.
Current automated brain structure segmentation algorithms predominantly rely on
atlas-based approaches, involving machine learning and sophisticated image registration
to a single probabilistic atlas of pre-labeled regions. FreeSurfer (Fischl, 2012; Fischl et
al., 2002), a robust method to segment both cortical and subcortical structures, is an atlas-
based approach and commonly used to study hippocampal volume in cognitively healthy
populations (Nobis et al., 2019; Ritchie et al., 2018) as well as in people with
neurodevelopmental, psychiatric, and neurodegenerative conditions (Hibar et al., 2016;
Müller-Ehrenberg et al., 2018; Schmaal et al., 2016; Van Erp et al., 2016; Zhao et al.,
2019). Recent studies by Khlif et al., (Khlif et al., 2019a, 2019b) compared automated
hippocampal segmentation methods, such as the gross hippocampal segmentation
available in FreeSurfer version 5.3, version 6.0, and the ‘sum of subfields’ segmentation
available in FreeSurfer version 6.0, in stroke populations. Khlif et al., (Khlif et al., 2019a,
2019b) reported that the FreeSurfer version 6.0 ‘sum of subfields’ segmentation was
among the most accurate methods for estimating hippocampal volume in healthy and
ischemic stroke populations with lesions outside the hippocampus.
FreeSurfer was specifically designed to account for structural brain abnormalities
common to AD and aging (Fischl, 2012), which share some overlapping features with
stroke populations (Mok et al., 2017; Yousufuddin and Young, 2019); perhaps as a result,
FreeSurfer has performed relatively well in stroke studies. However, large brain lesions
are distinct to stroke patients and can introduce large alterations to the expected spatial
distribution of brain structures, presenting a significant challenge to FreeSurfer.
34
FreeSurfer, and most other probabilistic atlas-based automated segmentation methods,
were not explicitly designed to accommodate significant brain injury pathology (Irimia et
al., 2011) and are more likely to fail in the presence of large lesions (Yang et al., 2016).
New methods that do not use single atlas-based automated segmentation methods may
better accommodate stroke pathology and help improve segmentation accuracies in
studies of the hippocampus in stroke. Related to this, recently, Hippodeep, a new
convolutional neural network-based (CNN) algorithm, emerged as a fast and robust
hippocampal segmentation method (Thyreau et al., 2018). Hippodeep relies on
hippocampal ‘appearance’ instead of a single atlas-based approach. Hippodeep has
better spatial agreement with manual segmentations than FreeSurfer version 6.0 ‘sum of
subfields’ segmentation in healthy aging populations, but has not yet been evaluated in a
stroke population (Nogovitsyn et al., 2019).
Our study sought to expand on previous findings (Khlif et al., 2019a, 2019b;
Nogovitsyn et al., 2019) and evaluate how Hippodeep compares to previously tested
methods for hippocampal segmentation in a stroke population. Using the Anatomical
Tracings of Lesions After Stroke dataset (ATLAS; Liew et al., 2018), we compared
Hippodeep, FreeSurfer version 6.0 gross hippocampal segmentation, and FreeSurfer
version 6.0 ‘sum of subfields’ segmentation in terms of 1) quality control (QC) and 2)
accuracy when compared to expert manual segmentations. QC and accuracy provide
different but complementary evaluations of hippocampal segmentation. QC was done by
visually inspecting segmentations to determine which segmentations failed to satisfy our
predetermined criteria for a good quality segmentation. We measure accuracy by
calculating intra-class correlation, which is a measure of how similar the volumes from
35
the automated segmentation methods are to their corresponding manual segmentations.
Overall, we hypothesized that Hippodeep’s CNN-based method would perform better on
lesioned brain anatomy, resulting in fewer segmentation failures and more accurate
hippocampal segmentations than either FreeSurfer method.
2.3 Methods
2.3.1 Data Acquisition
For our analyses, we used the ATLAS dataset (N=229), an open-source dataset
of anonymized T1-weighted structural brain MRI scans of stroke patients and
corresponding manually traced lesion masks (Liew et al., 2018). All 229 scans were
completed on 3-Tesla MRI scanners at a 1 mm
isotropic resolution, intensity normalized
and registered to the MNI-152 template space. T1-weighted MRIs, lesion masks, and
metadata are publicly available for download (Liew et al., 2018). We analyzed the
normalized data from these 229 participants as the input data to test the three automated
segmentation methods.
2.3.2. Hippocampal Segmentation Methods
2.3.2.a FreeSurfer version 6.0
As mentioned previously, Khlif et al. (Khlif, Egorova, et al., 2019; Khlif, Werden, et
al., 2019); found ‘sum of subfields’ segmentation available in FreeSurfer version 6.0 to be
one of the best performing segmentation methods for the stroke data they evaluated.
FreeSurfer is an atlas-based software that employs a Bayesian statistical approach to
segment and label brain regions (Fischl, 2012). It involves a series of data preprocessing
36
steps, such as intensity normalization, mapping of the input brain to a probabilistic brain
atlas, estimation of statistical distributions for the intensities of different tissue classes,
and labeling of cortical and subcortical structures based on known information on the
locations and adjacencies of specific brain substructures (Fischl et al., 2002).
FreeSurfer version 6.0 can output segmentations of 13 hippocampal subregions
using a refined probabilistic atlas (Fischl, 2012). This atlas was built from a combination
of ultra-high resolution ex vivo and in vivo MRI scans, to identify borders between
subregions of the hippocampus (Iglesias et al., 2015). The ex vivo scans included autopsy
samples of participants with AD and controls scanned with a 7T scanner at 0.13 mm
isotropic resolution that were then manually segmented by expert neuroanatomists. The
in vivo data consisted of manual segmentations from 1mm isometric resolution T1-
weighted MRI data acquired using a 1.5 T scanner from controls and participants with
mild dementia. In vivo and ex vivo segmentations were combined to create one single
computational atlas of hippocampal subfields. In this study, we combined the volumes of
the individually labeled hippocampal subfields output by FreeSurfer version 6.0 to create
a segmentation of the entire hippocampus, which we refer to as FS-Subfields-Sum
throughout our study.
FreeSurfer also outputs a separate hippocampal segmentation using a different
atlas, the Desikan-Killiany atlas (Desikan et al., 2006). The Desikan-Killiany atlas was
built using 40 T1-weighted 1x1x1.5 mm spatial resolution MRIs acquired on a 1.5T
scanner. These 40 participants were of ranging age and cognitive status with the intent
to include a range of anatomical variance common to aging and dementia in the atlas.
37
This hippocampal volume from the Desikan-Killiany atlas can be calculated using the
hippocampus labels of the aseg FreeSurfer output file.
FreeSurfer outputs segmentations to a FreeSurfer specific image space. The
FreeSurfer command, mri_label2vol, was used to transform the segmentation back to the
original MNI space used in the input for both FreeSurfer versions segmentations.
Segmentations from the aseg output are referred to as FS-Aseg throughout our study.
Prior studies have reported an inability to run FreeSurfer on certain participants
with large lesions (Bigler et al., 2018; Khlif et al., 2019a). In an effort to generate the
maximum number of segmentations, scans that were not segmented on the initial
FreeSurfer analysis were run a second time through FreeSurfer.
2.3.2.b Hippodeep
Hippodeep is a recent automated hippocampal segmentation algorithm that has
not yet been tested in stroke populations. Hippodeep does not warp individual images to
an atlas; instead, it relies on a hippocampal appearance model learned from existing
FreeSurfer v5.3 labeled online datasets as well as synthetic data (Thyreau et al., 2018).
Two types of synthetic data are included in training the Hippodeep CNN. The first
synthetic data is a manual segmentation of a synthetic high-resolution image of the
hippocampus generated from an average of 35 variations of MRI scans of a single healthy
participant. The purpose of segmenting the hippocampus on a high-resolution image (0.6
mm isotropic resolution) is to provide more detailed boundary information to the CNN that
might not be as clear on a lower resolution image. The second type of synthetic data used
to train the Hippodeep CNN are artificially geometrically distorted versions of the
38
FreeSurfer v5.3 training data. Some of the distortion goes beyond the range of clinically
plausible values but remains realistic enough to be easily delineated by a human rater.
The purpose of this distorted data is to provide relevant training guidance to the CNN. By
training the CNN on unconventional anatomy, Hippodeep may be more robust to severe
stroke pathology. Details on the specifics of how the synthetic data were generated can
be found in Thyreau et al. (2018).
Hippodeep outputs a probabilistic segmentation map calculated using a loss
function to allow for the uncertainty of voxels along the perimeter of the hippocampus in
native space, which can then be optionally thresholded. The probabilistic segmentation
was converted to a binary mask of the hippocampus, as recommended by Thyreau et al.,
(2018).
2.3.2.c Manual Segmentations
We tested the accuracy of the automated methods by randomly selecting 30
participants for whom all three automated segmentation algorithms (FS-Aseg, FS-
Subfields-Sum, and Hippodeep) were able to successfully output hippocampal
segmentations. Only participants with unilateral lesions were considered. The ATLAS
data was organized by lesion size and divided into thirds; small (range = 0.18 - 4.82 cubic
centimeters (cc)), medium (range = 4.98 - 22.7cc), and large (range = 23.6 - 291.0cc)
lesions. From each lesion size group, five participants with right hemisphere lesions and
five participants with left hemisphere lesions were randomly selected. In this way, we
examined the influence of lesion size across a broad range of lesion sizes, and with
lesions equally distributed across hemispheres. In this data sample, all lesions occurred
outside the medial temporal lobe.
39
Hippocampi for the subset of these 30 participants were manually traced by an
expert rater (AZP), strictly adhering to the EADC-ADNI harmonized protocol for manual
hippocampal segmentation (Boccardi et al., 2015; Frisoni et al., 2015). Coronal slices
were used to trace the hippocampi using ITK-Snap (Yushkevich et al., 2006). The sagittal
view was used to confirm hippocampal boundaries and edit the segmentations.
Hippocampi were segmented blindly based on participant ID alone, starting with the left
hippocampus, followed by the right hippocampus. Bilateral hippocampi were never
overlaid on the T1-weighted image at the same time to avoid using the segmentation from
one hemisphere to bias the other. The manual segmentations were checked for quality
by another expert in hippocampal neuroanatomy (MAT). All manual segmentations are
available for download here: https://github.com/npnl/Hippocampal_Segmentation
2.3.3. Analyses
2.3.3.a Quality Control (QC)
We manually assessed the quality of segmentations produced by each automated
hippocampal segmentation method in the full ATLAS dataset (N=229) using the ENIGMA
Stroke Recovery QC protocol (Liew et al., 2020). Briefly, a trained researcher (AZP)
reviewed nine slices of each brain (3 coronal, 3 axial, and 3 sagittal) with the bilateral
segmentations overlaid on the T1, which were generated for each participant (Figure
2.1). A segmentation failed QC if the segmentation grossly underestimated the
hippocampus (underestimated), overestimated by including regions of the brain outside
the hippocampus (overestimated), missed the hippocampus entirely (miss), or failed to
output a segmentation (no output) (Figure 2.2).
40
Figure 2.1 Hippocampal Segmentation Quality Control Here we show an example of
the 9 images per MRI used to perform Quality Control (QC). QC was done using 9 slices
of the brain to observe the bilateral hippocampus. The right (red) and left (blue)
hippocampi are overlaid onto the T1 and visually checked for quality. A segmentation
failed QC if the segmentation grossly underestimated the hippocampus (underestimated),
overestimated by including regions of the brain outside the hippocampus (overestimated),
missed the hippocampus entirely (miss), or failed to output a segmentation (no output).
The QC images for all 229 ATLAS scans can be found here:
https://github.com/npnl/Hippocampal_Segmentation.
41
Figure 2.2 Hippocampal Segmentation Quality Control Results Hippocampal
segmentations produced by automated segmentation methods (Hippodeep, FS-Aseg,
and FS-Subfields-Sum) on the 229 ATLAS participants were inspected for quality
according to the ENIGMA Stroke Recovery Working Group quality control (QC) protocol
(Liew et al., 2020). Segmentations failed QC for four possible reasons: 1) failing to output
a segmentation entirely (no output) 2) including voxels in the segmentation that are clearly
outside of the hippocampus (overestimating) 3) underestimating the hippocampus
(underestimating), or 4) producing a segmentation that misses the hippocampus entirely
(miss). In this figure, we report the total breakdown of the QC results by hemisphere. The
results are further broken down by location of lesion (LHL= left hemisphere lesion, RHL=
right hemisphere lesion). Percent fail for left and right hippocampi is calculated as the
total number of segmentations that failed QC for the specified hemisphere divided by 229.
Percent fail for total is the number of segmentations divided by 458.
42
QC was reported in two levels of stringency: 1) methods-wise QC and 2) across-
methods QC, similar to Sankar et al. (2017). For methods-wise QC, we calculated a QC
fail rate for each automated segmentation method by dividing the total number of
segmentations that failed for each segmentation method by the total number of
segmentations (229 participants * 2 hippocampi = 458 segmentations).
Across-methods QC fail rate was calculated as the total number of participants for
which all three automated algorithms failed QC on at least one of the hippocampi divided
by the total number of participants in the analysis (N=229).
QC images and scores for each of the automated hippocampal segmentation
methods on the 229 participants in ATLAS are available here:
https://github.com/npnl/Hippocampal_Segmentation.
2.3.3.b Statistical Analysis of Accuracy
All statistical analyses were conducted in R-Studio version 1.1.463. To promote
open science and reproducibility, all statistical analyses and code used for this study can
be found here: https://github.com/npnl/Hippocampal_Segmentation
2.3.3.b.i Volume Correlation Analysis
We evaluated the agreement in hippocampal volume across segmentation
methods in the dataset of 30 participants, by calculating the Pearson’s correlation
43
coefficient (R; Pearson, 1895) and the intra-class correlation coefficient (Shrout and
Fleiss, 1979) in the ipsilesional and contralesional hippocampi separately. We
predetermined the number of segmentation methods and we assumed no generalization
to a larger population. Therefore, we assumed fixed judges for the intra-class correlation
statistical analyses (ICC3).
2.4 Results
2.4.1. Quality Control
First, we performed a rigorous quality control analysis for the segmentations
generated by each automated method. This provided a sense of how robust each method
was for generating good quality segmentations on the stroke data. The method-wise QC
fail rate for FS-Aseg was 30.9% (N=144), 23.6% (N=108) for FS-Subfields-Sum, and
3.3% (N=15) for Hippodeep. The across-methods QC fail rate was 45.0% (N=103). A
summary of reasons for QC fails by hemisphere for each segmentation algorithm can be
found in Figure 2.2.
FS-Aseg did not output segmentations for 38 participants and FS-Subfields-Sum
did not output segmentations for 40 participants (the 38 that did not output from FS-Aseg
plus two additional participants). Of the 80 total hippocampi (40 participants * 2
hippocampi) that were not segmented by either FS-Aseg or FS-Subfields-Sum, 75 of
these hippocampi were successfully segmented by Hippodeep and passed QC.
Hippodeep also successfully segmented the remaining 5 hippocampi, but these did not
pass QC and were all underestimated ipsilesional hippocampi. QC images of Hippodeep
44
segmentations for participants who had no output by FS-Aseg or FS-Subfield-Sum are
compiled in a file here:
https://github.com/npnl/Hippocampal_Segmentation/Hippodeep_QC_for_no_output_FS
_scans.pdf
2.4.2. Accuracy
In the subset of 30 participants with manually segmented hippocampi, we
compared hippocampal volume between automated and manual segmentations. All three
segmentation methods overestimated both ipsilesional and contralesional hippocampal
volume, compared to the manual gold standard (Figure 2.3, Figure 2.4). Hippodeep and
FS-Subfields-Sum segmentations were not significantly different in volume (Figure 2.4a).
As expected, volumes from all three segmentation methods were strongly
correlated with volumes from the manual segmentations (Table 2.1). Volumes from FS-
Subfields-Sum had the strongest correlation with manual segmentation volumes
(ipsilesional ICC3 = 0.65; contralesional ICC3 = 0.83). Hippodeep measures were also
strongly correlated with manual segmentation volumes (ipsilesional ICC3 = 0.64;
contralesional ICC3 = 0.75). FS-Aseg was the least correlated with the manual
segmentation volumes (ipsilesional ICC3 = 0.50; contralesional ICC3 = 0.71). Volumes
from FS-Subfields-Sum and Hippodeep were strongly correlated with each other
(ipsilesional ICC3 = 0.91; contralesional ICC3 = 0.90). However, upper and lower bounds
for ICC3 indicated there were no significant differences among ICC3 values across
comparisons (Figure 2.4b). Volumes from all three segmentation methods were strongly
correlated with each other (Table 2.2).
45
Figure 2.3 Automated Hippocampal Segmentations Automated hippocampal
segmentations are overlaid, along with the manual segmentation, on MRI data from an
example participant. Each row shows the results of a different automated segmentation
method. The left column shows a sagittal view of the ipsilesional hemisphere, the middle
column shows a coronal view of the body of bilateral hippocampi, and the rightmost
column shows a sagittal view of the contralesional hemisphere.
46
Figure 2.4 Comparing Segmentation Accuracy A) Mean hippocampal volume is
plotted for manual and automated segmentation methods in the 30 participants with
manually segmented hippocampi. All three automated segmentation methods on average
overestimated the manually defined segmentation volume. This trend is consistently
found for scans with small, medium, and large lesions. Error bars represent standard
deviation. B) Intraclass Correlation Coefficient (ICC3) was calculated correlating volumes
from each automated segmentation algorithm with manual segmentations. The error bars
indicate the upper and lower bound of ICC3. FS-Subfields-Sum has the highest ICC3 with
manual segmentations, although none of the ICC3 results are significantly different
across automated methods.
47
Table 2.1 Accuracy Compared to Manual Segmentations Intraclass correlation
coefficient (ICC3), Pearson’s correlation coefficient (R), and p-values were calculated
correlating hippocampal volume from the automated segmentation methods to the
manual segmentations. Correlations between FS-Subfields-Sum and Hippodeep are also
shown because the resulting hippocampal volumes were very similar.
Ipsilesional Contralesional
ICC3 R p-value ICC3 R p-value
FS-Aseg vs. Manual 0.50 0.59 6.45 x10
-4
0.71 0.80 1.21 x10
-7
FS-Subfields-Sum vs. Manual 0.65 0.67 5.71 x10
-5
0.83 0.84 5.38 x10
-9
Hippodeep vs. Manual 0.64 0.69 2.18 x10
-5
0.75 0.75 1.91 x10
-6
Table 2.2 Accuracy Compared Across Automated Methods Intraclass correlation
coefficient (ICC3) was calculated to compare segmentations from each automated
method. Upper and lower boundaries of ICC3 are also reported. The volumes for all
hippocampal segmentations are highly correlated, implying inter-algorithm consistency.
Ipsilesional Contralesional
Method ICC3 Lower Bound Upper Bound ICC3 Lower Bound Upper Bound
FS-Subfields-Sum
Vs.
Hippodeep
0.91 0.83 0.95 0.90 0.82 0.94
FS-Subfields-Sum
vs
FS-Aseg
0.89 0.80 0.94 0.92 0.85 0.96
FS-Aseg
vs
Hippodeep
0.85 0.74 0.92 0.79 0.64 0.88
2.5 Discussion
In this study, we compared the quality control (QC) and accuracy of three automated
segmentation algorithms (Hippodeep, FS-Subfields-Sum, FS-Aseg) used to estimate
hippocampal volume in individuals with stroke. Our study found that Hippodeep was able
to generate the greatest number of segmentations that passed QC, while FS-Subfields-
48
Sum performed slightly higher in terms of intraclass correlations (ICC3), although not
significantly higher than Hippodeep and FS-Aseg. This suggests that all three automated
segmentation methods produce good volumetric correspondence with manual
hippocampal segmentations but Hippodeep was able to produce a significantly greater
number of total usable segmentations.
Hippodeep had the smallest methods-wise QC fail rate of the three automated
segmentations tested (3.3%). FS-Subfields-Sum had the second lowest methods-wise QC
fail rate (23.6%) followed by FS-Aseg (30.8%). Sankar et al., (2017) report high rates of
poor-quality hippocampal segmentation across multiple automated segmentation
algorithms, including FS-Aseg in elderly populations. While a certain amount of
segmentation failure is expected for automated methods, automated segmentations in
stroke populations are challenged by stroke pathology. An estimated 10-20% of FreeSurfer
subcortical segmentations do not pass quality control in the ENIGMA Stroke Recovery
Working Group data (Liew et al., 2020). In the stroke data we used here, Hippodeep was
able to generate segmentations of adequate quality for 27.5% more hippocampi than FS-
Aseg and 20.3% more than in FS-Subfields-Sum. Hippodeep generated volume estimates
for all of the participants whose data could not be run through FreeSurfer in our study, and
all but 5 of these segmentations passed QC. Therefore, Hippodeep can potentially help to
maximize the number of participants included in analyses whose data might not run
successfully through FreeSurfer, potentially boosting statistical power, and reducing the
bias that can come from excluding participants. Obtaining robust statistical power is of keen
interest to the stroke recovery field, as a recent review by Kim & Winstein (2017) found that
less than 30% of stroke recovery studies met the appropriate sample size criteria to achieve
49
sufficient statistical power for predicting recovery. Our understanding of the role of
hippocampal volume in stroke recovery will benefit from studies with larger, more
representative samples.
FS-Subfields-Sum and Hippodeep were both very competitive in terms of their
correlations with ipsilesional and contralesional volume estimates. FS-Subfields-Sum
segmentations were more highly correlated with manual segmentations (ICC3) than
Hippodeep, although both were high and considered very reliable (Koo & Li, 2016). There
were no significant differences across methods in ICC3 results. Ipsilesional FS-Aseg
volume estimates had the lowest ICC3, but this correlation was still high enough to be
considered moderately reliable (Koo and Li, 2016). For all three automated methods, as
expected, contralesional ICC3 was higher than ipsilesional ICC3. Hippodeep and FS-
Subfields-Sum may have performed better than FS-Aseg in terms of volumetric accuracy
because information from a high-resolution hippocampus is included in both the Hippodeep
and FS-Subfield-Sum algorithms. FS-Subfields-Sum is based on an atlas generated using
manual segmentations on an ultra-high resolution atlas (0.13 mm isotropic resolution;
Iglesias et al., 2015). Hippodeep uses information from a manually traced hippocampus on
a synthetic high-resolution image (0.6mm isotropic resolution; Thyreau et al., 2018). In
contrast, FS-Aseg uses the Desikan-Killiany atlas, which was built using only scans of
1x1x1.5mm resolution (Desikan et al., 2006). The Desikan-Killiany atlas was designed to
segment many structures across the brain, many of which are clearly delineated on low
resolution scans. Hippocampal boundary detection improves with stronger MRI field
strength and spatial resolution (Giuliano et al., 2017). Including more detailed information
on hippocampal boundaries that appear ambiguous on a low-resolution MRI may improve
50
segmentation performance. Due to these differences in underlying MRI scan resolution
used to develop each method, mild variability in the resulting segmentations and
correlations with manual segmentations is expected. Overall, further exploration of the
methodological aspects of successful automated segmentation methods may be helpful to
inform future development of methods in populations with irregular neuroanatomy.
Beyond QC and accuracy, there are other technical aspects to consider when
comparing Hippodeep, FS-Subfields-Sum, and FS-Aseg. Hippodeep requires less
computational power than FreeSurfer and runs within minutes, whereas FreeSurfer can
take over 24 hours on a typical CPU (Thyreau et al., 2018; Nogovitsyn et al., 2019).
However, Hippodeep only outputs estimates of the hippocampus and total intracranial
volume. In addition to FS-Subfields-Sum and FS-Aseg, FreeSurfer also estimates other
brain measures beyond hippocampal volume and intracranial volume, such as individual
subfield volumes (Iglesias et al., 2015), and cortical and subcortical volumes, as well as
thickness measures, and other vertex based measures and attributes that can be used for
surface-based statistical analyses (Fischl, 2012). Additionally, FreeSurfer has an extensive
archive of user questions for troubleshooting, while Hippodeep is a recent method that is
not as extensively documented. Therefore, selection of the appropriate hippocampal
segmentation method should be evaluated within the context of the study requirements and
constraints (Table 2.3).
51
Table 2.3 Pros and Cons of Each Segmentation Method Here we present a roadmap
for using each of the automated hippocampal segmentation method tested in this paper.
Study requirements and constraints should be considered when selecting which method
to apply.
How to Run Pros Cons
Anatomical
Variability
FS-Aseg
Link to website
recon-all -s subject
Use aparc+aseg.mgz in
subject/mri/ folder to
extract label 17 for the left
hippocampus and label 53
for the right hippocampus
-Widely used in the
literature
-Provides brain measures
beyond the hippocampus
(cortical and subcortical
volumes and thickness,
etc; Fischl, 2012)
-Extensive support archive
-Did not output
segmentations for all
of the stroke data
-Modest accuracy
with manual
segmentation
-Time consuming
-Resource intensive
-Low resolution atlas
-Atlas created
on 1x1x1.5 mm
resolution scan
FS-Subfields-Sum
Link to website
recon-all -s subject -
hippocampal-subfields-T1
rh.hippoSfLabels-
T1.v10.FSvoxelSpace_nat
ive.mgz and
lh.hippoSfLabels-
T1.v10.FSvoxelSpace_nat
ive.mgz in subject/mri
folder
-Strong accuracy with
manual segmentation
-Provides brain measures
beyond the hippocampus
(cortical and subcortical
volumes and thickness,
etc; Fischl, 2012)
-Provides information
about individual subfield
volumes (Iglesias et al.
2015)
-Widely used in the
literature
-Extensive support archive
-Did not output
segmentations for all
of the stroke data
-Time consuming
-Resource intensive
-Atlas created
on 0.13 mm
isotropic
resolution scan
Hippodeep
Link to website
deepseg1.sh
subject_t1.nii.gz
example_brain_t1_mask_
L.nii.gz and
example_brain_t1_mask_
R.nii.gz
-Strong accuracy with
manual segmentation
-Can help maximize
sample size
• Output
segmentations for
all of the stroke data
• Able to produce
good segmentations
for most of the
participants that
could not run
through FreeSurfer
-Short run-time
-Only outputs
hippocampal
segmentations and
total brain volume
-Not trained
specifically on stroke
data
-Newer method with
limited support
archives
-Includes a
manual
segmentation
built on 0.6 mm
isotropic
resolution scan
2.6 Limitations
A key methodological limitation to consider when comparing these segmentation
methods is that none of these approaches were designed specifically to accommodate
52
severe stroke pathology. The default atlases used in FreeSurfer, including FreeSurfer
subfields, were created based on data from cognitively healthy elderly adults and patients
with early AD pathology (Desikan et al., 2006; Iglesias et al., 2015). Stroke pathology, such
as large lesions, hydrocephalus ex vacuo of the lateral ventricle (Egorova et al., 2019;
Nelson, 2003), and midline shifts (Liao et al., 2018), can alter expected spatial distribution
of brain anatomy. As a result, stroke pathology can interfere with templates used by existing
atlas-based approaches, resulting in inaccurate hippocampal segmentations. Although the
CNN used in Hippodeep was not trained on data with stroke pathology, it is trained to
anticipate extreme anatomical variability from the synthetic data. Being robust to extreme
anatomical variability may explain why Hippodeep was able to perform well in stroke
participants. Stroke-specific CNN hippocampal segmentation models that include stroke
pathology in training data may further improve automated hippocampal segmentation in
this population.
2.7 Conclusion
In this study, we compared three automated hippocampal segmentation methods in
a large stroke population. While all three methods yielded similar volumes, Hippodeep had
the lowest method-wise QC fail, suggesting it may be the most robust to post-stroke
anatomical distortions. The use of more accurate automated hippocampal segmentation
methods may reveal clinical associations that are so far undetected. Additionally, future
work should aim to extract subfields from the Hippodeep segmentation to further enhance
our understanding of how the specific regions of the hippocampus are indirectly impacted
53
by stroke lesions. Overall, our results suggest that Hippodeep may be an optimal method
for accurate and robust hippocampal segmentation methods in diverse stroke populations.
54
Chapter 3: Associations between lesion size and post-stroke
hippocampal volume in patients with chronic stroke
This section is adapted from:
Zavaliangos-Petropulu A, …. Liew S-L (2021) Associations between hippocampal
volume and sensorimotor impairment in chronic stroke survivors: An ENIGMA Stroke
Recovery Analysis. In preparation
3.1 Abstract
The hippocampus is a brain region particularly vulnerable to post-stroke secondary
degeneration and is thought to be an important biomarker for predicting post-stroke
outcomes, such as dementia and depression. Prior cross-sectional studies have found
that chronic post-stroke hippocampal damage is observed in the ipsilesional hemisphere.
However, it is not yet well understood whether or not lesion size influences hippocampal
damage, with inconsistent findings reported in the literature. Some studies report no
significant association between lesion size and hippocampal volume while others report
that larger lesion size is significantly associated with smaller ipsilesional hippocampal
volume. In this chapter, we investigate the controversy in the literature by pooling cross-
sectional T1-weighted brain MRIs of 357 participants with chronic stroke from the
ENIGMA Stroke Recovery database with manually traced lesion masks. We used a
robust mixed effects model to test the association between lesion size and hippocampal
volume (ipsilesional and contralesional separately). We found that larger lesion size was
55
significantly associated with smaller ipsilesional hippocampal volume (p-value < 0.001)
but was not significantly associated with contralesional hippocampal volumes (p = 0.60).
