University of Southern California Dissertations and Theses

Relationship between brain structure and motor behavior in chronic stroke survivors

(USC Thesis Other)

Relationship between brain structure and motor behavior in chronic stroke survivors

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content RELATIONSHIP BETWEEN BRAIN STRUCTURE AND MOTOR BEHAVIOR IN
CHRONIC STROKE SURVIVORS

by

Bokkyu Kim

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(BIOKINESIOLOGY)

December 2017

Copyright 2017 Bokkyu Kim
ii

ACKNOWLEDGEMENTS

I would like to thank all the people who have contributed to my dissertation. I would
never have been able to finish my dissertation without the guidance of my Dissertation
Committee members, help from colleagues and friends, and support from my family.
I would like to express the deepest appreciation to my advisor, Dr. Carolee Winstein, for
her invaluable guidance, tolerance, and providing an exceptional environment for conducting
research. She always pushed me toward the limit of my capability, and taught and encouraged
me to challenge my limit and overcome it. Her scientific insights into neuroscience and
neurorehabilitation research were always facilitated my curiosity on the research matters. Every
conversation with her in regular meetings was full of fun and didactic, but not easy to defend
from her keen questions. She makes me feel that I am standing on the shoulders of giants to see
further and to look away the entire forest, not just a tree.
I would like to specially thank Dr. Gordon, who always challenged my scientific logical
thinking. His feedback on my dissertation works was critical and his criticism and skepticism on
my presentations were helpful to improve the logical connections in the dissertation. Although
sometimes it is extremely hard to defend from Dr. Gordon’s critical questions in seminars, these
were essential questions that I must be able to answer. This process evoked more scientific
inquiries, and bolstered my dissertation.
I would like to acknowledge Dr. Beth Fisher, who is a great mentor and inspiring
researcher. She helped me a lot to improve my writing, and every meeting with her to review my
writing was so useful to improve my dissertation quality. Further, her positive feedback on my
research always encouraged me a lot, and motivated me to more concentrate on my research.
iii

I would like to express my appreciation to Dr. Schweighofer, who helped me to improve
my Matlab programming and statistical analysis skills. I remember his modeling course was one
of the toughest courses that I have taken. Through this course, I was able to improve my Matlab
programming skills, and these skills were critical to conduct data analyses for this dissertation.
I would like to express gratitude to Dr. Richard Leahy and his research team for
developing an elaborate software to analyze brain structural imaging data and for teaching and
guiding me how to analyze brain imaging data. I believe I cannot complete my data analysis
without BrainSuite software, and I really appreciate that his research team’s support for my data
analysis, and their help to expand my knowledge on the neuroimaging data analysis.
My dissertation studies were interdisciplinary works. I used clinical and neuroimaging
data from a phase-I clinical trial conducted in Motor Behavior and Neurorehabilitation
Laboratory at the USC. I would like to thank all people conducting this clinical trial and stroke
patients who participated in this study. Further, I would like to acknowledge all collaborators of
my dissertation studies: Dr. Leahy’s research team (especially Dr. Justin Haldar and Soyoung
Choi) for providing technical support and research backgrounds on diffusion tensor imaging data
analysis; Dr. Sook-Lei Liew’s research team for providing the stroke lesion mask data; and Dr.
Brent Liu’s research team for working on the lateral ventricle mask data.
I would like to thank all faculty and staff in the Division of Biokinesiology and Physical
Therapy at the USC. Their support made me complete this dissertation. The faculty always
provided essential feedbacks on my presentations in Division Seminar and Neurorehabilitation
Seminar, and their feedback enhanced the quality of my dissertation. The staff were always
supportive and always listened to me. I will never forget the support from all the faculty and
staff.
iv

I would also like to express gratitude to all MBNL/NAIL lab mates – Alice, Yi-An, Yu-
Chen, Sujin, Hannah, Dorsa, Helen, Andrew, Irene, Alaa, Rini, and Alex. Their positive thinking
and energy always cheered me up. I enjoyed every conversation with them regarding research
and personal life. They were always supportive to complete my dissertation, and I could not have
done my dissertation without their support and help.
Finally, I would like to express appreciation to my family. My family always supported
me to complete my journey for the Ph.D. degree. My grandmother, Soon-Rye Choi, was the
person who motivated me to pursue stroke rehabilitation research. She was a stroke survivor who
had two incidences of ischemic stroke – one in left, and the other one in right hemisphere. After I
witnessed her tough time with post-stroke disability, I decided to go to a graduate school to help
people with neurological disorders like her. I used to help her to improve her gait and ADL
before I came here, and I remember she always enjoyed her gait exercise with me. Unfortunately,
she passed away in my second year at USC. As it was a middle of the semester, I was unable to
join her funeral. I regret that I cannot be with her in her last moment. My dad, Eung-Oh Kim,
passed away in the beginning of my third year (September 2014). I remember his last call to me
few days before he left. I regret that I cannot be with him when he left and I would have said I
love him so much when he called me last time. When I was young he always read books for me
and taught me mathematics. He always motivated me to pursue higher education to become a
better person. I have so many good memories with him, and I miss him so much. Although my
grandma and daddy are not physically here, I believe their spirits are with me, and protects our
family and helped me to complete this dissertation. I believe they will proud of me in the heaven.
My mom, Jae-Chun Yu, devoted entire her life for our family. This dissertation cannot be
completed without her support. She always encouraged me to pursue my dream, and she is
v