A follow-up power analysis revealed that a sample size of at least 79 participants is
necessary to achieve at 80% power, suggesting that conflicting evidence in the literature
may be attributed to underpowered samples. Overall, in this study we provide supporting
evidence to existing literature that lower hippocampal volumes post-stroke are likely a
consequence of damage within the lesioned hemisphere, beyond normal aging-related
neurodegeneration.
3.2 Introduction
The decreasing rate of stroke mortality has led to a growing population of stroke
survivors in need of rehabilitation (Johnson et al., 2019; Mozaffarian et al., 2016; Virani
et al., 2020). Stroke survivors are left with long-term symptoms such as hemiparesis,
cognitive impairment (Dichgans, 2019), depression (Robinson & Jorge, 2016), and
anxiety (Chun, Whiteley, Dennis, Mead, & Carson, 2018). To help clinicians, caregivers,
and patients make informed decisions about planning comprehensive rehabilitation
treatment, there is a critical need to identify reliable biomarkers of stroke outcomes (Boyd
et al., 2017). Although substantial damage that influences stroke outcomes occurs at the
primary site of ischemic injury, regions remote to the lesion can also incur damage in the
weeks to months following the ischemic incident and contribute to stroke outcomes as
well. This process is known as secondary degeneration. The last 10 years of stroke
rehabilitation research has ramped up efforts to improve our understanding of the role of
secondary degeneration in stroke recovery (Boyd et al., 2017; C M Stinear et al., 2017;
56
Cathy M. Stinear et al., 2020) as the delayed onset of damage from secondary
degeneration provides a feasible window for clinical intervention (Chen et al., 2015).
The hippocampus is a brain region particularly vulnerable to post-stroke secondary
degeneration and may act as an important biomarker for stroke outcomes. The
hippocampus is a key structure in the limbic system, which plays an important role in
emotional regulation and cognitive functioning (Bari, Niu, Langevin, & Fried, 2014). Limbic
system disruption is thought to drive cognitive impairment, depression, and anxiety
disorders (Bari et al., 2014). Stroke-related infarcts in the hippocampus are uncommon
(Szabo et al., 2009). Therefore, post-stroke hippocampal damage is attributed to
secondary degeneration and has been associated with post-stroke dementia
(Schaapsmeerders et al., 2015; Werden et al., 2017) and post-stroke depression (Shi et
al., 2017); both conditions that significantly influence long-term disability (Dichgans, 2019;
Robinson & Jorge, 2016).
Bilateral hippocampal damage is observed across a number of neurological and
psychiatric disorders, as well as risk factors for stroke such as hypertension (Fiford et al.,
2020), changes in estrogen (Albert et al., 2017), and ageing (Wierenga et al., 2020).
However, hippocampal damage that occurs following a stroke is thought to be a
consequence of the ischemic insult because it appears most prominently in the lesioned
hemisphere. Both rodent (Wang et al., 2004; Xie et al., 2011) and human (Brodtmann et
al., 2020; Haque et al., 2019; Khlif, Egorova, et al., 2019; Schaapsmeerders et al., 2015;
Tang et al., 2012; Xie et al., 2011) stroke studies of spared hippocampi show evidence of
severe ipsilesional hippocampal damage. Structural magnetic resonance imaging (MRI)
studies report smaller ipsilesional hippocampal volumes in chronic stroke patients, on
57
average, when compared to healthy controls (Brodtmann et al., 2020; Haque et al., 2019;
Khlif et al., 2018; Schaapsmeerders et al., 2015; Tang et al., 2012; Xie et al., 2011).
Magnetic resonance spectroscopy evidence of contralesional hippocampal neuronal loss
(Tang et al., 2012), and contralesional longitudinal hippocampal atrophy measured with
MRI (Brodtmann et al., 2020) have been reported, however contralesional damage is
more challenging to detect in a cross-sectional sample at a 1-mm isometric MRI
resolution. Overall, the effects of secondary degeneration are most evident on a macro
scale in the ipsilesional hippocampus starting in the sub-acute phase of recovery, three
months following a stroke (Brodtmann et al., 2020).
There is conflicting evidence regarding the impact of stroke-related lesion size on
hippocampal volume, with some studies reporting no significant associations between
lesion size and hippocampal volume (Tang et al., 2012; Xie et al., 2011), and others
reporting that larger lesion size is significantly associated with smaller ipsilesional
hippocampal volumes (Schaapsmeerders et al., 2015). In this chapter, we investigated
whether lesion size was associated with hippocampal volume, using a large sample of
brain MRI scans with manually segmented stroke lesions. It is possible that the
controversy in the literature is a result of underpowered studies. Both publications that
report no significant associations between lesion size and hippocampal volume analyzed
less than 50 participants with stroke (Tang et al., 2012; Xie et al., 2011). The publication
that found a significant association between lesion size and hippocampal volume had a
sample of 170 participants with stroke (Schaapsmeerders et al., 2015). Due to the
heterogeneity of stroke rehabilitation and neuroimaging research, large consortium based
multi-site studies such as the ENIGMA Stroke Recovery Working Group are important for
58
achieving large and diverse samples that can identify associations that may have
otherwise been undetectable in a smaller single-site sample (Thompson et al., 2020). In
addition, the diversity of data allows us to verify whether associations hold true beyond a
single cohort, or whether they depend on the cohort assessed, ultimately improving the
robustness and generalizability of research findings.
In this study, we aimed to investigate the relationship between lesion size and
ipsilesional and contralesional hippocampal volumes in 357 participants with chronic
stroke across 18 cohorts from the ENIGMA Stroke Recovery Working Group (Liew et al.,
2020). To our knowledge, associations between lesion size and hippocampal volume
have not been previously investigated in a multi-site study. We hypothesized that larger
lesion size would be significantly associated with smaller ipsilesional but not
contralesional hippocampal volumes. Furthermore, we hypothesized that the controversy
in the literature regarding the association between lesion size and hippocampal volume
is caused by underpowered samples. Additionally, to our knowledge, this is the first stroke
study to estimate hippocampal volume using Hippodeep, a convolutional neural network
hippocampal segmentation method that, in Chapter 2, we found outperforms other
automated hippocampal segmentation methods in the context of stroke pathology
(Zavaliangos ‐Petropulu et al., 2020). In a supplemental analysis using a subset of the
ENIGMA Stroke Recovery Working Group database of age, sex, and research cohort
matched participants with stroke and healthy controls (N=61 across 3 cohorts), we tested
to see if hippocampal segmentations generated with Hippodeep reproduce previous
findings that report ipsilesional hippocampal volume differences between participants with
stroke and healthy controls. We hypothesized that participants with stroke would have
59
significantly smaller ipsilesional hippocampal volumes than healthy controls but would not
differ in contralesional hippocampal volumes.
3.3 Methods
3.3.1 ENIGMA Stroke Recovery Dataset
We used cross-sectional data from the ENIGMA Stroke Recovery Working Group
database available as of December 15, 2020. Details of the ENIGMA Stroke Recovery
procedures and methods are available in Liew et al., 2020. The data were collected
across 18 research studies (cohorts) across 10 different research institutes in 6 countries.
ENIGMA Stroke Recovery participants with the following data were included: high
resolution (1-mm isotropic) T1-weighted brain MRI (T1w) acquired with a 3T MRI scanner,
a manually traced lesion mask, age, and sex. Exclusion criteria included site-reported
bilateral, brainstem, or cerebellar lesions. Additionally, to use the same sample for
analyses in Chapter 4, where we investigate associations between hippocampal volume
and lesion size, we excluded participants who had no sensorimotor impairment (Fugl-
Meyer Upper Extremity = 66). Each hippocampus was visually inspected with lesion
masks overlaid to ensure that no hippocampal lesions were included.
As we were interested in studying potential effects of secondary degeneration of
the hippocampus, we only included participants with chronic stroke (defined as data
acquired at least 180 days post-stroke; Bernhardt et al., 2017). The total initial sample
size was N=357 (age: median = 61 years, interquartile range (IQR) = 18, range = 23-93;
FMA-UE: median = 41, IQR = 28, range= 0-65; 135 women and 222 men) (Table 3.1).
60
Table 3.1 Demographics for ENIGMA Stroke Recovery Working Group participants
included in the study by cohort. Total sample size (N), number of women and men,
and information about age (years) and raw lesion size in cubic centimeters (cc) are listed.
Cohort N Women/Men
Median Age (years)
(IQR, min-max)
Median Lesion Size (cc)
(IQR, min-max)
Cohort 1 39 10/29 61 (17, 31-80) 6.1 (20.3, 0.04-120.8)
Cohort 2 12 6/6 69.5 (12, 39-85) 28.3 (28.5, 4.2-137.4)
Cohort 3 15 6/9 61 (17, 33-85) 21.1 (68.7, 0.6-182.2)
Cohort 4 19 6/13 44 (15, 30-68) 35.8 (54.4, 4.5-313.5)
Cohort 5 28 12/16 64 (18, 44-81) 1.9 (25.7, 0.1-237.7)
Cohort 6 10 3/7 61 (12.5, 49-72) 1.4 (1.1, 0.5-9.1)
Cohort 7 14 5/9 58 (12, 45-69) 2.0 (2.9, 0.04-6.9)
Cohort 8 11 4/7 56 (12, 45-74) 35.8 (50.2, 0.7-103.9)
Cohort 9 11 3/8 59 (3, 45-68) 2.6 (21.7, 0.7-53.7)
Cohort 10 8 4/4 58 (8, 46-73) 28.4 (43.2, 0.4-59)
Cohort 11 22 6/16 61.5 (11, 23-75) 5.6 (41.5, 0.4-201.4)
Cohort 12 13 4/9 57 (13, 32-80) 4.8 (18.2, 0.3-98)
Cohort 13 12 4/8 66 (16, 31-83) 4.4 (37.6, 0.2-107.5)
Cohort 14 29 18/11 50 (15, 25-79) 12.1 (28.6, 0.1-143.6)
Cohort 15 10 3/7 61.5 (11, 42-76) 9.1 (23.4, 3-186.1)
Cohort 16 40 14/26 66.5 (11, 43-93) 9.2 (26.1, 0.5-111.8)
Cohort 17 36 15/21 70 (14, 37-80) 7.6 (29.3, 0.3-188.4)
Cohort 18 28 12/16 64 (14, 34-85) 5 (29.4, 0.7-136.9)
Total 357 135/222 61 (18, 23-93) 7.6 (33.4, 0.04-313.5)
61
In a supplemental analysis investigating differences between hippocampal volume
in participants with stroke and healthy controls, we used a subset of age-, sex-, and
cohort- matched ENIGMA Stroke Recovery participants. Due to limitations in the control
population sample size, only participants with cohort-reported left hemisphere lesion
stroke were analyzed. Healthy controls were matched exactly for age, sex, and cohort
and within three years for age with replacement to participants with stroke. No more than
three controls were matched to a single participant with stroke. N=25 healthy controls
(age: median = 62, IQR = 9.7, range = 48-75; 9 women and 16 men) and N=36
participants with stroke (age: median = 59.0, IQR = 8.3, range = 46-76; 10 women and
25 men) were included from 3 cohorts (Table 3.2).
Table 3.2 Demographics for Stroke vs Control Demographics for the subset of
participants with left hemisphere stroke and age, sex, and cohort matched controls. Total
sample size (N), number of women and men, and information about age are listed
Healthy Controls Participants with Stroke
N Sex
(Women/Men)
Age N Sex
(women/men)
Age
Cohort 5 11 5/6 67 (11,48-75) 15 6/9 65 (10.5,46-76)
Cohort 9 6 1/5 56 (6.75,49-66) 8 2/6 58 (2,50-68)
Cohort 11 8 3/5 60.8 (5,51-71) 13 3/10 59 (5.6,49-,71)
TOTAL 25 9/16 62 (9.7,48-75) 36 10/25 59 (8.3,46-76)
3.3.2 MRI Data Analysis
Hippodeep, a convolutional neural network-based hippocampal segmentation
application, was used to estimate hippocampal volume and total intracranial volume
62
(eTIV) from the T1w MRI (Thyreau et al., 2018). Hippodeep was previously found to be
the most robust of the freely available methods for segmenting the hippocampus in people
with stroke pathology (Zavaliangos ‐Petropulu et al., 2020). Hippocampal segmentations
were visually inspected according to previously described protocols (Liew et al., 2020;
Zavaliangos ‐Petropulu et al., 2020). Any segmentations that were not properly
segmented were marked as failed and excluded from the analysis. 18 ipsilesional and 5
contralesional hippocampal segmentations failed quality control and were excluded from
the analysis.
To account for differences in head size, hippocampal volume was normalized for
head size by taking the ratio of hippocampal volume to eTIV for each participant and
multiplying it by the average eTIV across the sample, as done in previous studies of post-
stroke hippocampal volume (Schaapsmeerders et al., 2015; Tang et al., 2012; van
Norden et al., 2008).
3.3.3 Manually Segmented Lesions
Lesions were manually segmented on the T1w MRI by B.L., M.D., J.S., A.Z.P., and
S-L.L. according to an adapted version of the Anatomical Tracings of Lesions After Stroke
(ATLAS) protocol (Liew et al., 2017). Briefly, brain lesions were identified, and masks
were manually drawn on each individual brain in native space using ITK-Snap
(Yushkevich et al., 2006). Each lesion was checked for quality by at least two different
tracers. An expert neuroradiologist (G.B.) was also consulted to ensure lesion
segmentation quality and to differentiate stroke lesions from perivascular spaces.
63
Although all participants were listed by the providing research sites as having
unilateral lesions, additional secondary lesions were discovered in 100 participants while
manually tracing lesions, which may have occurred as a result of silent or prior strokes.
Secondary lesions were found in both hemispheres, the brainstem, and the cerebellum.
For the purpose of this paper, we refer to the primary lesioned hemisphere as the lesioned
hemisphere noted by the providing research site. We also performed follow-up analyses
excluding participants with secondary lesions, which did not significantly impact results.
Lesion probability maps were generated by nonlinearly normalizing lesion masks and
registering them to the MNI-152 template (Figure 3.1).
Finally, lesion volume was calculated by summing the voxels within each manually
traced lesion mask. Lesion size was also normalized for head size as previously
described for hippocampal volume in Methods Section 3.3.2. Lesion size was then log
transformed to normalize the distribution of the data.
64
Figure 3.1 Lesion Density Maps for Primary Lesions. Lesion density maps for primary
lesions from participants with cohort reported left and right hemisphere lesions are
overlaid on the MNI-152 template. Lesioned hemisphere refers to the primary lesion, as
reported by the research cohort. The color bar refers to the percentage of overlapping
lesions across participants
3.3.4 Statistical Analysis
3.3.4.a Lesion Size and Hippocampal Volume
To investigate associations between lesion size and hippocampal volume, we
performed a robust mixed effects regression with hippocampal volume as the dependent
variable. We tested ipsilesional and contralesional hippocampal volume separately.
Lesion size, age, sex (coded as a binary variable: women = 0, men = 1), and lesioned
hemisphere (coded as binary variable: left hemisphere lesion = 0.5, right hemisphere
lesion = 1.5) were included in the model as fixed effects, and cohort as a random effect.
Hippocampus ~ Lesion Size + Sex + Age + Lesioned Hemisphere + random (Cohort)
We applied a Bonferroni correction for two comparisons (ipsilesional, contralesional;
corrected p-value < 0.025).
65
To see if our findings remained significant after excluding participants with
secondary lesions, we also re-ran the model in a subset of the data in participants with
unilateral lesions determined by lesion mask findings (Figure 3.2).
We tested for associations between lesion size and hippocampal volume within
each cohort independently using a robust linear regression. We also performed a power
analysis to investigate what is the minimum sample size needed to detect associations
between lesion size and hippocampal volume.
Figure 3.2 Lesion Density Maps for Unilateral Lesions. Lesion density maps from
participants with unilateral left or right hemisphere lesions as determined by lesion masks
are overlaid on the MNI-152 template. In this supplementary analysis, participants with
secondary bilateral, brainstem, or cerebellar lesions are excluded.
3.3.4b Hippocampal Volume in Participants with Stroke versus Healthy Controls
A robust linear regression was used to test differences in hippocampal volume
between diagnoses (healthy control = 0, participant with stroke = 1). We analyzed
ipsilesional (left) and contralesional (right) hippocampi separately. Age and sex were
66
included in the model as fixed effects. Given the limited sample, the model could not
converge with cohort as a random effect, thus we also included cohort as a fixed effect.
Hippocampus ~ Diagnosis + Sex + Age + Lesioned Hemisphere + Cohort
We applied a Bonferroni correction for two comparisons (ipsilesional, contralesional;
corrected p-value < 0.025).
3.3.4.c Statistical Tools
All statistical analyses were performed in R (version 4.0.2; R Core Team, 2020).
We used the Mahalanobis distance to detect multivariate outliers, which were then
removed from the analyses. All mixed effects regressions were initially run as linear mixed
effects regressions (lmer function from nlme package). We ruled out collinearity for
variables in every model tested (variance inflation factor ≤ 2.5). We tested regression
assumptions of linearity, normality of the residuals, and homogeneity of the residual
variance by visually inspecting residuals versus fits plots as well as qq-plots. After
detecting influential observations using Cook’s distance in each analysis (Nieuwenhuis et
al., 2012), we repeated the analyses using robust mixed-effects regression. Robust mixed
effects regression (rlmer from the robustlmm package) avoids excluding data by reducing
the weight of influential observations (Greco et al., 2019). We therefore report the results
of the robust mixed effects regression. For all analyses, sample size (N), beta coefficients
for the factor of interest and confidence intervals (Beta(CI)), standard error (SE), t-value
and degrees of freedom (t(DF)), standardized effect size (d-value), and uncorrected p-
values were reported. We also used robust linear models (rlm) to test for associations
within each site.
67
3.4 Results
3.4.1 Lesion Effects on Hippocampal Volume
Larger lesion size was significantly associated with smaller ipsilesional (Beta = -
0.21, p-value < 0.001) but not contralesional hippocampal volume (Beta = -0.03, p-value
= 0.60), after adjusting for age, sex, lesioned hemisphere, and cohort (Table 3.3; Figure
3.3). Age and sex were significantly associated with both ipsilesional and contralesional
hippocampal volume, and lesioned hemisphere was significantly associated with
contralesional hippocampal volume only (Table 3.3). Lesion size remained significantly
associated with smaller ipsilesional hippocampal volume, even when excluding
participants with secondary lesions (Beta = -0.18, p-value = 0.003).
68
Table 3.3 Summary Statistics for Associations Between Lesion Size and
Hippocampal Volume. Summary statistics from robust mixed-effects regression to test
associations between ipsilesional hippocampal volume and lesion size (top) and
contralesional hippocampal volume and lesion size (bottom). The full model as well as
the beta coefficient (Beta) with 95% confidence interval (CI), standard error (SE), t-value
and degrees of freedom t(DF), standardized d-value, uncorrected p-value for all fixed
effect covariates are reported.
Hippocampus ~ Lesion Size + Sex + Lesioned Hemisphere + Age + random(Cohort)
Predictors Beta(CI) SE t(DF) d-value p-value
IPSILESIONAL HIPPOCAMPAL VOLUME (N=336; R
2
=0.33)
Lesion Size -0.21 (-0.31 - -0.12) 0.05 -4.23(334)
-0.46
<0.001
Sex -0.58 (-0.77 – -0.38) 0.10 -5.78(324)
-0.64
<0.001
Lesioned Hemisphere 0.16 (-0.03 – 0.36) 0.10 1.64(336)
0.18
0.10
Age -0.35 (-0.45 - -0.25) 0.05 -6.83(335)
-0.75
<0.001
CONTRALESIONAL HIPPOCAMPAL VOLUME (N=349; R
2
=0.30)
Lesion Size -0.03 (-0.12 – 0.07) 0.05 -0.53(348) -0.06 0.60
Sex -0.51 (-0.70 - -0.32) 0.10 -5.17(343) -0.56 <0.001
Lesioned Hemisphere -0.32 (-0.51 - -0.13) 0.10 -3.34(346) -0.36 0.001
Age -0.42 (-0.52 - -0.32) 0.05 -8.36(344) -0.90 <0.001
69
Figure 3.3 Lesion Size and Hippocampal Volume. (Top) effect sizes calculated with
standardized beta coefficients for associations between lesion size and ipsilesional (left)
and contralesional (right) hippocampal volume mapped onto a template brain. (Bottom)
Trend lines are plotted for the association between lesion size z-score (x-axis) and
hippocampal volumes z-score residuals (adjusted for age, sex, and eTIV) (y-axis) colored
by research cohort.
70
When testing for associations between lesion size and hippocampal volume within
each cohort individually, only Cohort 14 showed a significant association between lesion
size and hippocampal volume (p-value<0.005). Cohort 14 was one of the larger cohorts
(N=28) and had participants with a wide range of lesion sizes (Figure 3.4). A power
analysis determined that a sample of N=79 is necessary to achieve a power of 80%.
Figure 3.4 Associations Between Lesion Size and Ipsilesional Hippocampus by
Cohort. Boxplots of lesion size (left) and effect sizes (Beta; right) for associations
between hippocampal volume and lesion size by research cohort in order of sample size,
with the top being the smallest. Significant associations between diagnosis and
ipsilesional hippocampal volume were only detected after correcting for multiple
comparisons in Cohort 14 (p-value = 0.005), highlighted in green. Two additional sites
had p-values < 0.05 (Cohort 3 and Cohort 15, highlighted in yellow) but were not
significant after correcting for multiple comparisons.
3.4.2 Hippocampal Volume in Participants with Stroke versus Healthy Controls
Ipsilesional hippocampal volume in participants with stroke was significantly
smaller than averaged hippocampal volume in healthy controls (Beta = -0.59, p-value =
71
0.004). Contralesional hippocampal volume in participants with stroke was not
significantly different from healthy controls (Beta = -0.28, p-value = 0.21)(Figure 3.5).
Figure 3.5 Hippocampal Volume in Participants with Stroke vs Healthy Control
(Top) effect sizes calculated with standardized beta coefficients for ipsilesional (left) and
contralesional (right) hippocampal volume differences between participants with stroke
and age, sex, and cohort matched healthy controls mapped on to a template brain.
(Bottom) Boxplots comparing ipsilesional (left) and contralesional (right) hippocampal
volume z-score in participants with stroke and age, sex, and cohort matched healthy
controls.
3.5 Discussion
In this study, we investigated associations between lesion size, diagnosis, and
hippocampal volume in participants with chronic stroke from the ENIGMA Stroke
72
Recovery database. We found larger lesion size and stroke diagnosis were both
significantly associated with smaller ipsilesional hippocampal volume.
In our lesion analysis, we found larger lesion size was significantly associated with
smaller hippocampal volume, but only within the lesioned hemisphere. However, this
association only became apparent when pooling the multi-site data. When evaluating
cohorts individually, lesion size was significantly associated with ipsilesional hippocampal
volume in only one cohort. Our power analysis suggested that controversy in the literature
might be attributed to statistical power. Xie et al., 2011 and Tang et al., 2012 did not find
significant associations between lesion size and hippocampal volume, but both had
samples of less than 50 participants. Our pooled findings are similar to Schaapsmeerders
et al., (2015), which had a sample of 170 participants. This highlights the importance of
large, diverse samples necessary to detect associations between lesion size and
hippocampal volume. Additionally, we found that lesion size was significantly associated
with ipsilesional hippocampal volume regardless of whether or not secondary bilateral or
brainstem lesions were included. Further research is necessary to investigate the impact
of secondary lesions on hippocampal damage, as Werden et al., 2017 reports that
patients with recurrent stroke had smaller hippocampal volume compared to first-time
stroke.
In our supplemental analysis looking at hippocampal volume differences by
diagnosis, we found that only ipsilesional hippocampal volume in participants with stroke
was significantly smaller than hippocampal volume in healthy controls. To our knowledge,
this is the first study to estimate post-stroke hippocampal volumes with Hippodeep.
Despite the limited sample size, we were successfully able to reproduce previous findings
73
from a growing body of literature that report significantly smaller ipsilesional hippocampal
volumes in patients with stroke when compared to healthy controls (Brodtmann et al.,
2020; Haque et al., 2019; Khlif et al., 2018; Schaapsmeerders et al., 2015; Tang et al.,
2012; Xie et al., 2011).
Overall, our findings suggest that smaller hippocampal volumes observed in stroke
patients are specific to stroke-related damage within the lesioned hemisphere and are
likely not attributable to general age-related atrophy (Small et al., 2011) or other stroke
risk factors such as hypertension (Fiford et al., 2020) or changes in estrogen (Albert et
al., 2018), which are typically observed bilaterally. Researchers have speculated about a
number of different mechanisms that may underlie secondary degeneration of the
hippocampus. Wang et al., 2004 proposes that symptoms of ischemic injury such as brain
edema, swelling, and increased intracerebral pressure may constrict the hippocampus
and limit its blood supply. Secondary degeneration could also be a result of
disconnectivity, where hippocampal atrophy could be caused by retrograde degeneration
from the destruction of connected pathways (Xie et al., 2011, Chen et al., 2015).
Spreading depression (SD) might also explain secondary degeneration of the ipsilesional
hippocampus, where neurotoxic signals from the core of the lesion propagate to adjacent
grey matter regions and cause damage (Sueiras et al., 2021; Urbach et al., 2017). The
ionic imbalance that results from a blockage of blood supply during the acute phase of a
stroke causes a buildup of extracellular glutamate that leads to a self-propagating wave
of cell depolarization throughout neighboring gray matter (Sueiras et al., 2021; Urbach et
al., 2017). The hippocampus is filled with tightly packed, easily excitable glutamatergic
neurons and a high density of NMDA receptors (Nikonenko et al., 2009), making it more
74
susceptible to damage from SD. Over-excitation of the hippocampal glutamatergic
network leads to hippocampal excitotoxicity, resulting in hippocampal neuron apoptosis,
which is thought to be reflected on a macro scale as reduced hippocampal volume (Tang
et al., 2012). The damaging effects of SD are likely restricted to the lesioned hemisphere
because SD waves do not propagate easily through white matter (Chung et al., 2016),
therefore the waves cannot easily traverse to the contralesional hippocampus. However,
a magnetic resonance spectroscopy study reported evidence of hippocampal neuronal
loss in the contralesional hippocampus, although it was less severe and not detectable
using volumetric MRI (Tang et al., 2012). SD is still not well understood; therefore
contralesional hippocampal damage may be caused by mild SD or might be attributed to
other forms of secondary degeneration such as diaschisis (Zhang et al., 2012). The
available evidence in this study is insufficient to support SD as the key cause of reduced
ipsilesional hippocampal volumes observed in this study, however it cannot be excluded.
3.6 Limitations
Although secondary lesions were discovered while manually tracing lesion masks,
our findings did not change when participants with secondary lesions were excluded.
Further research is necessary to investigate the impact of lesion location on the
association between hippocampal volume and sensorimotor impairment.
This study lacks information about stroke risk factors that are thought to influence
hippocampal volume such as hypertension (Fiford et al., 2020) and changes in estrogen
(Albert et al., 2018). Furthermore, the current sample is cross-sectional and cannot
account for the extent of longitudinal hippocampal atrophy that may have occurred as a
75
result of stroke, pre-existing dementia, or stroke risk factors. However, the current cross-
sectional analysis can help guide what questions to take to longitudinal data collection,
which is very costly and time consuming.
3.7 Conclusion
We provide supporting evidence to existing literature that reduced hippocampal
volume is likely a consequence of stroke-related damage within the lesioned hemisphere.
We show the feasibility of multi-site data pooling using Hippodeep segmentations that
may be used for future research of the post-stroke hippocampus.
76
Chapter 4: Associations between sensorimotor impairment and
hippocampal volume in chronic stroke survivors
This section is adapted from:
Zavaliangos-Petropulu A, …. Liew S-L (2021) Associations between hippocampal
volume and sensorimotor impairment in chronic stroke survivors: An ENIGMA Stroke
Recovery Analysis. In preparation
4.1 Abstract
Long-lasting sensorimotor impairments after stroke can negatively impact quality of life.
The hippocampus is involved in sensorimotor circuits but has not been widely studied
within the context of post-stroke sensorimotor impairment. The purpose of this paper is
to investigate the association between non-lesioned hippocampal volumes and
sensorimotor impairment in people with chronic stroke. We pooled cross-sectional T1-
weighted brain MRIs from 357 participants with chronic stroke from 18 research cohorts
of the ENIGMA Stroke Recovery Working Group (age: median = 61 years, interquartile
range = 18, range = 23-93; 135 women and 222 men). Sensorimotor impairment was
estimated using the Fugl-Meyer Assessment of Upper Extremities. We used robust
mixed-effects models to test associations between post-stroke sensorimotor impairment
and non-lesioned hippocampal volumes (ipsilesional and contralesional separately;
Bonferroni-corrected, p-value < 0.025), controlling for age, sex, lesion volume, and
lesioned hemisphere. We also performed an exploratory analysis of whether sex
77
differences influence the relationship between sensorimotor impairment and hippocampal
volume. Greater sensorimotor impairment was significantly associated with smaller
ipsilesional (p-value = 0.001) but not contralesional (p = 0.05) hippocampal volume,
independent of lesion volume or other covariates. We also found preliminary evidence of
a sensorimotor impairment by sex interaction for both ipsilesional (p = 0.008) and
contralesional (p = 0.006) hippocampal volumes, where women showed progressively
smaller hippocampal volumes with worsening sensorimotor impairment compared to
men. We demonstrate a novel association between chronic post-stroke sensorimotor
impairment and ipsilesional, but not contralesional, hippocampal volume. This finding is
not due to lesion size and may be stronger in women.