always supportive of my decision. Also, she always emphasized the importance of education,
and at the same time, she always highlights that I must be a person who has a warm heart to help
others, not just has a smart brain. I always keep her dictation in mind, and I am trying to be a
good person who can help others, and I believe my research will contribute to help others
suffering from the post-stroke disability. My brother, Bongkyu Kim, and sister-in-law supported
me by taking care of all family stuff in South Korea during I am absent. I really appreciate for
their patience. My wife, Gyeonghye Shin, should be appreciated for giving me bountiful
supports. She gave up so many things, including her career as an occupational therapist in South
Korea, to accompany with me in my Ph.D. journey. I know her sacrifices made my dissertation
possible. Sometimes we had some argument due to my stubbornness, but I have no doubt on the
fact that she is always right. I love you so much, and I will pay back for your sacrifices. My little
one, Lucas Taeyang Kim, was a surprise new year gift to our couple for 2016. It was not easy to
deal with a new-born baby and my dissertation, but it was a full of fun and joy to taking care of a
baby and writing the dissertation at the same time. I want him to read this dissertation when he
grows up, and he will be happy with that he was a part of his daddy’s Ph.D. journey.

vi

TABLE OF CONTENTS

ACKNOWLEDGEMENTS ........................................................................................................ II
LIST OF FIGURES .................................................................................................................. VII
ABSTRACT ................................................................................................................................. IX
CHAPTER ONE: BACKGROUND AND OVERVIEW .......................................................... 1
CHAPTER TWO: CAN NEUROLOGICAL BIOMARKERS OF
BRAIN IMPAIRMENT BE USED TO PREDICT POST-STROKE MOTOR
RECOVERY? A SYSTEMATIC REVIEW ............................................................................. 10
CHAPTER THREE: A COMPARISON OF SEVEN DIFFERENT
DTI-DERIVED ESTIMATES OF CORTICOSPINAL TRACT STRUCTURAL
CHARACTERISTICS IN CHRONIC STROKE SURVIVORS ............................................ 55
CHAPTER FOUR: CAN A DTI-DERIVED CST BIOMARKER BE USED
TO PREDICT MOTOR IMPROVEMENT IN CHRONIC STROKE SURVIVORS? ....... 83
CHAPTER FIVE: SUMMARY AND GENERAL DISCUSSION ....................................... 117
APPENDIX A. EVIDENCE METHODOLOGICAL STRENGTH EVALUATION ........ 126
APPENDIX B. MOST COMMON PREDICTOR VARIABLES OF EACH
NEUROLOGICAL BIOMARKER TYPE ............................................................................. 130
APPENDIX C. INCLUSION AND EXCLUSION CRITERIA FOR DOSE
RANDOMIZED CONTROLLED TRIAL ............................................................................. 136
APPENDIX D. HISTOGRAM OF BEST LASSO LINEAR REGRESSION
MODELS FROM BOOTSTRAP SAMPLES ........................................................................ 137
REFERENCES .......................................................................................................................... 138