4.2. Introduction
Sensorimotor impairments are a major burden of disease in stroke survivors
(Gittler and Davis, 2018). For clinicians, caregivers, and patients to make informed
rehabilitation treatment decisions, accurate predictions of a patient’s potential for
sensorimotor recovery are necessary (Boyd et al., 2017). Therefore, there is a critical
need to identify reliable biomarkers of stroke sensorimotor impairment(Stinear, 2017).
Magnetic resonance imaging (MRI) studies of regional brain volumes suggest secondary
degeneration of adjacent or remote regions may contribute to sensorimotor impairment
and could be promising biomarkers of post-stroke sensorimotor outcomes (Liew et al.,
2021; Zhang et al., 2012).
The hippocampus is a brain region that is particularly vulnerable to post-stroke
secondary degeneration. Both rodent (Xie et al., 2011) and human (Brodtmann et al.,
78
2020; Haque et al., 2019; Khlif et al., 2019b; Schaapsmeerders et al., 2015; Tang et al.,
2012) stroke studies of non-lesioned hippocampi show evidence of severe ipsilesional
hippocampal damage. Structural magnetic resonance imaging (MRI) studies report
smaller ipsilesional hippocampal volumes in chronic stroke patients, on average,
compared to healthy controls (Brodtmann et al., 2020; Haque et al., 2019; Khlif et al.,
2019b; Schaapsmeerders et al., 2015; Tang et al., 2012). Studies have also reported
magnetic resonance spectroscopy evidence of contralesional hippocampal neuronal loss
(Tang et al., 2012) and contralesional longitudinal hippocampal atrophy measured with
MRI (Brodtmann et al., 2020). As stroke-related infarcts in the hippocampus are
uncommon (Chen et al., 2015; Szabo et al., 2009), post-stroke hippocampal atrophy is
attributed to secondary degeneration mechanisms such as spreading depression (Xie et
al., 2011) or reduced connectivity to lesioned structures (Chen et al., 2015).
As the hippocampus is widely known for its key role in memory, it is not surprising
that the primary focus of post-stroke hippocampal studies has been on the role of
hippocampal damage in cognitive impairment (Schaapsmeerders et al., 2015; Tang et
al., 2012; Xie et al., 2011). Although not typically considered a primary sensorimotor
region, there is evidence that the hippocampus is also involved in sensorimotor circuits.
The hippocampus is densely connected to important sensorimotor areas such as the
thalamus and basal ganglia through the spinal-limbic pathway (Maller et al., 2019).
Reports of hippocampal activity during sensorimotor behavior such as sensorimotor
integration (Suzuki, 2007), sensorimotor learning (Albouy et al., 2008; Burman, 2019;
Jacobacci et al., 2020), and motor control (Burman, 2019)
suggest the hippocampus plays
a role in sensorimotor circuits. Functional hippocampal connectivity in relation to
79
sensorimotor behavior has also been reported with the thalamus (Baumgartner et al.,
2018), sensorimotor cortex (Burman, 2019), and the supplementary motor area (Mukamel
et al., 2010). However, the relationship between hippocampal integrity and post-stroke
sensorimotor impairment remains unclear. Given the involvement of the hippocampus in
sensorimotor circuits, hippocampal damage due to secondary degeneration after stroke
could further weaken sensorimotor circuits, leading to worsened chronic sensorimotor
impairment. Alternatively, damage to the thalamus, basal ganglia, sensorimotor cortex,
or supplementary motor area, which are typically associated with greater sensorimotor
impairment, may lead to downstream degeneration of the hippocampus through
functional or structural connections. In this manuscript, we took a first step towards
examining this potential relationship using cross-sectional data to assess whether there
is an association between the volume of the non-lesioned post-stroke hippocampus and
sensorimotor impairment.
Additionally, data in dementia (Nebel et al., 2018) and healthy ageing (Wierenga
et al., 2020) research suggests that hippocampal atrophy may accelerate in women over
a certain age. Estrogen levels may play a mediating role in these trends (Albert et al.,
2017) and have been associated with stroke severity and mortality (Pappa et al., 2012).
Stroke-related outcomes including disability and quality of life are generally poorer in
women than men (Cordonnier et al., 2017; Dehlendorff et al., 2015; Gittler and Davis,
2018), although conclusive sex differences have not been reported in terms of post-stroke
sensorimotor impairment (Hawe et al., 2020b). Since both stroke severity (Cordonnier et
al., 2017; Dehlendorff et al., 2015; Gittler and Davis, 2018) and dementia-related
hippocampal atrophy (Nebel et al., 2018) are thought to be greater in women compared
80
to men, sex could moderate the relationship between sensorimotor impairment and
hippocampal volume. In particular, women may show a wider range of post-stroke
hippocampal volumes and sensorimotor impairment scores, potentially leading to
stronger effect sizes for associations as compared to men.
In this study, we aimed to investigate the relationship between sensorimotor
impairment and ipsilesional and contralesional hippocampal volumes (separately) in 357
participants with chronic stroke across 18 cohorts from the ENIGMA Stroke Recovery
Working Group (Liew et al., 2020). First, we investigated associations between
sensorimotor impairment and hippocampal volume, controlling for lesion size and
additional covariates of age, sex, and lesioned hemisphere. The Fugl-Meyer Assessment
of Upper Extremities (FMA-UE) was used to measure sensorimotor impairment (See et
al., 2013). We hypothesized that greater post-stroke sensorimotor impairment would be
correlated with smaller ipsilesional but not contralesional hippocampal volume. In an
exploratory analysis, we tested to see if sex had a moderating effect on the relationship
between sensorimotor impairment and hippocampal volume. Due to more severe
hippocampal vulnerability and poorer stroke outcome trends in women, we hypothesized
that women would have a stronger relationship between greater sensorimotor impairment
and smaller hippocampal volume than men. Finally, we hypothesized that due to the
involvement of the hippocampus in sensorimotor circuits, associations between
ipsilesional hippocampal volume and sensorimotor impairment would be independent of
lesion size.
81
4.3 Methods
4.3.1 ENIGMA Stroke Recovery Dataset
We used cross-sectional data from the ENIGMA Stroke Recovery Working Group
database available as of December 15, 2020. Details of the ENIGMA Stroke Recovery
procedures and methods are available in Liew et al., 2020. The data were collected
across 18 research studies (cohorts) across 10 different research institutes in 6 countries.
ENIGMA Stroke Recovery participants with the following data were included: high
resolution (1-mm isotropic) T1-weighted brain MRI (T1w) acquired with a 3T MRI scanner,
Fugl-Meyer Upper Extremity score (FMA-UE; On a scale from 0-66: 0 = severe
sensorimotor impairment, 66 = no sensorimotor impairment), age, and sex. As we were
interested in studying effects of secondary degeneration of the hippocampus, we only
included participants with chronic stroke (defined as data acquired at least 180 days post-
stroke (Bernhardt et al., 2017)). Exclusion criteria included site-reported bilateral,
brainstem, or cerebellar lesions, participants with no identifiable lesions, and participants
with no sensorimotor impairment (FMA-UE= 66). In addition, each hippocampus was
visually inspected with lesion masks overlaid and any brains with hippocampal lesions
were excluded. The total initial sample size was N=357 (age: median = 61 years,
interquartile range (IQR) = 18, range = 23-93; FMA-UE: median = 41, IQR = 28, range=
0-65; 135 women and 222 men) (Table 4.1).
82
Table 4.1 ENIGMA Stroke Recovery Demographics Demographics for ENIGMA Stroke
Recovery Working Group participants included in the study by cohort. Total sample size
(N), number of women and men, and information about age and Fugl-Meyer Assessment
of Upper Extremities (FMA-UE) are listed.
Cohort N Women/Men
Median Age (years)
(IQR, min-max)
Median FMA-UE
(IQR, min-max)
Cohort 1 39 10/29 61 (17, 31-80) 43 (16, 0-58)
Cohort 2 12 6/6 69.5 (12, 39-85) 33 (27, 13-48)
Cohort 3 15 6/9 61 (17, 33-85) 16 (13, 5-40)
Cohort 4 19 6/13 44 (15, 30-68) 10 (11, 1-34)
Cohort 5 28 12/16 64 (18, 44-81) 52 (33, 8-65)
Cohort 6 10 3/7 61 (12.5, 49-72) 65 (3, 45-65)
Cohort 7 14 5/9 58 (12, 45-69) 63 (14, 6-65)
Cohort 8 11 4/7 56 (12, 45-74) 48 (15, 25-55)
Cohort 9 11 3/8 59 (3, 45-68) 38 (18, 15-49)
Cohort 10 8 4/4 58 (8, 46-73) 48 (16, 35-59)
Cohort 11 22 6/16 61.5 (11, 23-75) 49 (22, 23-64)
Cohort 12 13 4/9 57 (13, 32-80) 54 (15, 38-63)
Cohort 13 12 4/8 66 (16, 31-83) 51 (26, 19-62)
Cohort 14 29 18/11 50 (15, 25-79) 41 (13, 24-53)
Cohort 15 10 3/7 61.5 (11, 42-76) 29 (16, 11-60)
Cohort 16 40 14/26 66.5 (11, 43-93) 47 (30, 4-65)
Cohort 17 36 15/21 70 (14, 37-80) 53 (27, 8-65)
Cohort 18 28 12/16 64 (14, 34-85) 27 (5, 14-34)
Total 357 135/222 61 (18, 23-93) 41 (28, 0-65)
83
4.3.2 MRI Data Analysis
Hippodeep, a convolutional neural network-based hippocampal segmentation
application, was used to produce hippocampal volume and total intracranial volume
(eTIV) estimates from the T1w MRI (Thyreau et al., 2018). Hippodeep was previously
found to be the most robust of the freely available methods for segmenting the
hippocampus in people with stroke pathology (Zavaliangos ‐Petropulu et al., 2020).
Hippocampal segmentations were visually inspected according to previously described
protocols (Liew et al., 2020; Zavaliangos ‐Petropulu et al., 2020). Any segmentations that
were not properly segmented were marked as failed and excluded from the analysis. 18
ipsilesional and 5 contralesional hippocampal segmentations failed quality control and
were excluded from the analysis.
To account for differences in head size, hippocampal volume was normalized for
head size by taking the ratio of hippocampal volume to eTIV for each participant and
multiplying it by the average eTIV across the sample, as done in previous studies of post-
stroke hippocampal volume (Schaapsmeerders et al., 2015; Tang et al., 2012; van
Norden et al., 2008).
4.3.3 Manually Segmented Lesions
Lesions were manually segmented on the T1w MRI by B.L., M.D., J.S., A.Z.P., and
S-L.L. according to an adapted version of the Anatomical Tracings of Lesions After Stroke
(ATLAS) protocol (Liew et al., 2018). For more information on lesion tracing, see Chapter
3.3.3. Due to secondary lesion discovered while tracing, we performed follow-up analyses
excluding participants with secondary lesions, which did not significantly impact results.
84
Lesion volume was calculated by summing the voxels within each manually traced
lesion mask. Lesion size was also normalized for head size as previously described for
hippocampal volume in Methods Section 4.3.2. Lesion size was then log transformed to
normalize the distribution of the data.
4.3.4 Statistical Analysis
First, we tested our main hypothesis that greater post-stroke sensorimotor
impairment is correlated with smaller ipsilesional but not contralesional hippocampal
volume, by performing a robust mixed-effects regression with hippocampal volume as the
dependent variable. Sensorimotor impairment was measured using the Fugl-Meyer
Assessment of Upper Extremities (FMA-UE)
31
FMA-UE, sex (coded as a binary variable:
women = 0, men = 1), age, and lesioned hemisphere (coded as binary variable: left
hemisphere lesion = 0.5, right hemisphere lesion = 1.5) were included in the model as
fixed effects and cohort was included in the model as a random effect.
Hippocampus ~ FMA-UE + Sex + Age + Lesioned Hemisphere + random (Cohort)
Next, we tested our exploratory hypothesis that sex may have a moderating
effect on the relationship between sensorimotor impairment and hippocampal volume by
including a FMA-UE*Sex interaction covariate as a fixed effect.
Hippocampus ~ FMA-UE*Sex + FMA-UE + Sex + Age + Lesioned Hemisphere + random (Cohort)
We also tested for sex differences in sensorimotor impairment, lesion size, and age using
an independent t-test.
Finally, we tested our hypothesis that sensorimotor impairment is independently
associated with hippocampal volume by including lesion size as a fixed covariate.
Hippocampus ~ Lesion Size + FMA-UE*Sex + FMA-UE + Sex + Age + Lesioned Hemisphere + random (Cohort)
85
For each model, we analyzed ipsilesional and contralesional hippocampi
separately. We applied a Bonferroni correction for two comparisons (ipsilesional,
contralesional; corrected p-value < 0.025).
All statistical analyses were performed in R (version 4.0.2; R Core Team, 2020).
We used the Mahalanobis distance to detect multivariate outliers, which were then
removed from the analyses. All mixed effects regressions were initially run as linear mixed
effects regressions (lmer function from nlme package). We ruled out collinearity for
variables in every model tested (variance inflation factor ≤ 2.5). We tested regression
assumptions of linearity, normality of the residuals, and homogeneity of the residual
variance. After detecting influential observations using Cook’s distance in each analysis
(Nieuwenhuis et al., 2012), we repeated the analyses using robust mixed-effects
regression. Robust mixed effects regression (rlmer from the robustlmm package) avoids
excluding data by reducing the weight of influential observations (Greco et al., 2019). We
therefore report the results of the robust mixed effects regression. For all analyses,
sample size (N), beta coefficients for the factor of interest and confidence intervals
(Beta(CI)), standard error (SE), t-value and degrees of freedom (t(DF)), standardized
effect size (d-value), and uncorrected p-values were reported.
4.4 Results
4.4.1 Hippocampal Volume and Sensorimotor Impairment
Greater sensorimotor impairment was significantly associated with smaller
ipsilesional (Beta = 0.31, p-value = 0.005, R
2
= 0.16) but not contralesional (Beta = 0.003,
86
p-value = 0.96, R
2
= 0.29) hippocampal volume after adjusting for age, sex, lesioned
hemisphere, and cohort (Table 4.2). We observed a better model fit and an increase in
effect size increased for the association between sensorimotor impairment and
ipsilesional hippocampal volume increased (Beta = 0.31, p-value < 0.001, R
2
= 0.30) when
including FMA-UE*Sex interaction as a covariate (Table 4.3). Furthermore, FMA-UE
remained independently associated with ipsilesional hippocampal volume after including
lesion size in the model (Beta = 0.26, p-value = 0.001, R
2
= 0.33; Table 4.4, Figure 4.1).
This association was still significant, even when excluding participants with secondary
lesions (Beta = 0.30, p-value = 0.001, R
2
= 0.35).
87
Table 4.2. Summary Statistics for Associations between Sensorimotor Impairment
and Hippocampal Volume. Summary statistics from robust mixed-effects regression to
test associations between ipsilesional hippocampal volume and sensorimotor impairment
(top) and contralesional hippocampal volume and sensorimotor impairment (bottom). The
full model as well as the sample size (N), conditional R
2
, beta coefficient (Beta) with 95%
confidence interval (CI), standard error (SE), t-value and degrees of freedom t(DF),
standardized d-value, uncorrected p-value for all fixed effect covariates are reported.
Hippocampus ~ FMA-UE*Sex + Lesioned Hemisphere + Age + random(Cohort)
Predictors Beta(CI) SE t(DF) d-value p-value
IPSILESIONAL HIPPOCAMPAL VOLUME (N=336; R
2
=0.27)
FMA-UE 0.16(0.05 – 0.27) 0.06 2.80(287)
0.33
0.005
Sex -0.53(-0.73 - -0.33) 0.10 -5.21(324)
-0.58
<0.001
Lesioned Hemisphere 0.19(-0.01 – 0.39) 0.10 1.84(336)
0.20
0.06
Age -0.32(-0.42 - -0.22) 0.05 -6.16(335)
-0.67
<0.001
CONTRALESIONAL HIPPOCAMPAL VOLUME (N=349; R
2
=0.29)
FMA-UE 0.003(-0.10 – 0.11) 0.05 0.05(238)
0.01
0.96
Sex -0.50(-0.69 - -0.31) 0.10 -5.14(343)
-0.56
<0.001
Lesioned Hemisphere -0.32(-0.51 - -0.13) 0.10 -3.30(346)
-0.35
0.001
Age -0.41(-0.51 – 0.32) 0.05 -8.30(346)
-0.89
<0.001
88
Figure 4.1. Associations between Sensorimotor Impairment and Hippocampal
Volume. Effect sizes (standardized Beta values) for ipsilesional and contralesional
hippocampi are mapped onto a template for associations between hippocampal volumes
and sensorimotor impairment, with warmer colors representing stronger negative
associations (left). A trend line (black line) is plotted for the association between
ipsilesional hippocampal volume z-scores FMA-UE z-scores (right). Scatter plot points
are colored by research cohort.
89
Table 4.3 Summary Statistics for Associations Between Sensorimotor
Impairment*Sex and Hippocampal Volume. Summary statistics from robust mixed-
effects regression to test associations between ipsilesional hippocampal volume and
sensorimotor impairment (top) and contralesional hippocampal volume and sensorimotor
impairment (bottom). The full model as well as the sample size (N), conditional R
2
, beta
coefficient (Beta) with 95% confidence interval (CI), standard error (SE), t-value and
degrees of freedom t(DF), standardized d-value, uncorrected p-value for all fixed effect
covariates are reported.
Hippocampus ~ FMA-UE*Sex + Lesioned Hemisphere + Age + random(Cohort)
Predictors Beta(CI) SE t(DF) d-value p-value
IPSILESIONAL HIPPOCAMPAL VOLUME (N=336; R
2
=0.30)
FMA-UE 0.31 (0.15 – 0.46) 0.08 3.86(336) 0.42 <0.001
FMA-UE*Sex -0.26 (-0.46 - -0.07) 0.10 -2.61(324) -0.29 0.009
Sex -0.53 (-0.73 - -0.33) 0.10 -5.29(336) -0.58 <0.001
Lesioned Hemisphere 0.17 (-0.03 – 0.37) 0.10 1.69(335) 0.18 0.09
Age -0.32 (-0.42 - -0.22) 0.05 -6.27(332) -0.69 <0.001
CONTRALESIONAL HIPPOCAMPAL VOLUME (N=349; R
2
=0.32)
FMA-UE 0.16 (0.01 – 0.31) 0.08 2.06(334) 0.23 0.04
FMA-UE*Sex -0.27 (-0.46 - -0.08) 0.10 -2.76(348) -0.30 0.006
Sex -0.51 (-0.70 - -0.32) 0.10 -5.28(343) -0.57 <0.001
Lesioned Hemisphere -0.35 (-0.54 - -0.16) 0.10 -3.58(343) -0.39 <0.001
Age -0.41 (-0.51 - -0.32) 0.05 -8.38(344) -0.90 <0.001
90
Table 4.4 Summary Statistics for Associations Between Sensorimotor Impairment
and Hippocampal Volume When Including Lesion Size. Summary statistics from
robust mixed-effects regression to test associations between ipsilesional hippocampal
volume and sensorimotor impairment (top) and contralesional hippocampal volume and
sensorimotor impairment (bottom) when including lesion size as a covariate. The full
model as well as the beta coefficient (Beta) with 95% confidence interval (CI), standard
error (SE), t-value and degrees of freedom t(DF), standardized d-value, uncorrected p-
value for all fixed effect covariates are reported.
Hippocampus ~ Lesion Size + FMA-UE*Sex + Lesioned Hemisphere + Age + random(Cohort)
Predictors Beta(CI) SE t(DF) d-value p-value
IPSILESIONAL HIPPOCAMPAL VOLUME (N=336; R
2
=0.33)
FMA-UE 0.26 (0.10 – 0.41) 0.08 3.28(332) 0.36 0.001
FMA-UE*Sex -0.26 (-0.45 - -0.07) 0.10 -2.65(332) -0.29 0.008
Lesion Size -0.19 (-0.29 - -0.09) 0.05 -3.75(333) -0.41 <0.001
Sex -0.58 (-0.78 - -0.39) 0.10 -5.91(325) -0.66 <0.001
Lesioned Hemisphere 0.17 (-0.03 – 0.36) 0.10 1.69(336) 0.18 0.09
Age -0.36 (-0.46 - -0.26) 0.05 -7.05(336) -0.77 <0.001
CONTRALESIONAL HIPPOCAMPAL VOLUME (N=349; R
2
=0.32)
FMA-UE 0.15 (0.00 – 0.30) 0.08 1.94(338) 0.21 0.05
FMA-UE*Sex -0.27 (-0.46 - -0.08) 0.10 -2.77(348) -0.30 0.006
Lesion Size -0.03 (-0.13 – 0.07) 0.05 -0.56(349) -0.06 0.58
Sex -0.52 (-0.71 - -0.33) 0.10 -5.30(343) -0.57 <0.001
Lesioned Hemisphere -0.35 (-0.54 - -0.16) 0.10 -3.59(343) -0.39 <0.001
Age -0.42 (-0.52 - -0.32) 0.05 -8.39(343) -0.91 <0.001
91
4.4.2 Sex Effects on the Association between Hippocampal Volume and Sensorimotor
Impairment
A t-test revealed no significant differences in FMA-UE (t(260) = -1.12, p-value =
0.26) or age (t(249) = 1.12, p-value = 0.26) between women and men, however women
did have significantly larger lesions than men (t(277) = 2.9, p-value = 0.004) (Figure 4.2).
The FMA-UE*sex interaction was a significant covariate for both ipsilesional (Beta = -
0.26, p-value = 0.009, R
2
= 0.30) and contralesional (Beta = -0.27, p-value = 0.006, R
2
=
0.32) hippocampal volume (Figure 4.2, Table 4.3), even after accounting for lesion size
(ipsilesional: Beta = -0.26, p-value = 0.001, R
2
= 0.33; contralesional: Beta = -0.27, p-
value = 0.006, R
2
= 0.32; Table 4.4). In the ipsilesional hippocampus, women had a
positive slope (β = 0.26) and men had a negative slope close to 0 (β = -0.002). In the
contralesional hippocampus, women had a positive slope (β = 0.15) while men had a
negative slope (β = -0.12). FMA-UE*sex interaction remained significantly associated with
both ipsilesional (Beta = -0.31, p-value = 0.008, R
2
= 0.35) and contralesional (Beta = -
0.28, p-value = 0.017, R
2
= 0.32) hippocampal volumes, even when excluding participants
with secondary lesions.
92
Figure 4.2. Sensorimotor Impairment*Sex Interaction Findings. Trend lines are
plotted for the association between FMA-UE z-score (x-axis) and hippocampal volumes
z-score (y-axis) for women (red) and men (blue) calculated from the FMA-UE*Sex
interactions. Histograms for FMA-UE scores (bottom left), age (bottom middle), and lesion
size (bottom right) are plotted by sex.
4.5 Discussion
In this study, we investigated associations between sensorimotor impairment and
hippocampal volume in participants with chronic stroke from the ENIGMA Stroke
Recovery database. We found greater sensorimotor impairment was significantly
associated with smaller ipsilesional hippocampal volume, an association that was
primarily driven by the population of women in the sample.
93
To our knowledge, this is the first study to report associations between
hippocampal volume and sensorimotor impairment in chronic stroke patients. We found
that greater sensorimotor impairment was independently associated with smaller
ipsilesional hippocampal volume, even after adjusting for lesion size, suggesting that
ipsilesional hippocampal integrity may be related to sensorimotor impairment. In addition
to hippocampal damage incurred by spreading depression (see Chapter 3 Discussion),
ipsilesional disruption to sensorimotor circuits may cause secondary degeneration of the
ipsilesional hippocampus, possibly as a result of anatomical connectivity (e.g., through
the thalamus (Baumgartner et al., 2018), basal ganglia (Maller et al., 2019), sensorimotor
cortex (Burman, 2019), or supplementary motor area (Mukamel et al., 2010)).
Furthermore, as the hippocampus is an important limbic system structure, it is heavily
involved in learning, memory, and emotion (Small et al., 2011). Post-stroke cognitive
impairment (Dichgans, 2019; Quinn et al., 2018), depression (Quinn et al., 2018), and
anxiety (Chun et al., 2018) are all common pervasive symptoms in stroke survivors that
interfere with rehabilitation and are associated with poor stroke outcomes (Dichgans,
2019; Quinn et al., 2018). Limbic system disruption caused by secondary post-stroke
hippocampal damage may cause cognitive impairment or aggravate symptoms of
depression and anxiety, which in turn may interfere with stroke sensorimotor rehabilitation
efforts. Further functional and longitudinal research is necessary to understand the impact
of hippocampal atrophy on sensorimotor circuits and sensorimotor rehabilitation.
In our exploratory analysis, we found significant sex differences in the association
between FMA-UE and bilateral hippocampal volume, where women showed
progressively smaller hippocampal volumes with increasing sensorimotor impairment
94
compared to men. This suggests that women with greater sensorimotor impairment may
also have more hippocampal damage compared to men. In addition, neither age nor
FMA-UE significantly differed by sex, suggesting that the sex differences we observed in
the relationship between sensorimotor impairment and the hippocampus cannot be
attributed to age or impairment differences. Although lesion size was significantly larger
in women, the FMA-UE*sex interaction covariate was independently associated with
hippocampal volume, even when accounting for lesion size. Overall, these findings should
be considered exploratory given the unequal number of men and women in the sample.
Further research is also needed to confirm these findings, as our sample was unable to
account for additional variables (Lisabeth et al., 2015; Reeves et al., 2008) thought to
influence the hippocampus in a sex-dependent way such as estrogen levels (Albert et al.,
2017), dementia (Ferretti et al., 2018; Fleisher et al., 2005; Zheng et al., 2017), and
depression (Shi et al., 2017). Furthermore, the extent to which sex differences observed
in stroke research are a result of physiological differences between sexes versus different
contextual factors such as treatment received by women post-stroke remains unclear
(Christensen and Bushnell, 2020; Dehlendorff et al., 2015). Further research on sex
differences in stroke is crucial to improve our understanding of the relationship between
hippocampal damage and sensorimotor impairment.
4.6 Limitations and Future Directions
In this paper, we only considered gross hippocampal volume. However, the
hippocampus is composed of structurally and functionally distinct subfields, each
differentially vulnerable to disease (Iglesias et al., 2015; Small et al., 2011). Structurally,
95
reduced neuron density has been observed in the cells of the CA1 but not CA2 subfield
of post-mortem stroke patients when compared to controls (Gemmell et al., 2012).
Functionally, while the posterior extents of the hippocampus along the long axis are
thought to be more involved with memory and cognitive processing, the anterior extents
have been implicated in sensorimotor integration (Small et al., 2011). Further research
investigating sensorimotor impairment and the hippocampus at a finer resolution, such as
at the level of hippocampal subfields (Iglesias et al., 2015) or vertex wise associations
(Hibar et al., 2017), may reveal more specific and robust relationships.
In addition, although secondary lesions were discovered while manually tracing
lesion masks, our findings did not change when participants with secondary lesions were
excluded. Further research is necessary to investigate the impact of lesion location on
the association between hippocampal volume and sensorimotor impairment.
Given the focus on hippocampal volumes, another limitation of this study is the
lack of cognitive and depression data. While cognitive and depressive scores are
available for a small number of cohorts in the ENIGMA Stroke Recovery database, the
participants with available data have very limited information. Many of the participating
stroke sensorimotor rehabilitation research studies also used cognitive impairment as an
exclusion criteria (Lingo VanGilder et al., 2020), resulting in participants with no or mild
cognitive deficits.
Finally, the current sample is cross-sectional and lacks information on the type or
dose of rehabilitation treatment received. Our cross-sectional sample also cannot account
for the extent of longitudinal hippocampal atrophy that may have occurred as a result of
stroke, pre-existing dementia, or normal aging. However, the current cross-sectional
96
analysis serves as a first step to examining the relationship between hippocampal
volumes, sensorimotor impairment, lesion volume and sex and can be used to guide
future questions to explore in a longitudinal study.
4.7 Conclusion
Our findings demonstrate a novel association between chronic post-stroke
sensorimotor impairment and hippocampal volume that may be modulated by sex.
Overall, these findings provide unique insight into the role that the hippocampus may play
in post-stroke sensorimotor impairment.
97
Chapter 5: Discussion
In this dissertation, I set the groundwork for studying post-stroke secondary
degeneration of the hippocampus using pooled data from the ENIGMA Stroke Recovery
database. This work represents an international effort to promote collaborative science
with accessible and importantly, quality-controlled data that can be used to study stroke
rehabilitation.
5.1 Summary of Key Findings
5.2.1 Chapter 2 Summary - Aim One
In Chapter 2, I aimed to identify an automated hippocampal segmentation
method that provides accurate hippocampal segmentations within the context of
stroke pathology. I compared three publicly available automated hippocampal
segmentation tools- Hippodeep (Thyreau et al., 2018), FreeSurfer version 6.0 gross
hippocampal segmentation (Fischl et al., 2012), and FreeSurfer version 6.0 ‘sum of
subfields’ (Iglesias et al., 2015). I evaluated the performance of each method in terms of
quality control and accuracy when compared to manual segmentations. I found that
although FreeSurfer version 6.0 ‘sum of subfields’ and Hippodeep performed comparably
in terms of segmentation accuracy, Hippodeep had the lowest method-wise quality control
fail rate, suggesting it may be the most robust to post-stroke anatomical distortions.
Therefore, Hippodeep can potentially help to maximize the number of participants
included in analyses whose data might not run successfully through FreeSurfer,
potentially boosting statistical power, and reducing the bias that can come from excluding
participants with more severe stroke pathology. Overall, our results suggest that
98
Hippodeep may be an optimal method for accurate and robust hippocampal
segmentation methods in diverse stroke populations.
5.2.2 Chapter 3 Summary - Aim Two
In Chapter 3, I aimed to investigate the association between lesion size and
post-stroke hippocampal volume in participants with chronic stroke from the
ENIGMA Stroke Recovery database. With the largest hippocampal volume analysis with
manually traced lesion masks to our knowledge, we found large lesion size was
significantly associated with smaller ipsilesional but not contralesional hippocampal
volume. This association was not evident in most cases when testing each cohort
individually and only became apparent when pooling the multi-site data. Our findings are
similar to Schaapsmeerders et al., (2015), and highlights the importance of large, diverse
samples necessary to detect associations between lesion size and hippocampal volume.
In a supplementary analysis, we also tested to see if hippocampal volumes differed in
participants with chronic stroke from age, sex, and research cohort matched healthy
controls and if lesion size was significantly associated with hippocampal volume. We
found that participants with stroke had significantly smaller hippocampal volumes than
healthy controls, but only within the ipsilesional hemisphere, reproducing several prior
studies that report similar findings (Schaapsmeerders et al, 2015; Haque et al., 2019;
Khlif et al., 2019; Brodtmann et al., 2020). Overall, in this study we provide supporting
evidence to existing literature that reduced hippocampal volume is likely a
consequence of stroke-related damage within the lesioned hemisphere. We also
99
show the feasibility of multi-site mega analyses using Hippodeep segmentations
that may be used for future research of the post-stroke hippocampus.
5.2.3 Chapter 4 Summary - Aim Three
In Chapter 4, I aimed to use ENIGMA Stroke Recovery data to investigate
possible associations between sensorimotor impairment and hippocampal volume
in patients with chronic stroke. We test associations between sensorimotor
impairment and hippocampal volume in patients with chronic stroke and did a follow up
exploratory analysis to see if sex moderates the association between sensorimotor
impairment and hippocampal volume. We found greater sensorimotor impairment was
significantly associated with smaller ipsilesional hippocampal volume, independent of
lesion size. We also found significant sex differences in the association between
sensorimotor impairment and bilateral hippocampal volume, where women showed
progressively smaller hippocampal volumes with increasing sensorimotor impairment
compared to men. Our findings demonstrate a novel association between chronic
post-stroke sensorimotor impairment and hippocampal volume that may be
modulated by sex. Overall, these findings provide unique insight into the role that
the hippocampus may play in post-stroke sensorimotor impairment.
5.3 Implications and Significance
5.3.1 Clinical Implications
Improving our understanding of secondary degeneration is of great clinical interest,
as the delayed onset of damage from secondary degeneration provides a feasible window
100
for clinical intervention (Wang et al., 2015). As discussed previously, identifying robust
biomarkers of stroke outcomes that generalize to diverse stroke populations is crucial for
understanding a patient’s ability to recovery and helping better target rehabilitation to
maximize outcomes (Boyd et al., 2017). In this work, we provide supporting evidence of
secondary degeneration of the hippocampus that generalizes in an international patient
population acquired at different research sites with varying inclusion/exclusion criteria at
each site. Specifically, stroke-related hippocampal damage occurs within the ipsilesional
hemisphere. Given that the hippocampus is densely connected to regions throughout the
brain and is involved in a number of functional circuits (Maller et al., 2019), it is important
for clinicians to take into considerations the range of symptoms that may occur as a result
of hippocampal damage, such as post-stroke dementia (Leys et al., 2005) and post-stroke
depression (Robinson and Jorge, 2016). Additionally, we introduce preliminary evidence
for how sensorimotor impairment may be tied to ipsilesional hippocampal damage that
should be taken into consideration during rehabilitation. Investigating the hippocampus
within the context of sensorimotor learning based treatment outcomes may provide a
clearer picture of circuits are involved in rehabilitation and sensorimotor impairment.
5.3.2 Research Implications
This work brings attention to the need to critically evaluate automated tools used
in stroke neuroimaging. The dataset of manual hippocampal segmentations that I
generated to test automated segmentation method accuracy in Chapter 2 were made
publicly available upon publication of our manuscript in late 2020 - Testing a convolutional
neural network-based hippocampal segmentation method in a stroke population
101
(Zavaliangos-Petropulu et al., 2020). These manual hippocampal segmentations have
already been used by an independent collaborator to publish an analysis that critically
evaluates stroke MRI normalization accuracy (Pappas et al., 2021).
In this work, we show the feasibility of a multi-site collaboration with a centralized
database. ENIGMA Stroke Recovery facilitates a venue for the leaders of the stroke
rehabilitation field to collaborate, identify gaps in the field, and share ideas on how to push
the field forward.
The importance of having centralized, available retrospective data became
glaringly important in March 2020, at the start of the global pandemic. While the world
came to a screeching halt, this work continued, uninterrupted. The ENIGMA Stroke
Recovery database will continue to serve its members in the case that the global
pandemic continues and in person data collection is suspended yet again.
Additionally, a centralized database with already collected and preprocessed data
stored on a computing cluster in the state of California is ecofriendly. As the world
grapples with the adverse reality of climate change, it is important to be mindful of our
carbon footprint as neuroimaging researchers. MRI data collection depends on liquid
helium to supercool the magnetic coil, which can only be acquired by drilling for fossil
fuels. Given the environmental cost of collecting MRI, the Organization for Human Brain
Sustainability and Environment Special Interest Group (OHBM-SE) encourages
neuroimaging researchers to consider the value of already collected MRI data more
greatly and focus on acquiring data with high reuse potential. Additionally, highly
computationally intensive MRI preprocessing consumes a significant amount of electricity
and other natural resources. OHBM-SE encourages centralized computing clusters to be
102
set up in areas that use clean energy. The state of California requires that at least 50%
of California’s electricity be produced through a renewable source (“State Programs —
California Renewable Energy”). Therefore, ENGIMA Stroke Recovery’s centralized
database of already preprocessed MRI located in Los Angeles, CA is aligned with
recommendations from OHBM-SE for eco-friendly neuroimaging practices.
5.4 Future Directions
There is much left to learn about the post-stroke hippocampus. Further research
on the clinical implications of hippocampal damage is important to creating a more
comprehensive understanding of the circuitry involved in post-stroke rehabilitation.
5.4.1 Hippocampal Subfields
In this paper, we only considered gross hippocampal volume. However, the
hippocampus is composed of structurally and functionally distinct subfields, each
differentially vulnerable to disease (Iglesias et al., 2015; Small et al., 2011). Structurally,
reduced neuron density has been observed in the cells of the CA1 but not CA2 subfield
of post-mortem stroke patients when compared to controls (Gemmell et al., 2012).
Functionally, while the posterior extents of the hippocampus along the long axis are
thought to be more involved with memory and cognitive processing, the anterior extents
have been implicated in sensorimotor integration (Small et al., 2011). Further research
investigating sensorimotor impairment and the hippocampus at a finer resolution, such as
at the level of hippocampal subfields (Iglesias et al., 2015) or vertex wise associations
(Hibar et al., 2017), may reveal more specific and robust relationships.
103
5.4.2 Hippocampal White Matter Pathways
Diffusion MRI (dMRI) can be used to further investigate secondary degeneration
of the hippocampus, specifically the microstructure of white matter tracts that interconnect
the hippocampus with the rest of the brain. Derived measures of dMRI such as fractional
anisotropy and mean diffusivity may reflect axonal injury and demyelination (Le Bihan
and Johansen-Berg, 2012) of temporal lobe pathways that pass through the hippocampus
such as the hippocampal-cingulum bundle, the fornix stria terminalis, and the uncinate
fasciculus. dMRI may be able to provide information about damage to hippocampal tracts
that may chronologically precede gross hippocampal volume changes. dMRI data for
ENIGMA Stroke Recovery is still in the early stages of data collection and preprocessing.
However, early on in my graduate studies, I investigated the impact of scanner and site
differences on derived dMRI measures and explored multi-site harmonization approaches
using the Alzheimer’s Disease Neuroimaging Data Initiative (ADNI) dataset (Zavaliangos-
Petropulu et al., 2019; Appendix B). We found that scanner manufacturer, acquisition
protocol, and site differences have a significant impact on derived dMRI measures and
must be statistically modeled when handling retrospective dMRI data. This research will
help inform future ENIGMA Stroke Recovery dMRI analyses.
5.4.3 Longitudinal Changes to Hippocampal Volume
The cross-sectional sample presented in this thesis cannot account for the extent
of longitudinal hippocampal atrophy that may have occurred as a result of stroke, pre-
existing dementia, normal aging, or heritability. Future longitudinal studies are important
104
for account for when hippocampal damage first appears, and how it may interfere with
sensorimotor rehabilitation.
5.4.4 Hippocampal Damage and Post-Stroke Cognitive Impairment, Depression, and
Anxiety
The hippocampus is a key structure of the limbic system and is involved in memory,
emotion and fear regulation (Rajmohan and Mohandas, 2007). Cognitive impairment,
depression and anxiety are all common chronic symptoms in stroke survivors and are all
associated with poorer stroke outcomes. Unfortunately, contemporary stroke
rehabilitation tends to focus on the physical (mostly sensorimotor) symptoms of stroke
and often overlook neuropsychological and neuropsychiatric elements (Quinn et al.,
2018). Sometimes independent, often comorbid, post-stroke dementia, depression, and
anxiety interfere with stroke rehabilitation efforts and prevent patients from returning to
their normal routines (Chun et al., 2018; Quinn et al., 2018).
Previous research suggests that post-stroke cognitive impairment may interfere
with a patient’s ability to learn or regain motor skills (Mullick et al., 2015, VanGilder et al.,
2020). In the ENIGMA Stroke Recovery database, most participants do not have severe
cognitive impairment, since it is common for stroke research studies to use cognitive
impairment as an exclusion criterion (Lingo VanGilder et al., 2020). However, cognitive
impairment is often screened in stroke research studies using blanket cognitive
assessments such as the Mini Mental Status Exam (MMSE). While the MMSE is quick
and easy to administer, it may not be sensitive to cognitive impairment common after
stroke (Nys et al., 2005), which may be better quantified by other more in-depth cognitive
105
assessments such as the Wechsler Adult Intelligence Scale (VanGilder et al., 2020).
Research suggests that this undetected cognitive impairment may be actually interfering
with sensorimotor learning based rehabilitation. In a clinical trial from 2006, FMA-UE was
significantly associated with cognitive status, and greater changes in FMA-UE post
treatment were related to fewer cognitive deficits (Cirstea et al., 2006). Further research
investigating the use of using hippocampal damage to predict post-stroke cognitive
impairment and its impact on rehabilitation efficacy is crucial for improving stroke recovery
outcomes.
There is conflicting evidence regarding associations between post-stroke
depression and hippocampal volumes that requires further investigation (Hong et al.,
2020; Shi et al., 2017). Unfortunately, despite the fact that post-stroke anxiety occurs in
an estimated 20-30% of stroke survivors, it remains understudied and undertreated (Chun
et al., 2018). It is possible that limbic system disruption caused by post-stroke
hippocampal damage aggravates symptoms of depression and anxiety that occur
because of the debilitating, stroke induced hemiparesis. Further research investigating
the role of hippocampal damage in post-stroke depression and anxiety are crucial for
identifying stroke survivors at risk and could lead to improved personalized rehabilitation
treatments.
106
References
Albert, K., Hiscox, J., Boyd, B., Dumas, J., Taylor, W., and Newhouse, P. (2017).
Estrogen enhances hippocampal gray-matter volume in young and older
postmenopausal women: a prospective dose-response study. Neurobiol. Aging 56,
1–6. doi:10.1016/j.neurobiolaging.2017.03.033.
Albouy, G., Sterpenich, V., Balteau, E., Vandewalle, G., Desseilles, M., Dang-Vu, T., et
al. (2008). Both the Hippocampus and Striatum Are Involved in Consolidation of
Motor Sequence Memory. Neuron 58, 261–272. doi:10.1016/j.neuron.2008.02.008.
Baumgartner, P., El Amki, M., Bracko, O., Luft, A. R., and Wegener, S. (2018).
Sensorimotor stroke alters hippocampo-thalamic network activity. Sci. Rep. 8.
doi:10.1038/s41598-018-34002-9.
Bean, L. A., Ianov, L., and Foster, T. C. (2014). Estrogen receptors, the hippocampus,
and memory. Neuroscientist 20, 534–545. doi:10.1177/1073858413519865.
Bernhardt, J., Hayward, K. S., Dancause, N., Lannin, N. A., Ward, N. S., Nudo, R. J., et
al. (2019). A stroke recovery trial development framework: Consensus-based core
recommendations from the Second Stroke Recovery and Rehabilitation
Roundtable: https://doi.org/10.1177/1747493019879657 14, 792–802.
doi:10.1177/1747493019879657.
Bernhardt, J., Hayward, K. S., Kwakkel, G., Ward, N. S., Wolf, S. L., Borschmann, K., et
al. (2017). Agreed definitions and a shared vision for new standards in stroke
recovery research: The Stroke Recovery and Rehabilitation Roundtable taskforce.
Int. J. Stroke 12, 444–450. doi:10.1177/1747493017711816.
Bigler, E. D., Skiles, M., Wade, B. S. C., Abildskov, T. J., Tustison, N. J., Scheibel, R.
107
S., et al. (2018). FreeSurfer 5.3 versus 6.0: are volumes comparable? A Chronic
Effects of Neurotrauma Consortium study. Brain Imaging Behav.
doi:10.1007/s11682-018-9994-x.
Boccardi, M., Bocchetta, M., Ganzola, R., Robitaille, N., Redolfi, A., Duchesne, S., et al.
(2015). Operationalizing protocol differences for EADC-ADNI manual hippocampal
segmentation. Alzheimer’s Dement. 11, 184–194. doi:10.1016/j.jalz.2013.03.001.
Borschmann, K., Hayward, K. S., Raffelt, A., Churilov, L., Kramer, S., and Bernhardt, J.
(2018). Rationale for intervention and dose is lacking in stroke recovery trials: A
systematic review. Stroke Res. Treat. 2018. doi:10.1155/2018/8087372.
Boyd, L. A., Hayward, K. S., Ward, N. S., Stinear, C. M., Rosso, C., Fisher, R. J., et al.
(2017). Biomarkers of stroke recovery: Consensus-based core recommendations
from the Stroke Recovery and Rehabilitation Roundtable. Int. J. Stroke 12, 480–
493. doi:10.1177/1747493017714176.
Braak, H., and Braak, E. (1991). Neuropathological stageing of Alzheimer-related
changes. Acta Neuropathol. 82, 239–59. Available at:
http://www.ncbi.nlm.nih.gov/pubmed/1759558 [Accessed July 21, 2018].
Braskie, M. N., and Thompson, P. M. (2014). A focus on structural brain imaging in the
Alzheimer’s Disease Neuroimaging Initiative. Biol. Psychiatry 75, 527–533.
doi:10.1016/j.biopsych.2013.11.020.
Brodtmann, A., Khlif, M. S., Egorova, N., Veldsman, M., Bird, L. J., and Werden, E.
(2020). Dynamic Regional Brain Atrophy Rates in the First Year after Ischemic
Stroke. Stroke 51, 183–192. doi:10.1161/STROKEAHA.120.030256.
Brodtmann, A., Werden, E., Pardoe, H., Li, Q., Jackson, G., Donnan, G., et al. (2014).
108
Charting cognitive and volumetric trajectories after stroke: Protocol for the
Cognition And Neocortical Volume After Stroke (CANVAS) study. Int. J. Stroke 9,
824–828. doi:10.1111/ijs.12301.
Burman, D. D. (2019). Hippocampal connectivity with sensorimotor cortex during
volitional finger movements: Laterality and relationship to motor learning. PLoS
One 14, e0222064. doi:10.1371/journal.pone.0222064.
Button, K. S., Ioannidis, J. P. A., Mokrysz, C., Nosek, B. A., Flint, J., Robinson, E. S. J.,
et al. (2013). Power failure: Why small sample size undermines the reliability of
neuroscience. Nat. Rev. Neurosci. 14, 365–376. doi:10.1038/nrn3475.
Chen, X., Dang, G., Dang, C., Liu, G., Xing, S., Chen, Y., et al. (2015). An ischemic
stroke model of nonhuman primates for remote lesion studies: A behavioral and
neuroimaging investigation. Restor. Neurol. Neurosci. 33, 131–142.
doi:10.3233/RNN-140440.
Christensen, H., and Bushnell, C. (2020). Stroke in Women. Contin. Lifelong Learn.
Neurol. 26, 363–385. doi:10.1212/CON.0000000000000836.
Chun, H., Whiteley, W., Dennis, M., Mead, G., and Carson, A. (2018). Anxiety After
Stroke: The Importance of Subtyping. Stroke 49, 556–564.
doi:10.1161/STROKEAHA.117.020078.
Chung, D. Y., Oka, F., and Ayata, C. (2016). Spreading depolarizations: A therapeutic
target against delayed cerebral ischemia after subarachnoid hemorrhage. J. Clin.
Neurophysiol. 33, 196. doi:10.1097/WNP.0000000000000275.
Cirstea, C. M., Ptito, A., and Levin, M. F. (2006). Feedback and cognition in arm motor
skill reacquisition after stroke. Stroke 37, 1237–1242.
109
doi:10.1161/01.STR.0000217417.89347.63.
Cordonnier, C., Sprigg, N., Sandset, E. C., Pavlovic, A., Sunnerhagen, K. S., Caso, V.,
et al. (2017). Stroke in women-from evidence to inequalities. Nat. Rev. Neurol. 13,
521–532. doi:10.1038/nrneurol.2017.95.
Dehlendorff, C., Andersen, K. K., and Olsen, T. S. (2015). Sex Disparities in Stroke:
Women Have More Severe Strokes but Better Survival Than Men. J. Am. Heart
Assoc. 4. doi:10.1161/JAHA.115.001967.
Desikan, R. S., Ségonne, F., Fischl, B., Quinn, B. T., Dickerson, B. C., Blacker, D., et al.
(2006). An automated labeling system for subdividing the human cerebral cortex on
MRI scans into gyral based regions of interest. Neuroimage 31, 968–980.
doi:10.1016/j.neuroimage.2006.01.021.
Dichgans, M. (2019). Dementia risk after transient ischaemic attack and stroke. Lancet.
Neurol. 18, 223–225. doi:10.1016/S1474-4422(18)30497-6.
Edwards, L. L., King, E. M., Buetefisch, C. M., and Borich, M. R. (2019). Putting the
“sensory” into sensorimotor control: The role of sensorimotor integration in goal-
directed hand movements after stroke. Front. Integr. Neurosci. 13, 16.
doi:10.3389/fnint.2019.00016.
Egorova, N., Gottlieb, E., Khlif, M. S., Spratt, N. J., and Brodtmann, A. (2019). Choroid
plexus volume after stroke. Int. J. Stroke 14, 923–930.
doi:10.1177/1747493019851277.
Endres, M., Engelhardt, B., Koistinaho, J., Lindvall, O., Meairs, S., Mohr, J., et al.
(2008). Improving outcome after stroke: overcoming the translational roadblock.
Cerebrovasc. Dis. 25, 268–278. doi:10.1159/000118039.
110
Feigin, V. L., Norrving, B., Mensah, G. A., Fisher, M., Iadecola, C., and Sacco, R.
(2017). Stroke Compendium Global Burden of Stroke Effects of Neurologic Injury
on Cardiovascular Function Vascular Cognitive Impairment.
doi:10.1161/CIRCRESAHA.116.308413.
Ferretti, M. T., Iulita, M. F., Cavedo, E., Chiesa, P. A., Dimech, A. S., Chadha, A. S., et
al. (2018). Sex differences in Alzheimer disease — The gateway to precision
medicine. Nat. Rev. Neurol. 14, 457–469. doi:10.1038/s41582-018-0032-9.
Fischl, B. (2012). FreeSurfer. Neuroimage 62, 774–781.
doi:10.1016/j.neuroimage.2012.01.021.
Fischl, B., Salat, D. H., Busa, E., Albert, M., Dieterich, M., Haselgrove, C., et al.
(2002a). Neurotechnique Whole Brain Segmentation: Automated Labeling of
Neuroanatomical Structures in the Human Brain. Neuron 33, 341–355. Available at:
https://surfer.nmr.mgh.harvard.edu/ftp/articles/fischl02-labeling.pdf [Accessed
February 6, 2018].
Fischl, B., Salat, D. H., Busa, E., Albert, M., Dieterich, M., Haselgrove, C., et al.
(2002b). Whole Brain Segmentation: Neurotechnique Automated Labeling of
NeuroanatomicalStructures in the Human Brain. Neuron 33, 341–355.
doi:10.1016/S0896-6273(02)00569-X.
Fleisher, A., Grundman, M., Jack, C. R., Petersen, R. C., Taylor, C., Kim, H. T., et al.
(2005). Sex, apolipoprotein E ε4 status, and hippocampal volume in mild cognitive
impairment. Arch. Neurol. 62, 953–957. doi:10.1001/archneur.62.6.953.
Frisoni, G. B., Jack, C. R., Bocchetta, M., Bauer, C., Frederiksen, K. S., Liu, Y., et al.
(2015). The EADC-ADNI harmonized protocol for manual hippocampal
111
segmentation on magnetic resonance: Evidence of validity. Alzheimer’s Dement.
11, 111–125. doi:10.1016/j.jalz.2014.05.1756.
Fugl-Meyer, A. R., Jääskö, L., Leyman, I., Olsson, S., and Steglind, S. (1975). The post-
stroke hemiplegic patient. 1. a method for evaluation of physical performance.
Scand. J. Rehabil. Med. 7, 13–31. Available at:
http://www.ncbi.nlm.nih.gov/pubmed/1135616 [Accessed August 20, 2019].
Gemmell, E., Bosomworth, H., Allan, L., Hall, R., Khundakar, A., Oakley, A. E., et al.
(2012). Hippocampal Neuronal Atrophy and Cognitive Function in Delayed
Poststroke and Aging-Related Dementias. Stroke 43, 808–814.
doi:10.1161/STROKEAHA.111.636498.
Gemmell, E., Tam, E., Allan, L., Hall, R., Khundakar, A., Oakley, A. E., et al. (2014).
Neuron Volumes in Hippocampal Subfields in Delayed Poststroke and Aging-
Related Dementias. J. Neuropathol. Exp. Neurol. 73, 305–311.
doi:10.1097/NEN.0000000000000054.
Gittler, M., and Davis, A. M. (2018). Guidelines for Adult Stroke Rehabilitation and
Recovery. JAMA 319, 820. doi:10.1001/jama.2017.22036.
Giuliano, A., Donatelli, G., Cosottini, M., Tosetti, M., Retico, A., and Fantacci, M. E.
(2017). Hippocampal subfields at ultra high field MRI: An overview of segmentation
and measurement methods. Hippocampus 27, 481–494. doi:10.1002/hipo.22717.
Gorgolewski, K. J., Auer, T., Calhoun, V. D., Craddock, R. C., Das, S., Duff, E. P., et al.
(2016). The brain imaging data structure, a format for organizing and describing
outputs of neuroimaging experiments. Sci. Data 2016 31 3, 1–9.
doi:10.1038/sdata.2016.44.
112
Greco, L., Luta, G., Krzywinski, M., and Altman, N. (2019). Analyzing outliers: robust
methods to the rescue. Nat. Methods 16, 275–276. doi:10.1038/s41592-019-0369-
z.
Halliday, G. (2017). Pathology and hippocampal atrophy in Alzheimer’s disease. Lancet
Neurol. 16, 862–864. doi:10.1016/S1474-4422(17)30343-5.
Han, X., Jovicich, J., Salat, D., van der Kouwe, A., Quinn, B., Czanner, S., et al. (2006).
Reliability of MRI-derived measurements of human cerebral cortical thickness: The
effects of field strength, scanner upgrade and manufacturer. Neuroimage 32, 180–
194. doi:10.1016/j.neuroimage.2006.02.051.
Haque, M. E., Gabr, R. E., Hasan, K. M., George, S., Arevalo, O. D., Zha, A., et al.
(2019). Ongoing Secondary Degeneration of the Limbic System in Patients With
Ischemic Stroke: A Longitudinal MRI Study. Front. Neurol. 10, 154.
doi:10.3389/fneur.2019.00154.
Hawe, R. L., Cluff, T., Dowlatshahi, D., Hill, M. D., and Dukelow, S. P. (2020a).
Assessment of Sex Differences in Recovery of Motor and Sensory Impairments
Poststroke. Neurorehabil. Neural Repair 34, 746–757.
doi:10.1177/1545968320935811.
Hawe, R. L., Cluff, T., Dowlatshahi, D., Hill, M. D., and Dukelow, S. P. (2020b).
Assessment of Sex Differences in Recovery of Motor and Sensory Impairments
Poststroke. Neurorehabil. Neural Repair 34, 746–757.
doi:10.1177/1545968320935811.
Hibar, D. P., Adams, H. H. H., Jahanshad, N., Chauhan, G., Stein, J. L., Hofer, E., et al.
(2017). Novel genetic loci associated with hippocampal volume. Nat. Commun. 8,
113
13624. doi:10.1038/ncomms13624.
Hibar, D. P., Westlye, L. T., Van Erp, T. G. M., Rasmussen, J., Leonardo, C. D.,
Faskowitz, J., et al. (2016). Subcortical volumetric abnormalities in bipolar disorder.
Mol. Psychiatry 21, 1710–1716. doi:10.1038/mp.2015.227.
Hijikata, N., Kawakami, M., Ishii, R., Tsuzuki, K., Nakamura, T., Okuyama, K., et al.
(2020). Item Difficulty of Fugl-Meyer Assessment for Upper Extremity in Persons
With Chronic Stroke With Moderate-to-Severe Upper Limb Impairment. Front.
Neurol. 11, 577855. doi:10.3389/fneur.2020.577855.
Hong, W., Zhao, Z., Wang, D., Li, M., Tang, C., Li, Z., et al. (2020). Altered gray matter
volumes in post-stroke depressive patients after subcortical stroke. Neuroimage
(Amst). 26. doi:10.1016/J.NICL.2020.102224.
Iglesias, J. E., Augustinack, J. C., Nguyen, K., Player, C. M., Player, A., Wright, M., et
al. (2015). A computational atlas of the hippocampal formation using ex vivo, ultra-
high resolution MRI: Application to adaptive segmentation of in vivo MRI.
Neuroimage 115, 117–37. doi:10.1016/j.neuroimage.2015.04.042.
Irimia, A., Chambers, M. C., Alger, J. R., Filippou, M., Prastawa, M. W., Wang, B., et al.
(2011). Comparison of Acute and Chronic Traumatic Brain Injury Using Semi-
Automatic Multimodal Segmentation of MR Volumes. J. Neurotrauma 28, 2287–
2306. doi:10.1089/neu.2011.1920.
Ito, K. L., Kim, H., and Liew, S.-L. (2019). A comparison of automated lesion
segmentation approaches for chronic stroke T1-weighted MRI data. Hum. Brain
Mapp. 40, 4669–4685. doi:10.1002/HBM.24729.
Jacobacci, F., Armony, J. L., Yeffal, A., Lerner, G., Amaro, E., Jovicich, J., et al. (2020).
114
Rapid hippocampal plasticity supports motor sequence learning. Proc. Natl. Acad.
Sci. U. S. A. 117, 23898–23903. doi:10.1073/pnas.2009576117.
Johnson, C. O., Nguyen, M., Roth, G. A., Nichols, E., Alam, T., Abate, D., et al. (2019).
Global, regional, and national burden of stroke, 1990–2016: a systematic analysis
for the Global Burden of Disease Study 2016. Lancet Neurol. 18, 439–458.
doi:10.1016/S1474-4422(19)30034-1.
Khlif, M. S., Egorova, N., Werden, E., Redolfi, A., Boccardi, M., DeCarli, C. S., et al.
(2018). A comparison of automated segmentation and manual tracing in estimating
hippocampal volume in ischemic stroke and healthy control participants.
NeuroImage Clin., 101581. doi:10.1016/j.nicl.2018.10.019.
Khlif, M. S., Egorova, N., Werden, E., Redolfi, A., Boccardi, M., DeCarli, C. S., et al.
(2019a). A comparison of automated segmentation and manual tracing in
estimating hippocampal volume in ischemic stroke and healthy control participants.
NeuroImage. Clin. 21, 101581. doi:10.1016/j.nicl.2018.10.019.
Khlif, M. S., Werden, E., Egorova, N., Boccardi, M., Redolfi, A., Bird, L., et al. (2019b).
Assessment of longitudinal hippocampal atrophy in the first year after ischemic
stroke using automatic segmentation techniques. NeuroImage Clin. 24, 102008.
doi:10.1016/j.nicl.2019.102008.