vii

LIST OF FIGURES
Figure 1-1. Illustration of specific aims .......................................................................................... 7
Figure 2-1. Evidence search strategy diagram. ............................................................................. 19
Figure 2-2. Trend in number of publications per year. ................................................................. 24
Figure 2-3. Frequency of each neurological biomarker type for each stroke phase. .................... 26
Figure 2-4. Summary of motor endpoint (dependent) measures. ................................................. 28
Figure 2-5. Distribution of evidence methodological quality scores. ........................................... 33
Figure 2-6. Comparison of statistical power and effect size between models using
neurological biomarkers alone and models using neurological biomarkers and
clinical measures. ................................................................................................................... 37
Figure 2-7. Forest plot for comparing effect sizes between different model groups. ................... 38
Figure 3-1. Visualization of a reconstructed CST tractography from a single participant. .......... 66
Figure 3-2. Difference in FA between ipsi- and contra-lesional CSTs from different
methods. ................................................................................................................................. 71
Figure 3-3. Comparison of FA between ipsi- and contra-lesional CST in axial slices.. ............... 72
Figure 3-4. Comparison of FA asymmetry from different methods. Red horizontal dash
line indicates the normative range of CST FA asymmetry in this cohort of chronic
stroke survivors with mild to moderate motor impairment. .................................................. 74
Figure 3-5. Partial correlation between individual tractography-based CST FA asymmetry
and mWMFT time score, controlled for age and chronicity. ................................................ 75
Figure 3-6. 3-D Template CST VOI and 3-D individual CST VOI from tractography
on the T1-weighted image. .................................................................................................... 79
Figure 4-1. Difference in WMFT-distal between two time-points. .............................................. 95
Figure 4-2. The clinically important difference (CID) in WMFT.. .............................................. 96
Figure 4-3. Linear regression between baseline CST FA asymmetry and Δ WMFT-distal. ........ 97
Figure 4-4. Cross-validated mean square error of lasso fit. .......................................................... 98
Figure 4-5. Regression between actual Δ WMFT-distal and predicted Δ WMFT-distal
from the lasso regression model. ........................................................................................... 99
Figure 4-6 Diagnostics of lasso regression model. ..................................................................... 100
viii

Figure 4-7. Logistic regression between baseline CST FA asymmetry and CID in
WMFT-total. ........................................................................................................................ 101
Figure 4-8. An independent t-test of baseline CST FA asymmetry between participants
with CID and without CID. ................................................................................................. 102
Figure 4-9. Cross-validated deviance of logistic lasso fit. .......................................................... 103
Figure 4-10. Logistic regression fit between the predicted CID from lasso logistic
regression and actual CID in WMFT. ................................................................................. 104
Figure 4-11. 2×2 repeated measures ANOVA on CST FA. Blue line indicates group
mean change in FA of the contralesional CST, and red dash line indicates group
mean change in FA of the ipsilesional CST. ....................................................................... 106
Figure 4-12. 2×2 repeated measures ANOVA on lateral ventricle volumes. ............................. 107
Figure 4-13. Linear regression between ΔCST FA asymmetry and ΔWMFT-distal. ................ 109
Figure 4-14. A hypothesized inverted “U” shape relationship between CST FA
asymmetry and motor improvement in chronic stroke survivors. ....................................... 111

ix

ABSTRACT
Prediction of functional recovery after stroke is crucial to improving the efficacy of rehabilitation
for stroke survivors. Previous studies reported that a Diffusion tensor imaging (DTI)-derived
corticospinal tract (CST) biomarker is predictive of motor recovery early after stroke. However,
little is known about whether or not a DTI-derived CST biomarker is an essential predictor of
motor improvement in chronic stroke survivors. This dissertation aimed to examine whether a
DTI-derived CST biomarker can be used to predict motor improvement in chronic stroke
survivors with mild-to-moderate motor impairment. Further, we examined whether subcortical
white matter structural changes over a 3-month period are measurable using research quality
diffusion images. Finally, we sought to determine if there was a relationship between CST brain
structural changes and improvement in motor behavior in this cohort of chronic stroke survivors.
Our results showed that the CST biomarker can be used to predict motor improvement over a 3-
month period. Further, a cross-validated multimodal model that included CST fractional
anisotropy (FA) asymmetry, upper extremity Fugl-Meyer impairment score, and age more
accurately predicted improvement in motor performance than a simple linear regression model.
However, there was no detectable change in subcortical white matter structure of either the
ipsilesional or contralesional CST. In spite of the lack of measurable subcortical white matter
structural changes, there was a small but significant improvement in motor behavior as measured
by the distal component of the Wolf Motor Function test (WMFT-Distal). These findings provide
strong evidence that a DTI-derived CST biomarker is a significant predictor of improvement in
motor behavior in chronic stroke survivors with mild-to-moderate motor impairment. Future
research is necessary to develop gold standard methods to quantify both structural and functional
brain changes for use in predicting functional recovery in stroke survivors.
Neurological Biomarkers of Post-Stroke Motor Recovery 1