Khoshnam, S. E., Winlow, W., Farzaneh, M., Farbood, Y., Hadi, &, and Moghaddam, F.
(2017). Pathogenic mechanisms following ischemic stroke. doi:10.1007/s10072-
017-2938-1.
Kim, B., and Winstein, C. (2017). Can Neurological Biomarkers of Brain Impairment Be
Used to Predict Poststroke Motor Recovery? A Systematic Review. Neurorehabil.
115
Neural Repair 31, 3–24. doi:10.1177/1545968316662708.
Koo, T. K., and Li, M. Y. (2016). A Guideline of Selecting and Reporting Intraclass
Correlation Coefficients for Reliability Research. J. Chiropr. Med. 15, 155–163.
doi:10.1016/j.jcm.2016.02.012.
Krakauer, J. W., and Carmichael, T. S. (2017). Broken Movement: The Neurobiology of
Motor Recovery After Stroke - John W. Krakauer, S. Thomas Carmichael - Google
Books. Available at: https://books.google.com/books?hl=en&lr=&id=Mpg-
DwAAQBAJ&oi=fnd&pg=PR7&dq=krakauer+broken+movement&ots=Wm2OKb6lb
n&sig=vVz7Q3fN6CM649xwqbujimKeUqk#v=onepage&q=krakauer broken
movement&f=false [Accessed July 16, 2019].
Le Bihan, D., and Johansen-Berg, H. (2012). Diffusion MRI at 25: Exploring brain tissue
structure and function. Neuroimage 61, 324–341.
doi:10.1016/j.neuroimage.2011.11.006.
Leys, D., Hénon, H., Mackowiak-Cordoliani, M.-A., and Pasquier, F. (2005). Poststroke
dementia. Lancet Neurol. 4, 752–759. doi:10.1016/S1474-4422(05)70221-0.
Liao, C. C., Chen, Y. F., and Xiao, F. (2018). Brain midline shift measurement and its
automation: A review of techniques and algorithms. Int. J. Biomed. Imaging 2018.
doi:10.1155/2018/4303161.
Liew, S.-L., Zavaliangos-Petropulu, A., Jahanshad, N., Lang, C. E., Hayward, K. S.,
Lohse, K., et al. (2020). The ENIGMA Stroke Recovery Working Group: Big data
neuroimaging to study brain-behavior relationships after stroke. Hum. Brain Mapp.
doi:10.31234/OSF.IO/WU7MH.
Liew, S.-L., Zavaliangos-Petropulu, A., Schweighofer, N., Jahanshad, N., Lang, C. E.,
116
Lohse, K. R., et al. (2021). Smaller spared subcortical nuclei are associated with
worse post-stroke sensorimotor outcomes in 28 cohorts worldwide. bioRxiv,
2020.11.04.366856. doi:10.1101/2020.11.04.366856.
Liew, S., Anglin, J. M., Banks, N. W., Sondag, M., Ito, K. L., Saldana, D., et al. (2017).
The Anatomical Tracings of Lesions After Stroke (ATLAS) Dataset - Release 1.1.
bioRxiv, 1–15.
Liew, S. L., Anglin, J. M., Banks, N. W., Sondag, M., Ito, K. L., Kim, H., et al. (2018). A
large, open source dataset of stroke anatomical brain images and manual lesion
segmentations. Sci. Data 5. doi:10.1038/sdata.2018.11.
Lingo VanGilder, J., Hooyman, A., Peterson, D. S., and Schaefer, S. Y. (2020). Post-
Stroke Cognitive Impairments and Responsiveness to Motor Rehabilitation: A
Review. Curr. Phys. Med. Rehabil. Reports 8, 461–468. doi:10.1007/s40141-020-
00283-3.
Lisabeth, L. D., Reeves, M. J., Baek, J., Skolarus, L. E., Brown, D. L., Zahuranec, D. B.,
et al. (2015). Factors influencing sex differences in poststroke functional outcome.
Stroke 46, 860–863. doi:10.1161/STROKEAHA.114.007985.
Maller, J. J., Welton, T., Middione, M., Callaghan, F. M., Rosenfeld, J. V., and Grieve, S.
M. (2019). Revealing the Hippocampal Connectome through Super-Resolution
1150-Direction Diffusion MRI. Sci. Rep. 9, 1–13. doi:10.1038/s41598-018-37905-9.
Mok, V. C. T., Lam, B. Y. K., Wong, A., Ko, H., Markus, H. S., and Wong, L. K. S.
(2017). Early-onset and delayed-onset poststroke dementia-revisiting the
mechanisms. Nat. Rev. Neurol. 13, 148–159. doi:10.1038/nrneurol.2017.16.
Mozaffarian, D., Benjamin, E. J., Go, A. S., Arnett, D. K., Blaha, M. J., Cushman, M., et
117
al. (2016). Heart disease and stroke statistics-2016 update a report from the
American Heart Association. doi:10.1161/CIR.0000000000000350.
Mukamel, R., Ekstrom, A. D., Kaplan, J., Iacoboni, M., and Fried, I. (2010). Single-
Neuron Responses in Humans during Execution and Observation of Actions. Curr.
Biol. 20, 750–756. doi:10.1016/j.cub.2010.02.045.
Müller-Ehrenberg, L., Riphagen, J. M., Verhey, F. R. J., Sack, A. T., and Jacobs, H. I. L.
(2018). Alzheimer’s Disease Biomarkers Have Distinct Associations with Specific
Hippocampal Subfield Volumes. J. Alzheimer’s Dis. 66, 811–823. doi:10.3233/JAD-
180676.
Munafò, M. R., Nosek, B. A., Bishop, D. V. M., Button, K. S., Chambers, C. D., Percie
du Sert, N., et al. (2017). A manifesto for reproducible science. Nat. Hum. Behav. 1,
0021. doi:10.1038/s41562-016-0021.
Nebel, R. A., Aggarwal, N. T., Barnes, L. L., Gallagher, A., Goldstein, J. M., Kantarci, K.,
et al. (2018). Understanding the impact of sex and gender in Alzheimer’s disease:
A call to action. Alzheimer’s Dement. 14, 1171–1183.
doi:10.1016/j.jalz.2018.04.008.
Nelson, J. S. (2003). Principles and practice of neuropathology. Oxford University
Press.
Nieuwenhuis, R., Te Grotenhuis, M., and Pelzer, B. (2012). influence.ME: Tools for
Detecting Influential Data in Mixed Effects Models. 4.
Nikonenko, A. G., Radenovic, L., Andjus, P. R., and Skibo, G. G. (2009). Structural
Features of Ischemic Damage in the Hippocampus. Anat. Rec. Adv. Integr. Anat.
Evol. Biol. 292, 1914–1921. doi:10.1002/ar.20969.
118
Nobis, L., Manohar, S. G., Smith, S. M., Alfaro-Almagro, F., Jenkinson, M., Mackay, C.
E., et al. (2019). Hippocampal volume across age: Nomograms derived from over
19,700 people in UK Biobank. NeuroImage Clin. 23.
doi:10.1016/j.nicl.2019.101904.
Nogovitsyn, N., Souza, R., Muller, M., Srajer, A., Hassel, S., Arnott, S. R., et al. (2019).
Testing a deep convolutional neural network for automated hippocampus
segmentation in a longitudinal sample of healthy participants. Neuroimage 197,
589–597. doi:10.1016/j.neuroimage.2019.05.017.
Nys, G. M. S., van Zandvoort, M. J. E., de Kort, P. L. M., Jansen, B. P. W., Kappelle, L.
J., and de Haan, E. H. F. (2005). Restrictions of the Mini-Mental State Examination
in acute stroke. Arch. Clin. Neuropsychol. 20, 623–629.
doi:10.1016/J.ACN.2005.04.001.
Pappa, T., Vemmos, K., Mantzou, E., Savvari, P., Stamatelopoulos, K., and Alevizaki,
M. (2012). Estradiol levels predict short-term adverse health outcomes in
postmenopausal acute stroke women. Eur. J. Neurol. 19, 1300–1304.
doi:10.1111/j.1468-1331.2012.03714.x.
Pappas, I., Hector, H., Haws, K., Curran, B., Kayser, A. S., and D’Esposito, M. (2021).
Improved normalization of lesioned brains via cohort-specific templates. Hum. Brain
Mapp. doi:10.1002/HBM.25474.
Pearson, K. (1895). [ 343 ] X. Contributions to the Mathematical Theory of Evolution.-JI.
Skew Variation in Homogeneous Matterial.
Quinn, T., Elliott, E., and Langhorne, P. (2018). Cognitive and Mood Assessment Tools
for Use in Stroke. Stroke 49, 483–490. doi:10.1161/STROKEAHA.117.016994.
119
Rajmohan, V., and Mohandas, E. (2007). The limbic system. Indian J. Psychiatry 49,
132. doi:10.4103/0019-5545.33264.
Reeves, M. J., Bushnell, C. D., Howard, G., Gargano, J. W., Duncan, P. W., Lynch, G.,
et al. (2008). Sex differences in stroke: epidemiology, clinical presentation, medical
care, and outcomes. Lancet Neurol. 7, 915–926. doi:10.1016/S1474-
4422(08)70193-5.
Ritchie, S. J., Cox, S. R., Shen, X., Lombardo, M. V., Reus, L. M., Alloza, C., et al.
(2018). Sex differences in the adult human brain: Evidence from 5216 UK biobank
participants. Cereb. Cortex 28, 2959–2975. doi:10.1093/cercor/bhy109.
Robinson, R. G., and Jorge, R. E. (2016). Post-stroke depression: A review. Am. J.
Psychiatry 173, 221–231. doi:10.1176/appi.ajp.2015.15030363.
Sachdev, P. S., Lo, J. W., Crawford, J. D., Mellon, L., Hickey, A., Williams, D., et al.
(2017). STROKOG (stroke and cognition consortium): An international consortium
to examine the epidemiology, diagnosis, and treatment of neurocognitive disorders
in relation to cerebrovascular disease. Alzheimer’s Dement. Diagnosis, Assess.
Dis. Monit. 7, 11–23. doi:10.1016/j.dadm.2016.10.006.
Sankar, T., Park, M. T. M., Jawa, T., Patel, R., Bhagwat, N., Voineskos, A. N., et al.
(2017). Your algorithm might think the hippocampus grows in Alzheimer’s disease:
Caveats of longitudinal automated hippocampal volumetry. Hum. Brain Mapp. 38,
2875–2896. doi:10.1002/hbm.23559.
Schaapsmeerders, P., van Uden, I. W. M., Tuladhar, A. M., Maaijwee, N. A. M., van
Dijk, E. J., Rutten ‐Jacobs, L. C. A., et al. (2015). Ipsilateral hippocampal atrophy is
associated with long ‐term memory dysfunction after ischemic stroke in young
120
adults. Hum. Brain Mapp. 36, 2432–2442. doi:10.1002/hbm.22782.
Schmaal, L., Veltman, D. J., Van Erp, T. G. M., Smann, P. G., Frodl, T., Jahanshad, N.,
et al. (2016). Subcortical brain alterations in major depressive disorder: Findings
from the ENIGMA Major Depressive Disorder working group. Mol. Psychiatry 21,
806–812. doi:10.1038/mp.2015.69.
See, J., Dodakian, L., Chou, C., Chan, V., McKenzie, A., Reinkensmeyer, D., et al.
(2013). A standardized approach to the Fugl-Meyer assessment and its
implications for clinical trials. Neurorehabil. Neural Repair 27, 732–741.
doi:10.1177/1545968313491000.
Shi, Y., Zeng, Y., Wu, L., Liu, W., Liu, Z., Zhang, S., et al. (2017). A Study of the Brain
Abnormalities of Post-Stroke Depression in Frontal Lobe Lesion. Sci. Rep. 7,
13203. doi:10.1038/s41598-017-13681-w.
Shrout, P. E., and Fleiss, J. L. (1979). Intraclass correlations: Uses in assessing rater
reliability. Psychol. Bull. 86, 420–428. doi:10.1037/0033-2909.86.2.420.
Small, S. A., Schobel, S. A., Buxton, R. B., Witter, M. P., and Barnes, C. A. (2011). A
pathophysiological framework of hippocampal dysfunction in ageing and disease.
Nat. Rev. Neurosci. 12, 585–601. doi:10.1038/nrn3085.
State Programs — California Renewable Energy Available at:
https://www.californiarenewableenergy.org/state-programs [Accessed July 17,
2021].
Stinear, C. M. (2017). Prediction of motor recovery after stroke: advances in
biomarkers. Lancet Neurol. 16, 826–836. doi:10.1016/S1474-4422(17)30283-1.
Stinear, C. M., Lang, C. E., Zeiler, S., and Byblow, W. D. (2020). Advances and
121
challenges in stroke rehabilitation. Lancet Neurol. 19, 348–360. doi:10.1016/S1474-
4422(19)30415-6.
Stinear, C. M., Stinear, C., and Stinear, C. M. (2017). Prediction of motor recovery after
stroke: advances in biomarkers. Lancet Neurol. 16, 826–836. doi:10.1016/S1474-
4422(17)30283-1.
Sueiras, M., Thonon, V., Santamarina, E., Sánchez-Guerrero, Á., Poca, M. A.,
Quintana, M., et al. (2021). Cortical Spreading Depression Phenomena Are
Frequent in Ischemic and Traumatic Penumbra: A Prospective Study in Patients
With Traumatic Brain Injury and Large Hemispheric Ischemic Stroke. J. Clin.
Neurophysiol. 38, 47–55. doi:10.1097/WNP.0000000000000648.
Suzuki, W. A. (2007). Integrating associative learning signals across the brain.
Hippocampus 17, 842–850. doi:10.1002/hipo.20321.
Szabo, K., Förster, A., Jäger, T., Kern, R., Griebe, M., Hennerici, M. G., et al. (2009).
Hippocampal lesion patterns in acute posterior cerebral artery stroke: clinical and
MRI findings. Stroke 40, 2042–5. doi:10.1161/STROKEAHA.108.536144.
Tang, X., Wang, C., Xia, L., Zhu, W., Zhao, L., and Zhu, W. (2012). Volumetric MRI and
1H MRS study of hippocampus in unilateral MCAO patients: Relationship between
hippocampal secondary damage and cognitive disorder following stroke. Eur. J.
Radiol. 81, 2788–2793. doi:10.1016/j.ejrad.2011.08.010.
Thompson, P. M., Ching, C. R. K., Dennis, E. L., Salminen, L. E., Turner, J. A., van Erp,
T. G. M., et al. (2020). “Big Data Initiatives in Psychiatry: Global Neuroimaging
Studies,” in Neuroimaging in Schizophrenia (Cham: Springer International
Publishing), 411–426. doi:10.1007/978-3-030-35206-6_21.
122
Thyreau, B., Sato, K., Fukuda, H., and Taki, Y. (2018). Segmentation of the
hippocampus by transferring algorithmic knowledge for large cohort processing.
Med. Image Anal. 43, 214–228. doi:10.1016/j.media.2017.11.004.
Urbach, A., Baum, E., Braun, F., and Witte, O. W. (2017). Cortical spreading
depolarization increases adult neurogenesis, and alters behavior and
hippocampus-dependent memory in mice. J. Cereb. Blood Flow Metab. 37, 1776.
doi:10.1177/0271678X16643736.
Van Erp, T. G. M., Hibar, D. P., Rasmussen, J. M., Glahn, D. C., Pearlson, G. D.,
Andreassen, O. A., et al. (2016). Subcortical brain volume abnormalities in 2028
individuals with schizophrenia and 2540 healthy controls via the ENIGMA
consortium. Mol. Psychiatry 21, 547–553. doi:10.1038/mp.2015.63.
van Norden, A., Fick, W., de Laat, K., van Uden, I., van Oudheusden, L., Tendolkar, I.,
et al. (2008). Subjective cognitive failures and hippocampal volume in elderly with
white matter lesions. Neurology 71, 1152–1159.
doi:10.1212/01.WNL.0000327564.44819.49.
Virani, S. S., Alonso, A., Benjamin, E. J., Bittencourt, M. S., Callaway, C. W., Carson, A.
P., et al. (2020). Heart disease and stroke statistics—2020 update: A report from
the American Heart Association. Circulation 141, E139–E596.
doi:10.1161/CIR.0000000000000757.
Wang, W., Redecker, C., Bidmon, H. J., and Witte, O. W. (2004). Delayed neuronal
death and damage of GDNF family receptors in CA1 following focal cerebral
ischemia. Brain Res. 1023, 92–101. doi:10.1016/j.brainres.2004.07.034.
Werden, E., Cumming, T., Li, Q., Bird, L., Veldsman, M., Pardoe, H. R., et al. (2017).
123
Structural MRI markers of brain aging early after ischemic stroke. Neurology 89,
116–124. doi:10.1212/WNL.0000000000004086.
Wierenga, L. M., Doucet, G. E., Dima, D., Agartz, I., Aghajani, M., Akudjedu, T. N., et al.
(2020). Greater male than female variability in regional brain structure across the
lifespan. Hum. Brain Mapp. 7, 96. doi:10.1002/HBM.25204.
Xie, M., Yi, C., Luo, X., Xu, S., Yu, Z., Tang, Y., et al. (2011). Glial gap junctional
communication involvement in hippocampal damage after middle cerebral artery
occlusion. Ann. Neurol. 70, 121–132. doi:10.1002/ana.22386.
Yang, X., Han, X., Park, E., Aylward, S., Kwitt, R., and Niethammer, M. (2016).
Registration of pathological images. in Lecture Notes in Computer Science
(including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in
Bioinformatics) (Springer Verlag), 97–107. doi:10.1007/978-3-319-46630-9_10.
Yousufuddin, M., and Young, N. (2019). Aging and ischemic stroke. Aging (Albany. NY).
11, 2542–2544. doi:10.18632/aging.101931.
Yushkevich, P. A., Piven, J., Hazlett, H. C., Smith, R. G., Ho, S., Gee, J. C., et al.
(2006). User-guided 3D active contour segmentation of anatomical structures:
Significantly improved efficiency and reliability. Neuroimage 31, 1116–1128.
doi:10.1016/j.neuroimage.2006.01.015.
Zavaliangos-Petropulu, A., Nir, T. M., Thomopoulos, S. I., Reid, R. I., Bernstein, M. A.,
Borowski, B., et al. (2019). Diffusion MRI Indices and Their Relation to Cognitive
Impairment in Brain Aging: The Updated Multi-protocol Approach in ADNI3. Front.
Neuroinform. 13, 2. doi:10.3389/fninf.2019.00002.
Zavaliangos ‐Petropulu, A., Tubi, M. A., Haddad, E., Zhu, A., Braskie, M. N., Jahanshad,
124
N., et al. (2020). Testing a convolutional neural network ‐based hippocampal
segmentation method in a stroke population. Hum. Brain Mapp., hbm.25210.
doi:10.1002/hbm.25210.
Zhang, J., Zhang, Y., Xing, S., Liang, Z., and Zeng, J. (2012). Secondary
neurodegeneration in remote regions after focal cerebral infarction: a new target for
stroke management? Stroke 43, 1700–1705.
doi:10.1161/STROKEAHA.111.632448.
Zhao, W., Wang, X., Yin, C., He, M., Li, S., and Han, Y. (2019). Trajectories of the
Hippocampal Subfields Atrophy in the Alzheimer’s Disease: A Structural Imaging
Study. Front. Neuroinform. 13, 13. doi:10.3389/fninf.2019.00013.
Zheng, Z., Li, R., Xiao, F., He, R., Zhang, S., and Li, J. (2017). Sex matters:
Hippocampal volume predicts individual differences in associative memory in
cognitively normal older women but not men. Front. Hum. Neurosci. 11.
doi:10.3389/fnhum.2017.00093.
125
Appendix A
List of Relevant Publications
Published:
1. Zavaliangos-Petropulu A, Tubi MA, Haddad E, Zhu A, Braskie MN, Jahanshad
N, Thompson PM, Liew S-L (2020) Testing a convolutional neural network-based
hippocampal segmentation method in a stroke population. Human Brain
Mapping. https://doi.org/10.1002/hbm.25210
2. Liew S-L, Zavaliangos-Petropulu A, Jahanshad N, Lang C E, Hayward K S,
Lohse K, Juliano JM, Assogna F, Baugh LA, Bhattacharya AK, Borich MR, Boyd
LA, Brodtmann A, Buetefisch CM, Byblow WD, Cassidy JM, Conforto AB,
Craddock RC, Dimyan MA, Dula AN, Ermer E, Etherton MR, Fercho KA, Gregory
CM, Hadidchi S, Holguin JA, Hwang DH, Jung S, Kautz SA, Khlif MS, Khoshab N,
Kim B, Kim H, Kuceyeski A, Lotze M, MacIntosh BJ, Margetis JL, Mohamed FB,
Piras F, Ramos-Murguialday A, Richard G, Roberts P, Robertson AD, Rondina
JM, Rost NS, Sanossian N, Schweighofer N, Shiroishi MS, Soekadar SR, Spalletta
G, Stinear CM, Suri A, Tang WKW, Thielman GT, Vecchio D, Villringer A, Ward
NS, Werden E, Westlye LT, Winstein C, Wittenberg GF, Wong KA, Yu C, Cramer
SC, & Thompson PM (2020). The ENIGMA Stroke Recovery Working Group: Big
data neuroimaging to study brain-behavior relationships after stroke. Human Brain
Mapping. https://doi.org/10.1002/hbm.25015
3. Zavaliangos-Petropulu A*, Nir TM*, Thomopoulos SI, Jahanshad N, Reid RI,
Bernstein MA, Borowski B, Jack CR, Weiner MW, Thompson PM, The Alzheimer's
Disease Neuroimaging Initiative (ADNI) (2019). Diffusion MRI indices of cognitive
126
impairment in brain aging: The updated multi-protocol approach in ADNI3.
Frontiers in Neuroinformatics. https://dx.doi.org/10.3389%2Ffninf.2019.00002
In Preparation:
1. Liew S-L, Zavaliangos-Petropulu A, Schweighofer N, Jahanshad N, Lang CE,
Lohse KR, Banaj N, Barisano G, Baugh LA, Bhattacharya AK, Bigjahan B, Borich
MR, Boyd LA, Brodtmann A, Beutefisch CM, Byblow WD, Cassidy JM, Ciullo V,
Conforto AB, Craddock RC, Dula AN, Egorova N, Feng W, Fercho KA, Gregory
CM, Hanlon CA, Hayward KS, Holguin JA, Hordacre B, Hwang DJ, Kautz SA, Khlif
MS, Kim B, Kim H, Kuceyeski A, Liu J, Lin D, Lotze M, MacIntosh BJ, Margetis JL,
Mohamed FB, Nordvik JE, Petoe MA, Piras F, Raju S, Ramos-Murguialday A,
Revill KP, Roberts P, Robertson AD, Schambra HM, Seo NJ, Shiroshi MS,
Soekadar SR, Spalletta G, Stinear CM, Suri A, Tang WK, Theilman GT, Thijs VN,
Vacchio D, Ward NS, Westlye LT, Winstein CJ, Wittenberg GF, Wong KA, Yu C,
Wolf SL, Cramer SC, Thompson PM (2021) Atrophy of spared subcortical nuclei
relates to worse post-stroke sensorimotor outcomes across 28 cohorts worldwide.
bioRxiv. https://doi.org/10.1101/2020.11.04.366856
2. Zavaliangos-Petropulu A, …. Liew S-L (2021) Associations between
hippocampal volume and sensorimotor impairment in chronic stroke survivors: An
ENIGMA Stroke Recovery Analysis. In preparation
127
Appendix B
Diffusion MRI Indices and their Relation to Cognitive
Impairment in Brain Aging: The updated multi-protocol
approach in ADNI3
Artemis Zavaliangos-Petropulu
1*
, Talia M. Nir
1*
, Sophia I. Thomopoulos
1
, Robert I. Reid
2
,
Matt A. Bernstein
3
, Bret Borowski
3
, Clifford R. Jack, Jr.
3
, Michael W. Weiner
4
, Neda
Jahanshad
1
, Paul M. Thompson
1
, for the Alzheimer’s Disease Neuroimaging Initiative
(ADNI)
+
1
Imaging Genetics Center, Mark & Mary Stevens Neuroimaging & Informatics Institute,
Keck School of Medicine, University of Southern California, Marina del Rey, CA, USA
2
Department of Information Technology, Mayo Clinic and Foundation, Rochester, MN,
USA
3
Department of Radiology, Mayo Clinic and Foundation, Rochester, MN, USA
4
Department of Radiology, University of California San Francisco School of Medicine,
San Francisco, CA, USA
*Denotes equal contribution
Correspondence: Paul Thompson
Professor of Neurology, Mark and Mary Stevens Neuroimaging & Informatics Institute,
Keck School of Medicine, University of Southern California, Los Angeles, CA, USA
pthomp@usc.edu
Keywords: Alzheimer’s disease, ADNI3, White Matter, DTI, Multi-site, Harmonization,
TDF, ComBat
128
Abstract
Brain imaging with diffusion-weighted MRI (dMRI) is sensitive to microstructural white
matter changes associated with brain aging and neurodegeneration. In its third phase,
the Alzheimer’s Disease Neuroimaging Initiative (ADNI3) is collecting data across multiple
sites and scanners using different dMRI acquisition protocols, to better understand
disease effects. It is vital to understand when data can be pooled across scanners, and
how the choice of dMRI protocol affects the sensitivity of extracted measures to
differences in clinical impairment. Here, we analyzed ADNI3 data from 317 participants
(mean age: 75.4±7.9 years; 143 men/174 women), who were each scanned at one of 47
sites with one of six dMRI protocols using scanners from three different manufacturers.
We computed four standard diffusion tensor imaging (DTI) indices including fractional
anisotropy (FA
DTI
) and mean, radial, and axial diffusivity, and one FA index based on the
tensor distribution function (FA
TDF
), in 24 bilaterally averaged white matter regions of
interest. We found that protocol differences significantly affected dMRI indices, in
particular FA
DTI
. We ranked the diffusion indices for their strength of association with four
clinical assessments. In addition to diagnosis, we evaluated cognitive impairment as
indexed by three commonly used screening tools for detecting dementia and Alzheimer’s
disease: the Alzheimer’s Disease Assessment Scale (ADAS-cog), the Mini-Mental State
Examination (MMSE), and the Clinical Dementia Rating scale sum-of-boxes (CDR-sob).
Using a nested random-effects model to account for protocol and site, we found that
across all dMRI indices and clinical measures, the hippocampal-cingulum and fornix
129
(crus) / stria terminalis regions most consistently showed strong associations with clinical
impairment. Overall, the greatest effect sizes were detected in the hippocampal-cingulum
and uncinate fasciculus for associations between axial or mean diffusivity and CDR-sob.
FA
TDF
detected robust widespread associations with clinical measures, while FA
DTI
was
the weakest of the five indices for detecting associations. Ultimately, we were able to
successfully pool dMRI data from multiple acquisition protocols from ADNI3 and detect
consistent and robust associations with clinical impairment and age.
1 Introduction
Alzheimer’s disease (AD) is the most common type of dementia, affecting
approximately 10% of the population over age 65 (Alzheimer’s Association, 2018). As life
expectancy increases, there is an ever-increasing need for sensitive biomarkers of AD -
to better understand the disease, and to serve as surrogate markers of disease burden
for use in treatment and prevention trials. The Alzheimer’s Disease Neuroimaging
Initiative (ADNI) is an ongoing large-scale, multi-center, longitudinal study designed to
improve methods for clinical trials by identifying brain imaging, clinical, cognitive, and
molecular biomarkers of AD and aging. Now in its third phase (ADNI3), ADNI continues
to incorporate newer technologies as they become established (Jack et al., 2015); data
from ADNI, collected at participating sites across the U.S. and Canada, is publicly
available and has been used in a diverse range of publications (Veitch et al., 2018).
ADNI’s second phase (ADNI2) introduced to the initiative the use of diffusion-
weighted MRI (dMRI) as an additional approach for tracking AD progression (Jack et al.,
2015). dMRI has since been used in numerous studies to understand the effects of AD
130
on white matter (WM) microstructure and brain connectivity (Daianu et al., 2013a,b; Nir
et al., 2013; Prasad et al., 2013). Some of these approaches assess dMRI indices in
normal appearing WM (Giulietti et al., 2018), while others use tractography and graph-
theoretic analysis to study abnormalities in structural brain networks (Nir et al., 2015; Hu
et al., 2016; Maggipinto et al., 2017; Sulaimany et al., 2017; Powell et al., 2018; Sanchez-
Rodriguez et al., 2018). In aggregate, these studies point to WM abnormalities in AD,
which may play a key role in early pathogenesis and diagnosis (Sachdev et al., 2013).
ADNI2 acquired dMRI data with one acquisition protocol from approximately one
third of enrolled participants at the subset of ADNI sites that used 3 tesla General Electric
(GE) scanners. To ensure that dMRI could be collected from all enrolled participants,
ADNI3 developed new dMRI protocols for all GE, Siemens and Philips scanners used
across ADNI sites. Now, data is being acquired with seven different dMRI acquisition
protocols (see methods for details; http://adni.loni.usc.edu/methods/documents/mri-
protocols/). ADNI3 began in October 2016, and has already acquired data from over 300
participants. dMRI spatial resolution was improved between ADNI2 and ADNI3 by
reducing the voxel size from 2.7x2.7x2.7 mm to 2.0x2.0x2.0 mm. While voxel size (i.e.,
spatial resolution) remains consistent across all seven ADNI3 protocols, angular
resolution (the number of gradient directions) varies across protocols to accommodate
scanner restrictions and to ensure that the multi-modal scanning session is completed in
under 60 minutes. Although many large-scale multi-site DTI studies have obtained
consistent results even when acquisition protocols across sites are not harmonized in
advance (Jahanshad et al., 2013; Kochunov et al., 2014; Acheson et al., 2017; Kelly et
al., 2018), differences in dMRI acquisition parameters, including vendor, voxel size, and
131
angular resolution, are known to affect derived dMRI measures (Alexander et al., 2001;
Cercignani et al., 2003; Zhan et al., 2010; Zhu et al., 2011). As a result, improved
harmonization of multi-site diffusion data is of great interest (Grech-Sollars et al., 2015;
Pohl et al., 2016; Palacios et al., 2017). For example, ComBat - originally developed to
model and remove batch effects from genomic microarray data (Johnson et al., 2007) -
was one of the most effective methods for harmonizing DTI measures in a recent
comparison of such techniques (Fortin et al., 2017).
Here we tested whether standard diffusion tensor imaging (DTI)-derived anisotropy
and diffusivity indices, calculated from multiple imaging protocols in ADNI3, can be pooled
and harmonized to show robust associations with age and four clinical assessments. In
addition to diagnosis, cognitive impairment was assessed with three commonly used
screening tools for detecting dementia and Alzheimer’s disease: the Alzheimer’s Disease
Assessment Scale (ADAS-cog; Rosen et al., 1984), the Mini-Mental State Examination
(MMSE; Folstein et al., 1975), and the Clinical Dementia Rating scale sum-of-boxes
(CDR-sob; Berg, 1988). For the rest of the paper we refer to these tools as “cognitive
measures”. In addition to standard DTI indices - the fractional anisotropy (FA
DTI
), mean
diffusivity (MD
DTI
), radial diffusivity (RD
DTI
), and axial diffusivity (AxD
DTI
) - we also
evaluated a modified measure of FA, derived from the tensor distribution function (FA
TDF
;
Leow et al., 2009) which can be more sensitive to neurodegenerative disease-related WM
abnormalities than FA
DTI
across high- and low-angular resolution dMRI (Nir et al., 2017).
The TDF model addresses well-established limitations of the standard single-tensor
diffusion model - which cannot resolve complex profiles of WM architecture such as
132
crossing or mixing fibers, present in up to 90% of WM voxels (Tournier et al., 2004;
Descoteaux et al., 2007, 2009; Jeurissen et al., 2013).
In 24 WM regions of interest (ROIs), we ranked these five anisotropy and diffusivity
indices, in terms of their strength of association with key clinical measures, to identify
dMRI indices that may help understand and track AD progression. We hypothesized that
the diffusion indices from ADNI2 (Nir et al., 2013, 2017) would still be associated with
clinical measures of disease burden in ADNI3 - despite the variation in protocols. We
hypothesized that when data were pooled across ADNI3 protocols: (1) higher diffusivity
and lower anisotropy in the temporal lobe white matter would be most sensitive to
cognitive impairment, with highest effect sizes for associations with CDR-sob, and (2)
FA
TDF
would detect associations with clinical impairment with higher effect sizes than
FA
DTI
.
2 Methods
2.1 ADNI participants
Baseline MRI, DTI, diagnosis, demographics, and cognitive measures were
downloaded from the ADNI database (https://ida.loni.usc.edu/). This analysis was
performed when data collection for ADNI3 was still ongoing (May 2018), and reflects the
data available on April 30, 2018. Of the 381 participants scanned to date, 55 were
excluded after quality assurance: this included ensuring complete clinical and
demographic information, and image-level quality control (removing scans with severe
133
motion, missing volumes, or corrupt files). To ensure sufficient statistical power to assess
differences in data collected with different protocols, we evaluated only those protocols
with complete available data for at least 10 participants at the time of download; we did
not assess protocol GE36, for which scans from 9 of 12 participants passed quality
assurance. Details on excluded participants are outlined in Supplementary Table 1.
317 remaining participants - from 47 scanning sites - were included in the analysis
(mean age: 75.4±7.9 yrs; 143 men, 174 women; Table 1): 211 were elderly cognitively
normal controls (CN; mean age: 74.5±7.3 yrs; 84 men, 127 women), 84 were diagnosed
with mild cognitive impairment (MCI); mean age: 76.3±8.1 yrs; 48 men, 36 women), and
22 were diagnosed with Alzheimer’s disease (AD; mean age: 80.6±10.5 yrs; 11 men, 11
women). We note that two of the ADNI2 diagnostic categories - CN and Significant
Memory Concern (SMC) - are combined and identified as CN in ADNI3. ADNI2’s early
and late MCI categories are combined and identified as MCI in ADNI3.
2.2 Clinical assessments
In addition to diagnosis, we indexed cognitive impairment using total scores from
commonly used screening tools for detecting dementia and AD (Table 1): the Alzheimer’s
Disease Assessment Scale 13 (ADAS-cog), the Mini-Mental State Examination (MMSE),
and the Clinical Dementia Rating scale sum-of-boxes (CDR-sob). We refer to these tools
as “cognitive measures”, but recognize the limitations of these assessments as proxy
measures of specific cognitive abilities (Balsis et al., 2015). The ADAS-cog is frequently
used in pharmaceutical trials, with scores ranging from 0-70; higher scores represent
134
more severe cognitive dysfunction (Rosen et al., 1984). MMSE is more often used by
clinicians and researchers in assessing cognitive aging. Scores for MMSE range from 0-
30; lower scores typically indicate greater cognitive dysfunction (Folstein et al., 1975).
CDR-sob is used primarily in clinical trials and in clinical practice for evaluating disease
severity including the mild and early symptomatic stages of dementia. It is calculated
based on the sum of severity ratings in six domains (‘boxes’) - memory, orientation,
judgment and problem solving, community affairs, home and hobbies, and personal care.
Scores range from 0 (no dementia) to 3 (severe dementia; Rosen et al., 1984). These
evaluations are among the measures used in diagnosing ADNI participants. Not all
cognitive measures were available for every participant (MMSE, N=315; CDR-sob,
N=316, and ADAS-cog, N=278; Supplementary Table 2 lists these by protocol).
135
Table 1. Demographic and clinical measures for participants in ADNI3, subdivided by
dMRI protocol. We report the average age, MMSE, CDR-sob, and ADAS-cog measures,
and their standard deviations.
Protocols Demographics
Clinical Assessments
Diagnosis Cognitive Measures
+
Name
Total
N
Sites
Age
(yrs)
Male CN MCI AD MMSE
*
CDR-
sob
*
ADAS-
cog
*
GE54 65 8
76.7 ±
7.3
32
(49.2%)
45 16 4
28.50 ±
3.26
0.78 ±
1.81
11.75 ±
6.81
P33 24 3
78.1 ±
7.1
13
(54.2%)
17 4 3
28.75 ±
2.03
1.31 ±
2.84
13.32 ±
6.76
P36 19 4
75.3 ±
6.6
7 (36.8%) 12 7 0
28.21 ±
2.39
0.76 ±
1.35
12.63 ±
5.12
S31 36 9
72.8 ±
8.6
15
(41.7%)
21 10 5
28.31 ±
2.77
0.79 ±
1.35
11.54 ±
5.25
S55 153 18
75.0 ±
8.4
66
(43.1%)
100 43 10
27.94 ±
3.28
0.95 ±
2.05
11.96 ±
5.65
S127 20 5
75.3 ±
5.4
10
(50.0%)
16 4 0
28.80 ±
1.70
0.33 ±
0.75
10.27 ±
2.83
TOTAL 317 47
75.4 ±
7.9
143
(45.1%)
211 84 22
28.23 ±
3.01
0.87 ±
1.91
11.89 ±
5.78
*
Data not available for all participants: MMSE N=315; CDR-sob N=316 and ADAS-cog
N=278.
+
We recognize the limitations of these assessments as proxy measures of specific cognitive
abilities (Balsis et al., 2015).
2.3 Diffusion MRI acquisition protocols
ADNI3 incorporated dMRI protocols for 3 tesla Siemens, Philips, and GE scanners.
ADNI2, the first phase of ADNI to include diffusion MRI, only prescribed dMRI protocols
for GE scanners. The available scanners span a wide range of software capabilities, such
as support (or the lack of it) for custom diffusion gradient tables and/or simultaneous multi-
slice acceleration. Including additional scanners while staying in a 7-10 minute scan
duration resulted in data acquired with seven different acquisition protocols - of which six
had sufficient sample sizes to be evaluated here. Protocols varied in the number of
136
diffusion weighted imaging (DWI) directions (i.e., angular resolution), and the number of
non-diffusion sensitized gradients (b0 images), which serve as a reference to assess
diffusion-related decay of the MR signal. Voxel size across all ADNI3 protocols was
2.0x2.0x2.0 mm
3
and 2.7x2.7x2.7 mm
3
in ADNI2. Table 2 summarizes the different
protocols.
Table 2. ADNI diffusion MRI acquisition protocols
Name Scanner Protocol
b0
Volumes
DWI
Volumes
Total
Volumes
Time
(min)
Total
N
ADNI3
GE36 GE
Basic
Widebore
25x
4 b=0
s/mm
2
32
b=1000
s/mm
2
36 9:52 --
GE54 GE Basic 25x
6 b=0
s/mm
2
48
b=1000
s/mm
2
54 7:09 65
P33 Philips
Basic
Widebore
1 b=0
s/mm
2
32
b=1000
s/mm
2
33 7:32 24
P36 Philips
Basic
Widebore
R3
1 b=0
s/mm
2
3 b=2
s/mm
2
32
b=1000
s/mm
2
36 6:54 19
P54 Philips Basic R5
1 b=0
s/mm
2
5 b=2
s/mm
2
48
b=1000
s/mm
2
54 8:05 --
S31 Siemens Basic VB17
1 b=0
s/mm
2
30
b=1000
s/mm
2
31 7:02 36
S55 Siemens
Basic Skyra
E11 &
Prisma D13
7 b=0
s/mm
2
48
b=1000
s/mm
2
55 9:18 153
S127 Siemens
Advanced
Prisma
VE11C
13 b=0
s/mm
2
48
b=1000
s/mm
2
61 7:25* 20
ADNI2 G46 GE
Discovery
MR750 &
MR750w,
Signa
HDx &
HDxt
5 b=0
s/mm
2
41
b=1000
s/mm
2
46
7:00-
10:00
59
*reflects the time to acquire the full multi-shell protocol (127 volumes), not the single-
shell subset
137
There is currently one multi-shell multiband protocol for Siemens Advanced Prisma
scanners (S127). As ADNI3 is still in its early stages, GE and Philips protocols for multi-
shell acquisition have not yet been finalized, so only 20 multi-shell scans were available
for analysis at the time of writing. Here our goal was to evaluate single-shell dMRI indices
across protocols, so we used a subsample of the 127 DWI volumes from the S127 multi-
shell protocol to include only 13 b=0 and 48 b=1000 s/mm
2
DWI volumes (removing
6 b=500 s/mm
2
and 60 b=2000 s/mm
2
volumes).
The Philips Basic Widebore R3 protocol (P36) included three b=2 s/mm
2
volumes
and one b=0 s/mm
2
, because Philips scanners cannot acquire more than one b=0 s/mm
2
.
The Philips Basic Widebore (P33) was not a prescribed protocol, but rather acquired from
Philips sites with a software version less than 5.0 that could not acquire the b=2 s/mm
2
volumes.
2.4 dMRI preprocessing and scalar indices
All DWI were preprocessed using the ADNI2 diffusion tensor imaging (DTI)
analysis protocol as in Nir et al., (2013). Briefly, we corrected for head motion and eddy
current distortion, removed extra-cerebral tissue, and registered each participant’s DWI
to the respective T1-weighted brain to correct for echo planar imaging (EPI) distortion.
Details of the preprocessing steps may be found here:
https://adni.bitbucket.io/reference/docs/DTIROI/DTI-ADNI_Methods-Thompson-
138
Oct2012.pdf. All DWI and T1-weighted images were visually checked for quality
assurance.
Scalar dMRI indices were derived from two reconstruction models: the single
tensor model (DTI; Basser et al., 1994) and the tensor distribution function (TDF; Leow
et al., 2009). From the single tensor model, FA
DTI
, AxD
DTI
, MD
DTI
, and RD
DTI
scalar maps
were generated. In contrast to the single tensor model, the TDF represents the diffusion
profile as a probabilistic mixture of tensors that optimally explain the observed DWI data,
allowing for the reconstruction of multiple underlying fibers per voxel, together with a
distribution of weights, from which the TDF-derived form of FA (FA
TDF
) was calculated
(Nir et al., 2017).
2.5 White matter tract atlas ROI summary measures
Images were processed as reported previously (Nir et al., 2013). Briefly, the FA
image from the Johns Hopkins University single subject Eve atlas (JHU-DTI-SS;
http://cmrm.med.jhmi.edu/cmrm/atlas/human_data/file/AtlasExplanation2.htm) was
registered to each participant’s corrected FA image using an inverse consistent mutual
information based registration (Leow et al., 2007); the transformation was then applied to
the atlas WM parcellation map (WMPM) ROI labels (as shown in Figure 7; Mori et al.,
2008) using nearest neighbor interpolation. Mean anisotropy and diffusivity indices were
extracted from 24 WM ROIs total (Table 3): 22 ROIs averaged bilaterally, the full corpus
callosum, and a summary across all ROIs (full WM).
139
Table 3. The following 24 ROIs from the JHU atlas (Mori et al., 2008) were analyzed.
CST Corticospinal tract SLF Superior longitudinal fasciculus
CP Cerebral peduncle SFO Superior fronto-occipital fasciculus
ALIC Anterior limb of internal capsule IFO Inferior fronto-occipital fasciculus
PLIC Posterior limb of internal capsule SS Sagittal stratum
RLIC Retrolenticular part of internal capsule EC External capsule
PTR Posterior thalamic radiation UNC Uncinate fasciculus
ACR Anterior corona radiata GCC Genu of corpus callosum
SCR Superior corona radiata BCC Body of corpus callosum
PCR Posterior corona radiata SCC Splenium of corpus callosum
CGC Cingulum (cingulate gyrus) CC Full corpus callosum
CGH Cingulum (hippocampal bundle) TAP Tapetum
Fx/ST Fornix (crus) / stria terminalis Full WM Full white matter
2.6 Comparing the ADNI2 and ADNI3 protocols in cognitively normal
participants
2.6.1 Sample sizes for the ADNI2 and ADNI3 cognitively normal participants
We evaluated the six ADNI3 protocols and the ADNI2 protocol using scans from
cognitively normal (CN) participants only. Of 85 CN participants in ADNI2 with dMRI, 30
rolled over to ADNI3. To avoid duplication, and boost the number of scans available for
each protocol, we did not include all these roll-over participants in the ADNI3 group. 26
CN roll-over participants were included in the ADNI3 group. Four CN roll-over participants
were scanned with the S55 protocol, and due to the larger sample size already available
for that protocol (N=156), we included these four in the ADNI2 group. In total, 59 out of
85 ADNI2 CN participants were included in the ADNI2 group and the remaining 26 were
140
kept in the ADNI3 group for a total of 207 ADNI3 CN participants (see Supplementary
Table 3 for CN demographics by ADNI phase and protocol).
2.6.2 Assessing age effects
In CN participants, multivariate random-effects linear regressions were used to
assess whether dMRI indices from each ADNI protocol individually were associated with
age, controlling for sex and age*sex interactions as fixed variables, and acquisition site
as a random variable. dMRI indices for the CN group were subsequently pooled across
ADNI3 protocols (N=207), or ADNI3 and ADNI2 protocols (N=266) and tested for
associations with age using an analogous model, but with protocol and acquisition site as
nested random variables (e.g., 8 sites used protocol GE54, and 3 sites used protocol P33,
so the acquisition site grouping variable is nested within the protocol grouping variable).
We used the false discovery rate (FDR) procedure to correct for multiple comparisons (q
= 0.05; Benjamini and Hochberg, 1995) across the 24 ROIs assessed for each dMRI
index. Regions that survive a more stringent Bonferroni correction at an alpha of 0.05 (p
≤ 0.05/24=0.0021) are also shown in the Supplements.
2.6.3 Effect of protocol on dMRI indices
In CN participants, we tested for significant differences in dMRI indices between
the seven ADNI protocols using analyses of covariance (ANCOVAs), adjusting for age,
sex, and age*sex interactions as fixed variables, and acquisition site as a random
141
variable. For each dMRI index, we used FDR to correct for multiple comparisons across
the 24 ROIs assessed. Pairwise tests were performed to directly compare protocols. In
total, there were 504 tests per dMRI index: 24 ROIs * 21 pairs of protocol comparisons
(protocol 1 vs 2, protocol 1 vs 3, etc). As before, we used FDR to account for multiple
comparisons.
2.6.4 dMRI harmonization with ComBat
ComBat uses an empirical Bayes framework to reduce unwanted variation in multi-
site data due to differences in acquisition protocol, while preserving the desired biological
variation in the data (Fortin et al., 2017). In the CN participants from ADNI2 and ADNI3,
we ran ComBat on each of the dMRI indices, including age, sex, age*sex, and information
from all 24 ROIs to inform the statistical properties of the protocol effects. Random-effects
regressions tested for dMRI microstructural associations with age, covarying for sex and
age*sex as fixed variables and site as a random variable; ANCOVAs and pairwise tests
of dMRI differences between protocols were repeated for the harmonized ROI data.
2.7 Clinical assessments and their relation to pooled ADNI3 diffusion indices
Multivariate random-effects linear regressions were used to test associations
between 5 dMRI indices in each of the 24 WM ROIs and the three cognitive measures
(ADAS, MMSE, CDR-sob), and with diagnosis. Due to the limited available sample size
for AD participants (N=22), and their uneven distribution across the acquisition protocols
142
tested here, we compared only groups of people with CN and MCI diagnoses. Age, sex,
and age*sex interactions were controlled for as fixed effects, and the protocol and
acquisition site were modeled as nested random variables. FDR was again used to
correct for 24 ROI tests (q = 0.05; Benjamini and Hochberg, 1995), in addition to
Bonferroni corrections (p ≤ 0.05/24=0.0021) available in the Supplements. Effect sizes
for associations were determined using the d-value standardized coefficient (Rosenthal
and Rosnow, 1991):
3 Results
3.1 ADNI2 and ADNI3 protocols in cognitively normal participants
3.1.1 Age effects in cognitively normal participants from ADNI2 and ADNI3
protocols
When data were pooled across ADNI2 and ADNI3, significant associations with
age were detected throughout the WM. Figure 1a shows effect sizes for ROIs significantly
associated with age after FDR multiple comparisons correction (tabulated results and
more stringent Bonferroni thresholds are shown in Supplementary Table 4). Lower
FA
TDF
and higher diffusivity indices were significantly associated with older age in all 24
ROIs. For FA
DTI
, 22 ROIs were significantly associated with age. The largest effect size
143
was detected with FA
TDF
in the Fornix (crus) / stria terminalis (Fx/ST; d = -1.459; p =
5.07x10
-21
). The Fx/ST, genu of corpus callosum (GCC) and full WM consistently showed
one of the 10 largest effect sizes across dMRI indices.
The mean ages of the CN participants assessed in the two phases of ADNI were
significantly different (p = 0.049; ADNI2 mean age: 72.4±6.6 yrs; ADNI3 mean age:
74.5±7.4 yrs; demographics in Supplementary Table 3). Pairwise tests comparing the
mean age of CN participants scanned with each protocol also showed significant
differences between those scanned with S31 and two other protocols: GE54 and S31 (p
= 0.026); P33 and S31 (p = 0.0037). Due to differences in age and sample size between
protocols and phases, effect sizes could not be directly compared (Button et al., 2013),
but the directions of associations with age were largely consistent for ADNI2 and ADNI3
phases separately, and each ADNI3 protocol (Figures 1-2). Each ADNI protocol showed
directionally consistent associations in more than 89% of tests (24 ROIs * 5 dMRI indices),
except for P36 which was consistent in 81%, but had the smallest sample size (N=12;
Figure 2b; Supplementary Tables 5-11). FA
TDF
and all three diffusivity indices were
consistent in ≥ 96% of tests (24 ROIs * 8 protocols/phases), while FA
DTI
was only
consistent in 88% of tests. Most of the associations detected in the unexpected direction
for each protocol were driven by FA
DTI
. None of the associations in the unexpected
direction were significant after multiple comparisons correction, and only 2 had a p ≤ 0.05.
144
Figure 1. (A) For each dMRI index, the absolute values of effect sizes (d-value) are
plotted for regional WM microstructural associations with age when all ADNI3 dMRI data
are pooled, adjusting for any site or protocol effects. For each test, we note the number
of significant ROIs, as indicated by filled shapes, and the corresponding FDR significance
p-value threshold (q = 0.05). See Supplementary Table 4 for complete tabulated results.
(B) Here, we plot the residuals of diffusivity and anisotropy indices in the full WM (y-axis)
against age (x-axis) after regressing out the effects of sex in CN participants from each
protocol separately. Individual level residuals from each protocol are plotted with a
different color. Despite protocol differences, age effects are evident across protocols.
145
Figure 2 shows consistent associations in the full WM by protocol. As
demographic and sample size variability between protocols affect detected effect sizes,
we also evaluated full WM dMRI associations with age in an age- and sex-matched subset
of 12 participants from each protocol (total N=84; demographics in Supplementary Table
3); a comparison of the effect sizes between protocols suggests that the protocols with
greatest total number of diffusion-weighted (b=1000 s/mm
2
) and non-diffusion sensitized
(b0) gradients may detect larger effects (S127 followed by S55; Supplementary Figure
1).
146
Figure 2. (A) Effect sizes (d-value) for each ADNI protocol and phase show the direction
of dMRI associations with age in the full WM are consistent. Due to differences in age
and sample size between protocols and phases, effect sizes could not be directly
compared. (B) For each protocol and phase, the number of ROIs (out of 24), that show
the expected association direction, regardless of significance, are reported for each dMRI
index, revealing consistent associations across tests, except for protocol P36 which has
the smallest sample size, and FA
DTI
, which shows the smallest effect sizes and fewest
significant associations across protocols when pooled.
147
3.1.2 Effect of protocol on dMRI indices from cognitively normal controls
The influence of dMRI acquisition protocol on mean values of the diffusion indices
is evident in boxplots of dMRI indices in the full WM for each protocol. When modeling
the mean full WM values for each diffusion index, the residuals of the statistical model
become closer to 0 after fitting the effect of protocol and site (nested as a random variable
with age, sex, and age*sex interactions as fixed effects) than when we plot the residuals
of just age, sex, and age*sex interactions (Figure 3).
148
Figure 3. Full WM mean (A) AxD
DTI
, MD
DTI
, and RD
DTI
and (B) FA
DTI
and FA
TDF
residuals
for each protocol after fitting effects of age, sex, and age*sex interactions are plotted here
in the top rows (red). Protocol has an effect on anisotropy and diffusivity measures. The
lower panels (blue) show residuals after additionally fitting protocol and site as nested
random-effects, after which the residuals across protocols are closer to 0.
149
ANCOVAs and pairwise tests for each ROI suggest there are significant
differences between protocols for all 5 dMRI indices across most ROIs (Figure 4).
ANCOVAs revealed significant protocol differences for 22 ROIs for FA
DTI
and FA
TDF
, with
the highest overall effect size detected in the anterior limb of the internal capsule (ALIC)
and the external capsule (EC) for FA
DTI
(ALIC: d = 0.648; EC: d = 0.652). AxD
DTI
had the
smallest effect size, overall, in the splenium of the corpus callosum (SCC; d = 0.106), and
only 13 ROIs showed significant AxD
DTI
differences between protocols.
150
Figure 4. (A) d-values from the ANCOVAs assessing differences in dMRI indices
between protocols, for each of the 24 ROIs; FA
DTI
shows the greatest significant
differences (largest d-values; dark red) between protocols and AxD
DTI
the fewest (dark
green). (B) We report the number of times each protocol and each dMRI index showed
significant differences in pairwise tests between protocols (out of 504 tests per index and
720 tests per protocol); AxD
DTI
was the most stable dMRI index across protocols, while
FA
DTI
was the least stable.
151
In pairwise analyses, AxD
DTI
was the most stable index across protocols, as
significant protocol differences were detected in only 20.6% of pairwise tests (24 ROIs *
21 pairwise tests), compared to FA
DTI
, the most variable index, which showed significant
protocol differences in 81.9% of tests (Figure 4b). ADNI2 was the most divergent protocol
across dMRI indices, showing differences in 36.3% of tests.
3.1.3 Diffusion MRI harmonization with ComBat
After using ComBat to harmonize dMRI indices across protocols, ANCOVAs
revealed that significant protocol differences in dMRI indices were all but eliminated
across ROIs (Supplementary Figure 2a); significant protocol differences were detected
only in the CST, for each of the dMRI indices. The number of pairwise tests for which
each protocol showed significant differences in dMRI indices decreased by 93.8% with
ComBat (Supplementary Figure 2b).
After harmonization, we still detected significant associations between age and
dMRI indices from ADNI2 and ADNI3 pooled in the same number of ROIs
(Supplementary Table 12). ComBat correction did not significantly change effect sizes,
while correcting for effects of protocol (Supplementary Figure 3). In Figure 5 we show
effect sizes before and after harmonization with ComBat in the Full WM, Fx/ST, and GCC,
the three ROIs that consistently showed one of the 10 largest effect sizes for associations
with age across all five diffusion indices (for changes by protocol see Supplementary
Figures 4-6). As harmonization with ComBat did not improve or change results found
152
with random-effect linear regressions, we proceeded to test clinical associations without
applying a ComBat transformation.
Figure 5. Beta-values and error bars representing standard error from the association
between each diffusion index and age in CN participants, before and after ComBat
harmonization. We show the three ROIs that consistently showed one of the 10 largest
effect sizes for associations with age across all five diffusion indices (see Supplementary
Figure 3 for all ROIs). Compared to pre-ComBat analyses, effect sizes are marginally
different across indices, but still within the standard error.
153
3.2 Cognitive measure associations with pooled ADNI3 dMRI indices
Pooling data across ADNI3, we detected significant associations between all three
cognitive measures and regional dMRI indices throughout the WM. Greater cognitive
impairment was associated with lower anisotropy and higher diffusivity. Figure 6a-c
shows effect sizes for ROIs significantly associated with each cognitive measure after
FDR multiple comparisons correction (for tabulated results and more stringent Bonferroni
corrections, please see Supplementary Tables 13-15). Across tests (5 dMRI indices * 3
cognitive measures), the hippocampal-cingulum (CGH), fornix (crus) / stria terminalis
region (Fx/ST), and the full WM consistently showed one of the 10 largest effect sizes
(see Supplementary Figures 7-9 for associations with indices in the CGH, Fx/ST, and
full WM, by protocol). In 14 of 15 tests, the CGH consistently showed one of the top two
largest effect sizes (CGH FA
DTI
association with CDR-sob was the third largest), along
with the uncinate fasciculus (UNC), which was top two in 12 of 15 tests (while significant,
cognitive associations with UNC FA
DTI
never showed one of the largest effect sizes).
154
Figure 6. For each dMRI index, the absolute values of effect sizes (d-value) are plotted
for regional WM microstructural associations with clinical measures. Lower anisotropy
and higher diffusivity were significantly associated with (A) higher CDR-sob, (B) lower
MMSE, (C) higher ADAS-cog, and (D) an MCI diagnosis, when all ADNI3 dMRI data are
pooled, adjusting for any site or protocol effects. For each test, we note the number of
significant ROIs, as indicated by filled shapes, and the corresponding FDR significance
p-value threshold (q = 0.05). See Supplementary Tables 13-16 for complete tabulated
results.
FA
DTI
showed significant associations in the fewest ROIs: 55 out of 72 tests
(76.4%; 24 ROIs * 3 cognitive measures) were significant. FA
TDF
showed more
widespread associations with cognitive measures throughout WM ROIs: 69 out of 72 tests
(94.4%; 24 ROIs * 3 cognitive measures) were significant. Effect sizes were consistently
lower for FA
DTI
than for the other dMRI indices, across all 3 cognitive measures; the
largest FA
DTI
effect size was consistently found in the Fx/ST, followed by the CGH or the
155
GCC. The strongest FA
DTI
association was in the Fx/ST with CDR-sob (d = -0.681, p =
7.01x10
-8
). Compared to FA
DTI
, FA
TDF
showed larger effect sizes; across cognitive tests,
the strongest FA
TDF
associations were detected in the uncinate fasciculus (UNC) with
CDR-sob (d = -1.244; p = 1.39x10
-20
), followed by the CGH (d = -1.213; p = 8.86x10
-20
).
CDR-sob effect sizes for FA
DTI
and FA
TDF
in the CGH, UNC, Fx/ST, and full WM are
depicted by protocol in Supplementary Figure 10, revealing consistently larger effect
sizes for FA
TDF
across protocols.
Cognitive associations with all of the diffusivity indices were widespread: significant
associations were detected in 207 out of 216 tests (95.8%; 24 ROIs * 3 cognitive
measures * 3 diffusivity indices). Regional measures of AxD
DTI
consistently showed the
largest effect sizes across all cognitive measures (CDR-sob and the UNC: d = 1.344, p =
3.13x10
-23
; MMSE and the CGH: d = -1.178, p = 7.87x10
-19
; ADAS-cog and the UNC: d
= 1.048, p = 1.09x10
-13
).
Of the three cognitive measures, CDR-sob associations showed the largest effect
sizes across dMRI indices (in the UNC followed by the CGH for all indices except FA
DTI
);
the largest effect sizes across all tests were detected with AxD
DTI
(UNC: d = 1.344) and
MD
DTI
(UNC: d = 1.342, p = 3.47x10
-23
). Figure 7 shows the distribution of the effect sizes
for CDR-sob throughout the brain. Temporal lobe regions (UNC, CGH, IFO, SS)
frequently showed greatest effect sizes (for ADAS-cog and MMSE figures, see
Supplementary Figures 11-12). Effect size was not correlated with ROI size
(Supplementary Figure 13), consistent with prior studies of other disorders (Kelly et al.,
2018).
156
Figure 7. Effect size (absolute d-value) maps of WM regions that show significant
associations with CDR-sob - the cognitive measure with the largest effect sizes - reveal
widespread associations throughout the WM, with particularly strong associations in the
temporal lobes (SS, IFO, UNC, and CGH; light green regions show the largest effect
sizes). As expected, positive associations were detected between CDR-sob and (A)
AxD
DTI
(FDR critical threshold p = 1.78x10
-4
) (B) MD
DTI
(FDR critical threshold p =
3.64x10
-4
) and (C) RD
DTI
(FDR critical threshold p = 6.92x10
-3
); higher diffusivity was
associated with greater cognitive impairment. Lower (D) FA
DTI
(FDR critical threshold p =
0.025) and (E) FA
TDF
(FDR critical threshold p = 7.73x10
-3
) were also associated with
greater impairment, but FA
DTI
associations were detected in fewer regions with weaker
effect sizes compared to FA
TDF
.
157
3.3 CN vs MCI diagnosis associations with pooled ADNI3 dMRI indices
For each diffusion index, Figure 6d shows the significant regional effect sizes for
differences between CN and MCI participants. Widespread diffusivity differences were
detected, with significantly higher diffusivity in MCI participants in 21 out of 24 ROIs
(Supplementary Table 16 and Supplementary Figure 14). Only three regions showed
significantly lower FA
DTI
in MCI participants – Fx/ST (d = -0.460; p = 3.89x10
-4
), CGH (d
= -0.410; p = 1.53x10
-3
), and the posterior thalamic radiation (PTR; d = 0.367; p = 4.55x10
-
3
). On the other hand, FA
TDF
was significant in 20 out of 24 ROIs, similar to diffusivity
indices. FA
TDF
and diffusivity indices in the CGH showed the largest effect sizes (AxD
DTI
d = 0.681; p = 2.26x10
-7
, MD
DTI
d = 0.700; p = 1.15x10
-7
; RD
DTI
d = 0.679; p = 2.41x10
-7
;
FA
TDF
d = -0.622; p = 2.00x10
-6
).
For all three cognitive measures, and in the comparison between CN and MCI
participants, the CGH and Fx/ST were the only regions that survived multiple
comparisons correction across all dMRI indices. The Fx/ST always had the largest effect
size in FA
DTI
tests. The UNC showed either the first or second largest effect size
(alternating with CGH) across diffusivity indices and FA
TDF
tests, but was significant only
for cognitive measure associations with FA
DTI
(i.e., 3 of 4 clinical tests).
158
4 Discussion
This study has three main findings: (1) When data were pooled from the six
available diffusion MRI protocols used in ADNI3, anisotropy and diffusivity indices showed
robust associations with MCI diagnosis, and with three common cognitive measures:
MMSE, ADAS-cog, and CDR-sob; (2) When using a higher-order diffusion model, the
tensor distribution function (TDF), the derived measure of anisotropy (FA
TDF
) showed
stronger and more widespread associations with clinical impairment than the standard
DTI anisotropy measure (FA
DTI
); (3) Despite significant differences in protocols, for each
dMRI index, we were able to detect consistent associations with clinical measures in
ADNI3 participants, and age in ADNI2 and ADNI3 CN participants.
Accumulation of amyloid plaques and neurofibrillary tangles (NFT) in the brain
(Braak & Braak, 1991; Braak & Braak 1996; Frank et al., 2003, Shaw et al., 2007) can
directly impact WM (Lee et al., 2004; Roth et al., 2005), promoting myelin degeneration
and axonal loss (Braak and Braak 1996; Kneynsberg et al., 2017). While many factors
drive anisotropy and diffusivity measures from DTI, higher anisotropy values may
indicate, in part, more coherent intact axons, while lower anisotropy and higher diffusivity
may reflect factors such as axonal injury and demyelination, among other factors
(Beaulieu et al., 2002; Song et al., 2003; Song et al., 2005; Harsan et al., 2006; Le Bihan
and Johansen-Berg 2012; Kantarci et al., 2017; Moore et al., 2018). In this paper, lower
anisotropy values and higher diffusivity values were correlated with clinical impairment,
most strongly in the hippocampal-cingulum and uncinate fasciculus. Along with the full
WM, reflecting global WM effects, the largest effect sizes were most frequently detected
159
in the hippocampal-cingulum and fornix (crus) / stria terminalis, WM bundles connecting
hippocampal and parahippocampal regions to the rest of the brain, consistent with
patterns of AD pathology. The histopathological validity of these findings has been
supported, specifically in a recent study that compared NFT stages in autopsy material
along with ante-mortem MRI. Elevated MD
DTI
and lower FA
DTI
significantly correlated with
higher postmortem NFT stage, particularly in the crus of the fornix, the ventral cingulum
tracts, the precuneus, and entorhinal WM (Kantarci et al., 2017).
The participants recruited for ADNI3 tend to be younger and healthier, on average,
than those in ADNI2, as they were recruited with the intention of studying the transition
from CN to mild AD (Jack et al., 2015). With few AD patients enrolled so far in ADNI3, the
primary focus of this paper was to assess three cognitive assessments (ADAS-cog, CDR-
sob, and MMSE), and to compare CN to MCI participants. MCI is now the focus of intense
research, as it is essential to find ways to clinically categorize the transitional stages
between normal aging and AD to evaluate targeted treatments, as pathophysiological
mechanisms may differ or change throughout the course of AD (Mueller et al., 2005). As
in our prior analysis of ADNI2 (Nir et al., 2013), FA
DTI
was the least sensitive DTI measure.
In ADNI3, AxD
DTI
and MD
DTI
showed the largest effect sizes. Lower FA
DTI
and higher
MD
DTI
are most frequently reported in studies of AD (Kavcic et al., 2008; Clerx et al., 2012;
Nir et al., 2013; Maggipinto et al., 2017; Mayo et al., 2017), but AxD
DTI
may be more
sensitive to unspecific microscopic cellular loss earlier in the disease (O’Dwyer et al.,
2011), perhaps making it more sensitive in the healthier participants of the ADNI3 dataset.
Similarly, in ADNI2, AxD
DTI
was the most sensitive to differences between CN and MCI
diagnoses (Nir et al., 2013).
160
Among the three cognitive assessments, CDR-sob showed the strongest
correlations with dMRI indices, in line with prior ADNI brain imaging studies (Hua et al.,
2009; Nir et al., 2013). The largest of these effects were found in temporal WM tracts
including the hippocampal-cingulum, uncinate fasciculus, sagittal stratum, and inferior
fronto-occipital fasciculus. These are all regions that show early degenerative changes
in MCI and AD (Mielke et al., 2009; Nir et al., 2013; Maggipinto et al., 2017; Powell et al.,
2018). While associations with clinical impairment were detected throughout the WM, the
region that most frequently showed the lowest effect sizes and was significant in only 3
of the 20 clinical tests, was the corticospinal tract (CST). However, the CST ROI from the
JHU WMPM atlas is limited to a small region in the inferior portion of the brain and has
been shown to be the least reliable and reproducible ROI (Jahanshad et al., 2013;
Acheson et al., 2017), suggesting alternate approaches, such as tractography-based
evaluations (Jin et al., 2017), or the use of the probabilistic JHU atlas (Hua et al., 2008),
may be more appropriate for studying the CST. Our analysis focused on white matter
microstructure, but future work assessing tract geometry and properties of anatomical
brain networks using tractography may reveal more detailed information. The validation
and harmonization of tractography methods and derived network metrics is a vast field of
research with active ongoing work (Maier-Hein et al., 2017).
DTI is widely recognized as a useful tool for studying neurodegenerative disorders
such as AD (Oishi et al., 2011; Müller and Kassubek, 2013; Abhinav et al., 2014; Acosta-
Cabronero et al., 2014; Maggipinto et al., 2017). However, at the spatial resolutions now
used, a single voxel typically captures partial volumes of different tissue compartments,
161
e.g., the intra- and extra-cellular compartments, the vascular compartment, the CSF and
myelin; each affects water diffusion and the MR signal. The DTI model cannot differentiate
these components or even crossing fibers (Tuch et al., 2002; Jbabdi et al., 2010), which
are estimated to occur in up to 90% of WM voxels at the typical dMRI resolution
(Descoteaux et al., 2009; Jeurissen et al., 2013). In healthy tissue with crossing fibers,
the DTI model may show low FA. FA
DTI
may paradoxically appear to increase in regions
where crossing fibers deteriorate in neurodegenerative diseases such as AD (Douaud et
al., 2011). FA
TDF
addresses this limitation even in low angular resolution data (Nir et al.,
2017). Here, compared to FA
DTI
, FA
TDF
showed more widespread associations with
cognitive measures and diagnosis throughout WM ROIs: FA
TDF
was significant in 89 of
the 96 tests (92.7%; 24 regions in 4 clinical tests), while FA
DTI
was only significant in 58
(60.4%). The greatest difference was seen for diagnostic associations (CN vs MCI): FA
TDF
was significant in 20 out of 24 ROIs while FA
DTI
was only significant in 3. FA
TDF
also
showed stronger effect sizes across the protocols, suggesting that tensor limitations have
likely confounded previous diffusion studies of cognitive decline that have found little or
no effects with FA (Acosta-Cabronero et al., 2010). Recently proposed biophysical
models of brain tissue may help to relate diffusion signals directly to underlying
microstructure and different tissue compartments (Harms et al., 2017). We may be able
to further disentangle questions of orientation coherence (dispersing and ‘kissing’ fibers),
fiber diameter, fiber density, membrane permeability, and myelination, which all influence
classic anisotropy and diffusivity measures derived from DTI. Several AD studies have
already used multi-shell protocols to compute diffusion indices from models that do not
assume mono-exponential decay, such as diffusion kurtosis imaging (DKI; Jensen et al.,
162
2005; Chen et al., 2017; Cheng et al., 2018; Wang et al., 2018), and multi-compartment
models such as neurite orientation dispersion and density imaging (NODDI; Zhang et al.,
2012; Colgan et al., 2016; Slattery et al., 2017; Parker et al., 2018). To date,
approximately 20 participants in ADNI have been scanned with multi-shell diffusion
protocols; in a future report, we will relate these measures to those examined here.
Large-scale, multi-site neuroimaging studies can increase the power of statistical
analyses and establish greater confidence and generalizability for findings. Most multi-
site neuroimaging studies are susceptible to variability across sites. Variability in dMRI
studies is due in part to heterogeneity in acquisition protocols, scanning parameters, and
scanner manufacturers (Zhu et al., 2009; Zhu et al., 2011; Zhu et al., 2018). Anisotropy
and diffusivity maps are affected by angular and spatial resolution (Alexander et al., 2001;
Kim et al., 2006; Zhan et al., 2010), the number of DWI directions (Giannelli et al., 2009),
and the number of acquired b-values (Correia et al., 2009). All five dMRI indices were
significantly different between protocols; AxD
DTI
was the most stable index, while FA
DTI
was the least stable, reflective of their performance in detecting associations with
cognitive measures. ADNI2 was the most divergent protocol across dMRI indices,
perhaps due to the larger voxel size in ADNI2 (2.7 mm versus 2.0 mm isotropic voxels
used in ADNI3). This is consistent with the notion that DTI measures vary with voxel size
due to partial voluming (Zhan et al., 2013). Despite differences in protocols, the directions
of associations were consistent across protocols.
ADNI3 extends dMRI acquisitions across scanner manufacturers and platforms to
maximize the number of participants scanned with dMRI; this makes it necessary to
account for site-related heterogeneities and confounds in analytical models where data
163
are pooled. Multi-site dMRI studies are becoming increasingly common, and new data
harmonization methods to adjust for site and acquisition protocol are being developed
and tested. A thorough investigation of dMRI harmonization methods is now possible with
ADNI3, one of the few publically available multi-site datasets acquired with multiple
protocols. As regional dMRI measures are available for download as part of the ADNI
database, we highlight two ways that the data may be pooled across sites: 1) performing
statistical analyses with nested random-effects models to account for site and acquisition
protocol differences, and 2) harmonizing the derived regional measures before
aggregating the data across sites. In a preliminary analysis, we showed that one
harmonization method performed on these regional measures, ComBat, reduced cross-
site differences in dMRI indices, while preserving biological relationships with age in CN
controls. The only region where differences remained after ComBat, was the CST, the
ROI with the weakest associations with clinical measures, and previously identified as
least reliable (Acheson et al., 2017). In Fortin et al. (2017), compared to other
harmonization methods, ComBat increased the number of voxels where significant
associations between age and FA
DTI
or MD
DTI
were detected. Here, the number of
significant ROIs and the magnitude of effect sizes were comparable for ComBat and
nested random-effects model approaches. This discrepancy between our findings and
that of Fortin et al, may be due to differences between studies: 1) ADNI3 includes more
sites and protocols, 2) the number of ROIs is far less than the number of participants, and
3) the age effects in the elderly populations tested here are stronger than the effects
tested in adolescents in Fortin et al. When effects are more readily detected, one
harmonization approach may not be more advantageous than others. In addition to
164
exploring additional harmonization techniques, future work should evaluate voxel-wise
ComBat approaches and the effects of harmonization beyond CN participants (i.e., across
the entire ADNI cohort).
In addition to ComBat, a number of harmonization approaches have recently been
proposed at various stages of analysis (Tax et al., 2018; Zhu et al., 2018). Site differences
can be accounted for at the time of overall group inference, such as with the random-
effects regression level correction used here, or by using a meta-analysis approach in
lieu of pooling data (Thompson et al., 2014). The data may also be transformed prior to
multi-site group-level statistics. Some methods, such as ComBat and RAVEL, use the
distribution of derived features, such as diffusivity and anisotropy measures (Fortin et al.,
2016, 2017). Alternatively, several proposed methods use information from the raw image
to adjust for acquisition variability (Zhu et al., 2018). For example, Kochunov et al. (2018)
calculated the signal to noise ratio for each protocol and include it in their regression
models. Mirzaalian et al. (2018) use voxel-wise spherical harmonic residual networks to
derive local correction parameters. Finding the best method to harmonize dMRI data is
an active topic at ‘hackathons’ and technical challenges; in 2017 and 2018, the
International Conference on Medical Image Computing and Computer Assisted
Intervention (MICCAI) hosted a computational diffusion MRI challenge to explore
approaches for data harmonization. With so many available approaches, the preliminary
random-effects regression and ComBat results from this paper serve as a first step
towards future work establishing robust approaches for combining data in ADNI3 and
other multi-site studies.
165
The current study is limited in that the sample sizes, and sample demographics,
available for each protocol vary, complicating direct comparison of the protocols (Button
et al., 2013). A matched comparison might be possible if a group of participants or a
phantom were scanned using every protocol. Even so, separating protocol differences
from differences in scanner manufacturer is difficult. We also could not directly compare
all diagnostic groups in ADNI3, as few participants with AD were scanned.
A more complete picture of brain changes in aging and AD would include imaging
metrics from other modalities, such as perfusion imaging, resting state functional MRI
(Wang et al., 2017), and radiotracer methods such as FDG-PET (Popuri et al., 2018), or
amyloid- and tau-sensitive PET (Grothe et al., 2017; Phillips et al., 2018). Genetic and
other ‘omics’ data could be analyzed as well, and may help to predict diagnostic
classification and brain aging, when combined with other neuroimaging markers (Ding et
al., 2018; Kauppi et al., 2018). While these data are all being collected as part of ADNI3
and other studies of brain aging, our focus here was on the variety of available dMRI
measures, calculated using different protocols. With this in mind, the optimal dMRI indices
to include in a multimodal study may be those that contribute the greatest independent
information beyond that available from anatomical MRI and other standard imaging
modalities. Multivariate methods - such as machine learning (Zhou et al., 2017; Wang et
al., 2018) and even deep learning (Liu et al., 2017) - may also help to extract and
capitalize on features that predict clinical decline beyond those studied here.
In addition to providing a roadmap for the new ADNI3 dMRI data, these preliminary
analyses show that despite differences in the updated dMRI protocols, diffusion indices
166
can be pooled to detect white matter microstructural differences associated with aging
and Alzheimer’s disease.
Conflict of Interest Statement
Michael W. Weiner has served on the scientific advisory boards for Lilly, Araclon, and
Institut Catala de Neurociencies Aplicades, Gulf War Veterans Illnesses Advisory
Committee, VACO, Biogen Idec, and Pfizer; has served as a consultant for Astra Zeneca,
Araclon, Medivation/Pfizer, Ipsen, TauRx Therapeutics LTD, Bayer Healthcare, Biogen
Idec, Exonhit Therapeutics, SA, Servier, Synarc, Pfizer, and Janssen; has received
funding for travel from NeuroVigil, Inc., CHRU-Hopital Roger Salengro, Siemens,
AstraZeneca, Geneva University Hospitals, Lilly, University of California, San Diego–
ADNI, Paris University, Institut Catala de Neurociencies Aplicades, University of New
Mexico School of Medicine, Ipsen, CTAD (Clinical Trials on Alzheimer’s Disease), Pfizer,
AD PD meeting, Paul Sabatier University, Novartis, Tohoku University; has served on the
editorial advisory boards for Alzheimer’s & Dementia and MRI; has received honoraria
from NeuroVigil, Inc., Insitut Catala de Neurociencies Aplicades, PMDA/Japanese
Ministry of Health, Labour, and Welfare, and Tohoku University; has received commercial
research support from Merck and Avid; has received government research support from
DOD and VA; has stock options in Synarc and Elan; and declares the following
organizations as contributors to the Foundation for NIH and thus to the NIA funded
Alzheimer’s Disease Neuroimaging Initiative: Abbott, Alzheimer’s Association,
167
Alzheimer’s Drug Discovery Foundation, Anonymous Foundation, AstraZeneca, Bayer
Healthcare, BioClinica, Inc. (ADNI 2), Bristol-Myers Squibb, Cure Alzheimer’s Fund, Eisai,
Elan, Gene Network Sciences, Genentech, GE Healthcare, GlaxoSmithKline,
Innogenetics, Johnson & Johnson, Eli Lilly & Company, Medpace, Merck, Novartis, Pfizer
Inc., Roche, Schering Plough, Synarc, and Wyeth.
Clifford R. Jack has provided consulting services for Janssen Research & Development,
LLC, and Eli Lilly.
Matt A. Bernstein was a former employee of GE Medical Systems from 1987-1998.
The authors have no commercial or financial relationships that would involve a conflict of
interest.
Author Contributions Statement
RIR, MAB, BB, CRJ, PT and MWW designed the ADNI3 diffusion MRI study. ST, AZP,
and TMN performed the image analysis. TMN, AZP, NJ and PT conceived and designed
the image analysis study. TMN, AZP, ST, NJ and PT drafted the manuscript. All authors
contributed to interpreting the results and critically revised the manuscript for intellectual
content.
168
Acknowledgements
This study builds on preliminary findings in a conference paper entitled, Ranking Diffusion
Tensor Measures of Brain Aging & Alzheimer’s Disease, which may be found in the
conference proceedings from the 14th International Symposium on Medical Information
Processing and Analysis (SIPAIM; Zavaliangos-Petropulu et al., 2018).
Funding
Data collection and sharing for ADNI was funded by National Institutes of Health Grant
U01 AG024904 and the DOD (Department of Defense award number W81XWH-12-2-
0012). Additional support was provided by NIA grant RF1 AG04191, P01 AG026572-13
and P41 EB015922. ADNI is funded by the National Institute on Aging, the National
Institute of Biomedical Imaging and Bioengineering, and through generous contributions
from the following: AbbVie, Alzheimer’s Association; Alzheimer’s Drug Discovery
Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol-Myers Squibb Company;
CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company;
EuroImmun; F. Hoffmann-La Roche Ltd and its affiliated company Genentech, Inc.;
Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research &
Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.;
Lumosity; Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx
Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.;
Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition
Therapeutics. The Canadian Institutes of Health Research is providing funds to support
169
ADNI clinical sites in Canada. Private sector contributions are facilitated by the
Foundation for the National Institutes of Health (www.fnih.org). The grantee organization
is the Northern California Institute for Research and Education, and the study is
coordinated by the Alzheimer’s Therapeutic Research Institute at the University of
Southern California. ADNI data are disseminated by the Laboratory for Neuro Imaging at
the University of Southern California. Samples from the National Centralized Repository
for Alzheimer's Disease and Related Dementias (NCRAD), which receives government
support under a cooperative agreement grant (U24 AG21886) awarded by the National
Institute on Aging (NIA), were used in this study. We thank contributors who collected
samples used in this study, as well as patients and their families, whose help and
participation made this work possible.
References
Abhinav, K., Yeh, F.-C., Pathak, S., Suski, V., Lacomis, D., Friedlander, R. M., et al.
(2014). Advanced diffusion MRI fiber tracking in neurosurgical and
neurodegenerative disorders and neuroanatomical studies: A review. Biochim.
Biophys. Acta - Mol. Basis Dis. 1842, 2286–2297.
doi:10.1016/J.BBADIS.2014.08.002.
Acheson, A., Wijtenburg, S. A., Rowland, L. M., Winkler, A., Mathias, C. W., Hong, L. E.,
et al. (2017). Reproducibility of tract-based white matter microstructural measures
using the ENIGMA-DTI protocol. Brain Behav. 7, e00615. doi:10.1002/brb3.615.
170
Acosta-Cabronero, J., Williams, G. B., Pengas, G., and Nestor, P. J. (2010). Absolute
diffusivities define the landscape of white matter degeneration in Alzheimer’s
disease. Brain 133, 529–539. doi:10.1093/brain/awp257.
Acosta-Cabronero, J., and Nestor, P. J. (2014). Diffusion tensor imaging in Alzheimer’s
disease: insights into the limbic-diencephalic network and methodological
considerations. Front. Aging Neurosci. 6, 266. doi:10.3389/fnagi.2014.00266.
Alexander, A. L., Hasan, K. M., Lazar, M., Tsuruda, J. S., and Parker, D. L. (2001).
Analysis of partial volume effects in diffusion-tensor MRI. Magn. Reson. Med. 45,
770–780. doi:10.1002/mrm.1105.
Alzheimer’s Association (2018). 2018 Alzheimer’s disease facts and figures. Alzheimer’s
Dement. 14, 367–429. doi:10.1016/j.jalz.2018.02.001.
Balsis, S., Benge, J. F., Lowe, D. A., Geraci, L., and Doody, R. S. (2015). How do scores
on the ADAS-Cog, MMSE, and CDR-SOB correspond? Clin. Neuropsychol. 29,
1002–1009. doi:10.1080/13854046.2015.1119312.
Basser, P. J., Mattiello, J., and LeBihan, D. (1994). MR diffusion tensor spectroscopy and
imaging. Biophys. J. 66, 259–67. doi:10.1016/S0006-3495(94)80775-1.
Beaulieu, C. (2002). The basis of anisotropic water diffusion in the nervous system - A
technical review. NMR Biomed. 15, 435–455. doi:10.1002/nbm.782.
Benjamini, Y., and Hochberg, Y. (1995). Controlling the false discovery rate: A practical
and powerful approach to multiple testing. J. R. Stat. Soc. Ser. B 57, 289–300.
doi:10.2307/2346101.
171
Berg, L. (1988). Clinical Dementia Rating (CDR). Psychopharmacol. Bull. 24, 637–9.
Available at: http://www.ncbi.nlm.nih.gov/pubmed/3249765 [Accessed July 14,
2018].
Braak, H., and Braak, E. (1991). Neuropathological stageing of Alzheimer-related
changes. Acta Neuropathol. 82, 239–59. Available at:
http://www.ncbi.nlm.nih.gov/pubmed/1759558 [Accessed July 21, 2018].
Braak, H., and Braak, E. (1996). Evolution of the neuropathology of Alzheimer’s disease.
Acta Neurol. Scand. Suppl. 165, 3–12. Available at:
http://www.ncbi.nlm.nih.gov/pubmed/8740983 [Accessed July 21, 2018].
Button, K. S., Ioannidis, J. P. A., Mokrysz, C., Nosek, B. A., Flint, J., Robinson, E. S. J.,
et al. (2013). Power failure: Why small sample size undermines the reliability of
neuroscience. Nat. Rev. Neurosci. 14, 365–376. doi:10.1038/nrn3475.
Cercignani, M., Bammer, R., Sormani, M. P., Fazekas, F., and Filippi, M. (2003). Inter-
sequence and inter-imaging unit variability of diffusion tensor MR imaging histogram-
derived metrics of the brain in healthy volunteers. Am. J. Neuroradiol. 24, 638–643.
Chen, Y., Sha, M., Zhao, X., Ma, J., Ni, H., Gao, W., et al. (2017). Automated detection of
pathologic white matter alterations in Alzheimer’s disease using combined diffusivity
and kurtosis method. Psychiatry Res. Neuroimaging 264, 35–45.
doi:10.1016/J.PSCYCHRESNS.2017.04.004.
Cheng, J.-X., Zhang, H.-Y., Peng, Z.-K., Xu, Y., Tang, H., Wu, J.-T., et al. (2018).
Divergent topological networks in Alzheimer’s disease: A diffusion kurtosis imaging
analysis. Transl. Neurodegener. 7, 10. doi:10.1186/s40035-018-0115-y.
172
Clerx, L., Visser, P. J., Verhey, F., and Aalten, P. (2012). New MRI markers for
Alzheimer’s disease: A meta-analysis of diffusion tensor imaging and a comparison
with Medial temporal lobe measurements. J. Alzheimer’s Dis. 29, 405–429.
doi:10.3233/JAD-2011-110797.
Colgan, N., Siow, B., O’Callaghan, J. M., Harrison, I. F., Wells, J. A., Holmes, H. E., et al.
(2016). Application of neurite orientation dispersion and density imaging (NODDI) to
a tau pathology model of Alzheimer’s disease. Neuroimage 125, 739–744.
doi:10.1016/j.neuroimage.2015.10.043.
Correia, M. M., Carpenter, T. A., and Williams, G. B. (2009). Looking for the optimal DTI
acquisition scheme given a maximum scan time: Are more b-values a waste of time?
Magn. Reson. Imaging 27, 163–175. doi:10.1016/J.MRI.2008.06.011.
Daianu, M., Dennis, E. L., Jahanshad, N., Nir, T. M., Toga, A. W., Jack, C. R., et al.
(2013a). Alzheimer’s disease disrupts rich club organization in brain connectivity
networks. Proc. IEEE Int. Symp. Biomed. Imaging, 266–269.
doi:10.1109/ISBI.2013.6556463.
Daianu, M., Jahanshad, N., Nir, T. M., Toga, A. W., Jack, C. R., Weiner, M. W., et al.
(2013b). Breakdown of brain connectivity between normal aging and Alzheimer’s
disease: A structural k-core network analysis. Brain Connect. 3, 407–22.
doi:10.1089/brain.2012.0137.
Descoteaux, M., Angelino, E., Fitzgibbons, S., and Deriche, R. (2007). Regularized, fast,
and robust analytical Q-ball imaging. Magn. Reson. Med. 58, 497–510.
doi:10.1002/mrm.21277.
173
Descoteaux, M., Deriche, R., Knösche, T., and Anwander, A. (2009). Deterministic and
probabilistic tractography based on complex fiber orientation distributions. IEEE
Trans. Med. Imaging 28, 269–286. doi:10.1109/TMI.2008.2004424.
Ding, L., Zhu, A. H., Saremi, A., Faskowitz, J. I., Håberg, A., Thompson, P. M., et al.
(2018). Voxelwise meta-analysis of brain structural associations with genome-wide
polygenic risk for Alzheimer’s disease. In 14th International Symposium on Medical
Information Processing and Analysis. (Under Review).
Douaud, G., Jbabdi, S., Behrens, T. E. J., Menke, R. A., Gass, A., Monsch, A. U., et al.
(2011). DTI measures in crossing-fibre areas: Increased diffusion anisotropy reveals
early white matter alteration in MCI and mild Alzheimer’s disease. Neuroimage 55,
880–890. doi:10.1016/j.neuroimage.2010.12.008.
Folstein, M. F., Folstein, S. E., and McHugh, P. R. (1975). “Mini-mental state”: A practical
method for grading the cognitive state of patients for the clinician. J. Psychiatr. Res.
12, 189–198. doi:10.1016/0022-3956(75)90026-6.
Fortin, J.-P., Sweeney, E. M., Muschelli, J., Crainiceanu, C. M., and Shinohara, R. T.
(2016). Removing inter-subject technical variability in magnetic resonance imaging
studies. Neuroimage, 198–212. doi:doi:10.1016/j.neuroimage.2016.02.036.
Fortin, J.-P., Parker, D., Tunç, B., Watanabe, T., Elliott, M. A., Ruparel, K., et al. (2017).
Harmonization of multi-site diffusion tensor imaging data. Neuroimage 161, 149–170.
doi:10.1016/J.NEUROIMAGE.2017.08.047.
Frank, R. A., Galasko, D., Hampel, H., Hardy, J., De Leon, M. J., Mehta, P. D., et al.
(2003). Biological markers for therapeutic trials in Alzheimer’s disease: Proceedings
174
of the biological markers working group; NIA initiative on neuroimaging in Alzheimer’s
disease. Neurobiol. Aging 24, 521–536. doi:10.1016/S0197-4580(03)00002-2.
Giannelli, M., Cosottini, M., Michelassi, M. C., Lazzarotti, G., Belmonte, G., Bartolozzi, C.,
et al. (2009). Dependence of brain DTI maps of fractional anisotropy and mean
diffusivity on the number of diffusion weighting directions. J. Appl. Clin. Med. Phys.
11, 2927. Available at: http://www.ncbi.nlm.nih.gov/pubmed/20160677 [Accessed
July 23, 2018].
Giulietti, G., Torso, M., Serra, L., Spanò, B., Marra, C., Caltagirone, C., et al. (2018).
Whole brain white matter histogram analysis of diffusion tensor imaging data detects
microstructural damage in mild cognitive impairment and Alzheimer’s disease
patients. J. Magn. Reson. Imaging. doi:10.1002/jmri.25947.
Grech-Sollars, M., Hales, P. W., Miyazaki, K., Raschke, F., Rodriguez, D., Wilson, M., et
al. (2015). Multi-centre reproducibility of diffusion MRI parameters for clinical
sequences in the brain. NMR Biomed. 28, 468–485. doi:10.1002/nbm.3269.
Grothe, M. J., Barthel, H., Sepulcre, J., Dyrba, M., Sabri, O., and Teipel, S. J. (2017). In
vivo staging of regional amyloid deposition. Neurology 89, 2031–2038.
doi:10.1212/WNL.0000000000004643.
Harms, R. L., Fritz, F. J., Tobisch, A., Goebel, R., and Roebroeck, A. (2017). Robust and
fast nonlinear optimization of diffusion MRI microstructure models. Neuroimage 155,
82–96. doi:10.1016/J.NEUROIMAGE.2017.04.064.
Harsan, L. A., Poulet, P., Guignard, B., Steibel, J., Parizel, N., Loureiro de Sousa, P., et
al. (2006). Brain dysmyelination and recovery assessment by noninvasive in vivo
175
diffusion tensor magnetic resonance imaging. J. Neurosci. Res. 83, 392–402.
doi:10.1002/jnr.
Hu, C., Hua, X., Ying, J., Thompson, P. M., Fakhri, G. E., and Li, Q. (2016). Localizing
sources of brain disease progression with network diffusion model. IEEE J. Sel. Top.
Signal Process. 10, 1214–1225. doi:10.1109/JSTSP.2016.2601695.
Hua, K., Zhang, J., Wakana, S., Jiang, H., Li, X., Reich, D. S., et al. (2008). Tract
probability maps in stereotaxic spaces: Analyses of white matter anatomy and tract-
specific quantification. Neuroimage 39, 336–347.
doi:10.1016/j.neuroimage.2007.07.053.
Hua, X., Lee, S., Yanovsky, I., Leow, A. D., Chou, Y.-Y., Ho, A. J., et al. (2009). Optimizing
power to track brain degeneration in Alzheimer’s disease and mild cognitive
impairment with tensor-based morphometry: An ADNI study of 515 subjects.
Neuroimage 48, 668–681. doi:10.1016/J.NEUROIMAGE.2009.07.011.
Jack, C. R., Barnes, J., Bernstein, M. A., Borowski, B. J., Brewer, J., Clegg, S., et al.
(2015). Magnetic Resonance Imaging in ADNI HHS Public Access. Alzheimer’s
Dement. 11, 740–756. doi:10.1016/j.jalz.2015.05.002.
Jahanshad, N., Kochunov, P. V., Sprooten, E., Mandl, R. C., Nichols, T. E., Almasy, L., et
al. (2013). Multi-site genetic analysis of diffusion images and voxelwise heritability
analysis: A pilot project of the ENIGMA–DTI working group. Neuroimage 81, 455–
469. doi:10.1016/j.neuroimage.2013.04.061.
Jbabdi, S., Behrens, T. E. J., and Smith, S. M. (2010). Crossing fibres in tract-based
spatial statistics. Neuroimage 49, 249–256. doi:10.1016/j.neuroimage.2009.08.039.
176
Jensen, J. H., Helpern, J. A., Ramani, A., Lu, H., and Kaczynski, K. (2005). Diffusional
kurtosis imaging: The quantification of non-Gaussian water diffusion by means of
magnetic resonance imaging. Magn. Reson. Med. 53, 1432–1440.
doi:10.1002/mrm.20508.
Jeurissen, B., Leemans, A., Tournier, J.-D., Jones, D. K., and Sijbers, J. (2013).
Investigating the prevalence of complex fiber configurations in white matter tissue
with diffusion magnetic resonance imaging. Hum. Brain Mapp. 34, 2747–2766.
doi:10.1002/hbm.22099.
Jin, Y., Huang, C., Daianu, M., Zhan, L., Dennis, E. L., Reid, R. I., et al. (2017). 3D tract-
specific local and global analysis of white matter integrity in Alzheimer’s disease.
Hum. Brain Mapp. 38, 1191–1207. doi:10.1002/hbm.23448.
Johnson, W. E., Li, C., and Rabinovic, A. (2007). Adjusting batch effects in microarray
expression data using empirical Bayes methods. Biostatistics 8, 118–127.
doi:10.1093/biostatistics/kxj037.
Kantarci, K., Avula, R., Senjem, M. L., Samikoglu, A. R., Zhang, B., Weigand, S. D., et al.
(2010). Dementia with Lewy bodies and Alzheimer disease: Neurodegenerative
patterns characterized by DTI. Neurology 74, 1814–1821.
doi:10.1212/WNL.0b013e3181e0f7cf.
Kantarci, K., Murray, M. E., Schwarz, C. G., Reid, R. I., Przybelski, S. A., Lesnick, T., et
al. (2017). White-matter integrity on DTI and the pathologic staging of Alzheimer’s
disease. Neurobiol. Aging 56, 172–179.
doi:10.1016/J.NEUROBIOLAGING.2017.04.024.
177
Kauppi, K., Fan, C. C., McEvoy, L. K., Holland, D., Tan, C. H., Chen, C.-H., et al. (2018).
Combining polygenic hazard score with volumetric MRI and cognitive measures
improves prediction of progression from mild cognitive impairment to Alzheimer’s
disease. Front. Neurosci. 12, 260. doi:10.3389/fnins.2018.00260.
Kavcic, V., Ni, H., Zhu, T., Zhong, J., and Duffy, C. J. (2008). White matter integrity linked
to functional impairments in aging and early Alzheimer’s disease. Alzheimer’s
Dement. 4, 381–9. doi:10.1016/j.jalz.2008.07.001.
Kelly, S., Jahanshad, N., Zalesky, A., Kochunov, P., Agartz, I., Alloza, C., et al. (2018).
Widespread white matter microstructural differences in schizophrenia across 4322
individuals: results from the ENIGMA Schizophrenia DTI Working Group. Mol.
Psychiatry 23, 1261–1269. doi:10.1038/mp.2017.170.
Kim, M., Ronen, I., Ugurbil, K., and Kim, D.-S. (2006). Spatial resolution dependence of
DTI tractography in human occipito-callosal region. Neuroimage 32, 1243–1249.
doi:10.1016/J.NEUROIMAGE.2006.06.006.
Kneynsberg, A., Combs, B., Christensen, K., Morfini, G., and Kanaan, N. M. (2017).
Axonal degeneration in tauopathies: Disease relevance and underlying mechanisms.
Front. Neurosci. 11, 572. doi:10.3389/fnins.2017.00572.
Kochunov, P., Jahanshad, N., Sprooten, E., Nichols, T. E., Mandl, R. C., Almasy, L., et
al. (2014). Multi-site study of additive genetic effects on fractional anisotropy of
cerebral white matter: Comparing meta and mega-analytical approaches for data
pooling. Neuroimage 95, 136–150. doi:10.1016/j.neuroimage.2014.03.033.
Kochunov, P., Dickie, E. W., Viviano, J. D., Turner, J., Kingsley, P. B., Jahanshad, N., et
al. (2018). Integration of routine QA data into mega-analysis may improve quality and
178
sensitivity of multisite diffusion tensor imaging studies. Hum. Brain Mapp. 39, 1015–
1023. doi:10.1002/hbm.23900.
Le Bihan, D., and Johansen-Berg, H. (2012). Diffusion MRI at 25: Exploring brain tissue
structure and function. Neuroimage 61, 324–341.
doi:10.1016/j.neuroimage.2011.11.006.
Lee, J.-T., Xu, J., Lee, J.-M., Ku, G., Han, X., Yang, D.-I., et al. (2004). Amyloid-β peptide
induces oligodendrocyte death by activating the neutral sphingomyelinase–ceramide
pathway. J. Cell Biol. 164, 123–131. doi:10.1083/jcb.200307017.
Leow, A. D., Lee, A. D., Klunder, A. D., Lu, A., Becker, J. T., Davis, S. W., et al. (2007).
Statistical properties of Jacobian maps and inverse-consistent deformations in non-
linear image registration. IEEE Trans. Med. Imaging 26, 822–832.
doi:10.1109/TMI.2007.892646.
Leow, A. D., Zhu, S., Zhan, L., McMahon, K., de Zubicaray, G. I., Meredith, M., et al.
(2009a). The tensor distribution function. Magn. Reson. Med. 61, 205–14.
doi:10.1002/mrm.21852.
Leow, A. D., Zhu, S., Zhan, L., McMahon, K., de Zubicaray, G. I., Meredith, M., et al.
(2009b). The tensor distribution function. Magn. Reson. Med. 61, 205–214.
doi:10.1002/mrm.21852.
Liu, M., Zhang, J., Adeli, E., and Shen, D. (2017). Deep multi-task multi-channel learning
for joint classification and regression of brain status. Med. Image Comput. Comput.
Assist. Interv. 10435, 3–11. doi:10.1007/978-3-319-66179-7_1.
179
Maggipinto, T., Bellotti, R., Amoroso, N., Diacono, D., Donvito, G., Lella, E., et al. (2017).
DTI measurements for Alzheimer’s classification. Phys. Med. Biol. 62, 2361–2375.
doi:10.1088/1361-6560/aa5dbe.
Maier-Hein, K. H., Neher, P. F., Houde, J. C., Côté, M. A., Garyfallidis, E., Zhong, J., et
al. (2017). The challenge of mapping the human connectome based on diffusion
tractography. Nat. Commun. 8. doi:10.1038/s41467-017-01285-x.
Mayo, C. D., Mazerolle, E. L., Ritchie, L., Fisk, J. D., and Gawryluk, J. R. (2017).
Longitudinal changes in microstructural white matter metrics in Alzheimer’s disease.
NeuroImage Clin. 13, 330–338. doi:10.1016/j.nicl.2016.12.012.
Mielke, M. M., Kozauer, N. A., Chan, K. C. G., George, M., Toroney, J., Zerrate, M., et al.
(2009). Regionally-specific diffusion tensor imaging in mild cognitive impairment and
Alzheimer’s disease. Neuroimage 46, 47–55.
doi:10.1016/j.neuroimage.2009.01.054.
Mirzaalian, H., Ning, L., Savadjiev, P., Pasternak, O., Bouix, S., Michailovich, O., et al.
(2018). Multi-site harmonization of diffusion MRI data in a registration framework.
Brain Imaging Behav. 12, 284–295. doi:10.1007/s11682-016-9670-y.
Moore, E. E., Hohman, T. J., Badami, F. S., Pechman, K. R., Osborn, K. E., Acosta, L. M.
Y., et al. (2018). Neurofilament relates to white matter microstructure in older adults.
Neurobiol. Aging 70, 233–241. doi:10.1016/j.neurobiolaging.2018.06.023.
Mori, S., Oishi, K., Jiang, H., Jiang, L., Li, X., Akhter, K., et al. (2008). Stereotaxic white
matter atlas based on diffusion tensor imaging in an ICBM template. Neuroimage 40,
570–582. Available at:
180
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2478641/pdf/nihms44189.pdf
[Accessed June 9, 2018].
Mueller, S. G., Weiner, M. W., Thal, L. J., Petersen, R. C., Jack, C. R., Jagust, W., et al.
(2005). Ways toward an early diagnosis in Alzheimer’s disease: the Alzheimer’s
Disease Neuroimaging Initiative (ADNI). Alzheimer’s Dement. 1, 55–66.
doi:10.1016/j.jalz.2005.06.003.
Müller, H.-P., and Kassubek, J. (2013). Diffusion Tensor Magnetic Resonance Imaging in
the Analysis of Neurodegenerative Diseases. J. Vis. Exp. doi:10.3791/50427.
Nir, T. M., Jahanshad, N., Villalon-Reina, J. E., Toga, A. W., Jack, C. R., Weiner, M. W.,
et al. (2013). Effectiveness of regional DTI measures in distinguishing Alzheimer’s
disease, MCI, and normal aging. NeuroImage Clin. 3, 180–195.
doi:10.1016/J.NICL.2013.07.006.
Nir, T. M., Jahanshad, N., Toga, A. W., Jack, C. R., Weiner, M. W., and Thompson, P. M.
(2012). Connectivity Network Breakdown Predicts Imminent Volumetric Atrophy in
Early Mild Cognitive Impairment. Neurobiology of Aging. 36, 41–50. doi:10.1007/978-
3-642-33530-3_4.
Nir, T. M., Jahanshad, N., Villalon-Reina, J. E., Isaev, D., Zavaliangos-Petropulu, A.,
Zhan, L., et al. (2017). Fractional anisotropy derived from the diffusion tensor
distribution function boosts power to detect Alzheimer’s disease deficits. Magn.
Reson. Med. 78, 2322–2333. doi:10.1002/mrm.26623.
O’Dwyer, L., Lamberton, F., Bokde, A. L. W., Ewers, M., Faluyi, Y. O., Tanner, C., et al.
(2011). Multiple indices of diffusion identifies white matter damage in mild cognitive
181
impairment and Alzheimer’s disease. PLoS One 6, e21745.
doi:10.1371/journal.pone.0021745.
Oishi, K., Mielke, M. M., Albert, M., Lyketsos, C. G., and Mori, S. (2011). DTI analyses
and clinical applications in Alzheimer’s disease. J. Alzheimers. Dis. 26 Suppl 3, 287–
96. doi:10.3233/JAD-2011-0007.
Palacios, E. M., Martin, A. J., Boss, M. A., Ezekiel, F., Chang, Y. S., Yuh, E. L., et al.
(2017). Toward precision and reproducibility of diffusion tensor imaging: A multicenter
diffusion phantom and traveling volunteer study. Am. J. Neuroradiol. 38, 537–545.
doi:10.3174/ajnr.A5025.
Parker, T. D., Slattery, C. F., Zhang, J., Nicholas, J. M., Paterson, R. W., Foulkes, A. J.
M., et al. (2018). Cortical microstructure in young onset Alzheimer’s disease using
neurite orientation dispersion and density imaging. Hum. Brain Mapp. 39, 3005–
3017. doi:10.1002/hbm.24056.
Phillips, J. S., Das, S. R., McMillan, C. T., Irwin, D. J., Roll, E. E., Da Re, F., et al. (2018).
Tau PET imaging predicts cognition in atypical variants of Alzheimer’s disease. Hum.
Brain Mapp. 39, 691–708. doi:10.1002/hbm.23874.
Pohl, K. M., Sullivan, E. V., Rohlfing, T., Chu, W., Kwon, D., Nichols, B. N., et al. (2016).
Harmonizing DTI measurements across scanners to examine the development of
white matter microstructure in 803 adolescents of the NCANDA study. Neuroimage
130, 194–213. doi:10.1016/j.neuroimage.2016.01.061.
Popuri, K., Balachandar, R., Alpert, K., Lu, D., Bhalla, M., Mackenzie, I. R., et al. (2018).
Development and validation of a novel dementia of Alzheimer’s type (DAT) score
182
based on metabolism FDG-PET imaging. NeuroImage. Clin. 18, 802–813.
doi:10.1016/j.nicl.2018.03.007.
Powell, F., Tosun, D., Sadeghi, R., Weiner, M., Raj, A., and Initiative, for the A. D. N.
(2018). Preserved structural network organization mediates pathology spread in
Alzheimer’s disease spectrum despite loss of white matter tract integrity. J.
Alzheimer’s Dis. Preprint, 1–18. doi:10.3233/JAD-170798.
Prasad, G., Nir, T. M., Toga, A. W., and Thompson, P. M. (2013). Tractography density
and network measures in Alzheimer’s disease. Proceedings. IEEE Int. Symp.
Biomed. Imaging 2013, 692–695. doi:10.1109/ISBI.2013.6556569.
Rosen, W. G., Mohs, R. C., and Davis, K. L. (1984). A new rating scale for Alzheimer’s
disease. Am. J. Psychiatry 141, 1356–1364. doi:10.1176/ajp.141.11.1356.
Rosenthal, R., and Rosnow, R. L. (1991). Essentials of behavioral research: methods and
data analysis. McGraw-Hill Available at:
https://books.google.com/books/about/Essentials_of_Behavioral_Research.html?id
=aPEMAQAAMAAJ [Accessed July 11, 2018].
Roth, A. D., Ramírez, G., Alarcón, R., and Von Bernhardi, R. (2005). Oligodendrocytes
damage in Alzheimer’s disease: Beta amyloid toxicity and inflammation. Biol. Res.
38, 381–387. doi:10.4067/S0716-97602005000400011.
Sachdev, P. S., Zhuang, L., Braidy, N., and Wen, W. (2013). Is Alzheimer’s a disease of
the white matter? Curr. Opin. Psychiatry 26, 244–251.
Sanchez-Rodriguez, L. M., Iturria-Medina, Y., Baines, E. A., Mallo, S. C., Dousty, M.,
Sotero, R. C., et al. (2018). Design of optimal nonlinear network controllers for
183
Alzheimer’s disease. PLoS Comput. Biol. 14, e1006136.
doi:10.1371/journal.pcbi.1006136.
Shaw, L. M., Korecka, M., Clark, C. M., Lee, V. M.-Y., and Trojanowski, J. Q. (2007).
Biomarkers of neurodegeneration for diagnosis and monitoring therapeutics. Nat.
Rev. Drug Discov. 6, 295–303. doi:10.1038/nrd2176.
Slattery, C. F., Zhang, J., Paterson, R. W., Foulkes, A. J. M., Carton, A., Macpherson, K.,
et al. (2017). ApoE influences regional white-matter axonal density loss in
Alzheimer’s disease. Neurobiol. Aging 57, 8–17.
doi:10.1016/j.neurobiolaging.2017.04.021.
Song, S.-K., Sun, S.-W., Ju, W.-K., Lin, S.-J., Cross, A. H., and Neufeld, A. H. (2003).
Diffusion tensor imaging detects and differentiates axon and myelin degeneration in
mouse optic nerve after retinal ischemia. Neuroimage 20, 1714–22. Available at:
http://www.ncbi.nlm.nih.gov/pubmed/14642481 [Accessed July 21, 2018].
Song, S.-K., Yoshino, J., Le, T. Q., Lin, S.-J., Sun, S.-W., Cross, A. H., et al. (2005).
Demyelination increases radial diffusivity in corpus callosum of mouse brain.
Neuroimage 26, 132–140. doi:10.1016/j.neuroimage.2005.01.028.
Sulaimany, S., Khansari, M., Zarrineh, P., Daianu, M., Jahanshad, N., Thompson, P. M.,
et al. (2017). Predicting brain network changes in Alzheimer’s disease with link
prediction algorithms. Mol. Biosyst. 13, 725–735. doi:10.1039/C6MB00815A.
Tax, C. M. W., Grussu, F., Kaden, E., Ning, L., Rudrapatna, U., Evans, J., et al. (2018).
Cross-vendor and Cross-protocol harmonisation of diffusion MRI data: A comparative
study. Proc. Jt. Annu. Meet. ISMRM-ESMRMB, 471. Available at:
http://indexsmart.mirasmart.com/ISMRM2018/PDFfiles/0471.html.
184
Thompson, P. M., Stein, J. L., Medland, S. E., Hibar, D. P., Vasquez, A. A., Renteria, M.
E., et al. (2014). The ENIGMA Consortium: large-scale collaborative analyses of
neuroimaging and genetic data. Brain Imaging Behav. 8, 153–82.
doi:10.1007/s11682-013-9269-5.
Tournier, J.-D., Calamante, F., Gadian, D. G., and Connelly, A. (2004). Direct estimation
of the fiber orientation density function from diffusion-weighted MRI data using
spherical deconvolution. Neuroimage 23, 1176–1185.
doi:10.1016/J.NEUROIMAGE.2004.07.037.
Tuch, D. S., Reese, T. G., Wiegell, M. R., Makris, N., Belliveau, J. W., and Wedeen, V. J.
(2002). High angular resolution diffusion imaging reveals intravoxel white matter fiber
heterogeneity. Magn. Reson. Med. 48, 577–582. doi:10.1002/mrm.10268.
Veitch, D. P., Weiner, M. W., Aisen, P. S., Beckett, L. A., Cairns, N. J., Green, R. C., et
al. (2018). Understanding disease progression and improving Alzheimer’s disease
clinical trials: Recent highlights from the Alzheimer’s Disease Neuroimaging Initiative.
Alzheimer’s Dement., 1–46. doi:10.1016/j.jalz.2018.08.005.
Wang, B., Niu, Y., Miao, L., Cao, R., Yan, P., Guo, H., et al. (2017). Decreased complexity
in Alzheimer’s disease: Resting-state fMRI evidence of brain entropy mapping. Front.
Aging Neurosci. 9, 378. doi:10.3389/fnagi.2017.00378.
Wang, M.-L., Wei, X.-E., Fu, J.-L., Li, W., Yu, M.-M., Li, P.-Y., et al. (2018). Subcortical
nuclei in Alzheimer’s disease: A volumetric and diffusion kurtosis imaging study. Acta
Radiol. 59, 1365–1371. doi:10.1177/0284185118758122.
185
Wang, X., Zhen, X., Li, Q., Shen, D., and Huang, H. (2018). Cognitive assessment
prediction in Alzheimer’s disease by multi-layer multi-target regression.
Neuroinformatics, 1–10. doi:10.1007/s12021-018-9381-1.
Zavaliangos-Petropulu, A., Nir, T. M., Thomopoulos, S. I., Jahanshad, N., Reid, R. I.,
Bernstein, M. A., et al. (2018). Ranking Diffusion Tensor Measures of Brain Aging &
Alzheimer’s Disease. Proc. SPIE Int. Symp. Med. Inf. Process. Anal.
Zhan, L., Leow, A. D., Jahanshad, N., Chiang, M.-C., Barysheva, M., Lee, A. D., et al.
(2010). How does angular resolution affect diffusion imaging measures? Neuroimage
49, 1357–71. doi:10.1016/j.neuroimage.2009.09.057.
Zhan, L., Jahanshad, N., Ennis, D. B., Jin, Y., Bernstein, M. A., Borowski, B. J., et al.
(2013). Angular versus spatial resolution trade-offs for diffusion imaging under time
constraints. Hum. Brain Mapp. 34, 2688–2706. doi:10.1002/hbm.22094.
Zhang, H., Schneider, T., Wheeler-Kingshott, C. A., and Alexander, D. C. (2012). NODDI:
Practical in vivo neurite orientation dispersion and density imaging of the human
brain. Neuroimage 61, 1000–1016. doi:10.1016/J.NEUROIMAGE.2012.03.072.
Zhou, T., Thung, K.-H., Zhu, X., and Shen, D. (2017). Feature learning and fusion of
multimodality neuroimaging and genetic data for multi-status dementia diagnosis.
Mach. Learn. Med. imaging. MLMI 10541, 132–140. doi:10.1007/978-3-319-67389-
9_16.
Zhu, A. H., Moyer, D. C., Nir, T. M., Thompson, P. M., and Jahanshad, N. (2018).
Challenges and opportunities in diffusion MRI data harmonization. In: Computational
Diffusion MRI and Brain Connectivity. Mathematics and Visualization. Springer,
Cham.
186
Zhu, T., Liu, X., Gaugh, M. D., Connelly, P. R., Ni, H., Ekholm, S., et al. (2009). Evaluation
of measurement uncertainties in human diffusion tensor imaging (DTI)-derived
parameters and optimization of clinical DTI protocols with a wild bootstrap analysis.
J. Magn. Reson. Imaging 29, 422–435. doi:10.1002/jmri.21647.
Zhu, T., Hu, R., Qiu, X., Taylor, M., Tso, Y., Yiannoutsos, C., et al. (2011).
Quantification of accuracy and precision of multi-center DTI measurements: A
diffusion phantom and human brain study. Neuroimage 56, 1398–1411.
doi:10.1016/J.NEUROIMAGE.2011.02.010.
The Supplementary Material for this article can be found online
at: https://www.frontiersin.org/articles/10.3389/fninf.2019.00002/full#supplementary-
material
Abstract (if available)
Abstract
The decrease in stroke mortality has led to a growing population of people in need of rehabilitation. To help clinicians and caregivers make informed decisions about rehabilitation planning, there is a critical need to identify reliable biomarkers of stroke outcomes that help make accurate predictions of a patient’s potential to recovery. However, the generalizability of stroke recovery biomarkers must be tested in diverse patient populations to improve the probability of successful use in the clinic. This has led to an increasing interest in multi-site research consortia to obtain large, diverse patient samples. One such consortium is ENIGMA Stroke Recovery. ENIGMA Stroke Recovery is an international multi-site collaboration of stroke researchers dedicated to providing a reliable infrastructure for the collection and analyses of large, diverse datasets of poststroke brain magnetic resonance imaging (MRI) and behavioral measures that can be used to identify robust, reproducible biomarkers of stroke outcomes. In this dissertation, I discuss my contributions to ENIGMA Stroke Recovery and walk through the curation of MRI data for the hippocampusㅡa brain region particularly vulnerable to stroke related secondary degeneration and thought to be a promising biomarker of stroke outcomes. The overarching goal of this work was to set the foundation for future large-scale research of post-stroke hippocampus by using ENIGMA Stroke Recovery data to 1) identify a robust automated hippocampal segmentation method, 2) investigate the association between lesion size and post-stroke hippocampal volume, and 3) explore the association between post-stroke sensorimotor impairment and hippocampal damage.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Neuroimaging markers of risk & resilience to brain aging and dementia
PDF
Relationship between brain structure and motor behavior in chronic stroke survivors
PDF
Neuroimaging in complex polygenic disorders
PDF
Feasibility, acceptability, and implementation context of a complex telerehabilitation intervention for post-stroke upper extremity recovery
PDF
Development and implementation of a modular muscle-computer interface for personalized motor rehabilitation after stroke
PDF
Novel multi-site brain imaging approaches to map HIV-related neuropathology
PDF
The effects of fast walking, biofeedback, and cognitive impairment on post-stroke gait
PDF
Arm choice post-stroke
PDF
Engagement of the action observation network through functional magnetic resonance imaging with implications for stroke rehabilitation
PDF
Deficits and rehabilitation of upper-extremity multi-joint movements in individuals with chronic stroke
PDF
Vascular contributions to brain aging along the Alzheimer's disease continuum
PDF
Locomotor skill learning in virtual reality in healthy adults and people with Parkinson disease
PDF
Precision care to optimize safe return to function following surgical flexor tendon repair
PDF
Reproducibility and management of big data in brain MRI studies
PDF
Hemisphere-specific deficits in the control of bimanual movements after stroke
PDF
The neuroimaging of cancer and chemotherapy
PDF
Anemia and white matter: a diffusion MRI analysis of the human brain
PDF
Computational model of stroke therapy and long term recovery
PDF
Decoding the neurological and genetic underpinnings of chronic pain
PDF
Virtual surgeries as a tool for studying motor learning
Asset Metadata
Creator
Zavaliangos-Petropulu, Artemis
(author)
Core Title
A multi-site neuroimaging approach to studying hippocampal damage in chronic stroke
School
Keck School of Medicine
Degree
Doctor of Philosophy
Degree Program
Neuroscience
Degree Conferral Date
2021-12
Publication Date
11/30/2021
Defense Date
09/23/2021
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
hippocampus,MRI,neuroimaging,OAI-PMH Harvest,sensorimotor impairment,stroke
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Bradley, Nina (
committee chair
), Liew, Sook-Lei (
committee member
), Schweighofer, Nicolas (
committee member
), Thompson, Paul M. (
committee member
)
Creator Email
azavaliangos@gmail.com,zavalian@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-oUC17789615
Unique identifier
UC17789615
Legacy Identifier
etd-Zavaliango-10266
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Zavaliangos-Petropulu, Artemis
Type
texts
Source
20211206-wayne-usctheses-batch-901-nissen
(batch),
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Repository Email
cisadmin@lib.usc.edu
Tags
hippocampus
MRI
neuroimaging
sensorimotor impairment
stroke