CHAPTER ONE: BACKGROUND AND OVERVIEW

Background
Precise prognosis of motor recovery after stroke is a key to improve the efficacy of stroke
rehabilitation (Stinear et al., 2017). Attempts have been made using advanced neurological
measures in conjunction with standard clinical outcome measures to predict motor recovery. The
neurological measures are including structural and function magnetic resonance imaging (MRI),
diffusion tensor imaging (DTI), and transcranial magnetic stimulation (TMS). Accurate
prediction of post-stroke motor recovery allows clinicians to establish achievable goals and to
determine the optimal therapeutic approach for each individual patient following stroke (Heiss
and Kidwell, 2014; Stinear and Stinear, 2010). Diffusion tensor imaging (DTI has factored into
this determination greatly by identifying the influence that the brain structural characteristics
have on motor behavior (Assaf and Pasternak, 2008; Gregg et al., 2007; Scholz et al., 2009).
DTI-derived metrics of the cortico-spinal tracts (CST) are associated with upper limb motor
behavior in both acute and chronic stroke (Lindenberg et al., 2010; Park et al., 2013; Puig et al.,
2017, 2010). I operationally defined cortico-spinal tract (CST) as the descending motor pathways
including the direct cortico-spinal pathway (i.e., medial and lateral cortico-spinal tracts) as well
as the indirect pathways (i.e., cortico-rubro-spinal and cortico-reticulo-spinal pathways and etc).
Fractional anisotropy (FA) is a representative DTI-derived variable, defined as a
normalized measure of the directionality of water molecules surrounding axons and ranges from
0 to 1 (Soares et al., 2013). It has been widely used to quantify the microstructural characteristics
of the CST in the lesioned hemisphere (i.e., ipsi-lesional) of individuals with stroke (Lindenberg
et al., 2010; Park et al., 2013; Puig et al., 2010). In general, FA of damaged or demyelinated
Neurological Biomarkers of Post-Stroke Motor Recovery 2

axons is lower than intact axons in the central nervous system (CNS) (Concha et al., 2006; Gregg
et al., 2007; Groisser et al., 2014; Gupta et al., 2006; Han et al., 2009; Lindenberg et al., 2012;
Soares et al., 2013). Specifically, a decrease in FA of CST in the lesioned hemisphere after
stroke has been demonstrated in both animal and human studies (Lindenberg et al., 2012; Puig et
al., 2010; Yu et al., 2009). FA of ipsilesional CST indicates the degree of demyelination and/or
Wallerian degeneration of axons (Lindenberg et al., 2012; Puig et al., 2010; Yu et al., 2009).
A number of cross-sectional and longitudinal clinical studies have shown a strong
relationship between FA of the ipsilesional CST and upper extremity (UE) motor deficits during
the acute, sub-acute (Puig et al., 2013, 2011, 2010; Stinear et al., 2012; Yoshioka et al., 2008)
and chronic stages of stroke (Lindenberg et al., 2012, 2010, 2009; Qiu et al., 2011; Song et al.,
2012; Stinear et al., 2007). As motor commands are delivered through descending motor
pathways from sensorimotor cortices to the spinal cord (Lindenberg et al., 2012; Puig et al.,
2010), axonal damage or demyelination disturbs delivery of motor commands to the contralateral
arm muscles thus resulting in deficits in motor skill performance (Lindenberg et al., 2012). As
such, a lesion that disrupts the CST is the major determinant of motor impairment (Corbetta et
al., 2015).
Further, A number of longitudinal prognostic studies have shown that the DTI-derived
metrics of CST are significant predictors of motor recovery and response to therapy for both
acute and chronic stroke (Borich et al., 2014; Cho et al., 2007a; Jang et al., 2005; Lindenberg et
al., 2010; Puig et al., 2017, 2013, Qiu et al., 2011, 2008; Stinear et al., 2007; Yoshioka et al.,
2008). In general, people with greater FA value of the ipsilesional CST at 2 weeks after stroke
onset showed better long-term UE motor recovery during the acute phase (Puig et al., 2011,
Neurological Biomarkers of Post-Stroke Motor Recovery 3

2010) and greater response to therapeutic interventions during the chronic phase (Lindenberg et
al., 2012; Stinear et al., 2007).
As a biomarker of brain impairment, DTI measures of brain structure also capture
learning-dependent brain white matter structural changes, such as axonal remodeling (May,
2011; Puig et al., 2017). The axonal remodeling includes increases in axon diameter, myelin
thickness, and the number of myelinated axons. The white matter structural changes may
contribute to motor skill learning by optimizing the speed or synchrony of brain signal delivery
(Zatorre et al., 2012). A number of motor skill learning studies with non-disabled young adults
have shown brain white matter remodeling after motor skill practice captured by DTI and its
relationship to motor skill learning (Dayan and Cohen, 2011; Fields, 2008; Scholz et al., 2009).
Further, stroke animal studies have demonstrated that the white matter remodeling of
sensorimotor pathways is crucial to functional recovery after stroke (Gregg et al., 2007; Quallo et
al., 2009; Ramos-Cejudo et al., 2015; Sato et al., 2009; Schlaug et al., 2009; Wan et al., 2014).

Gaps in current research literature

Limited evidence for DTI biomarkers for individuals with mild-to-moderate motor
impairment in the chronic stage. A significant number of diffusion MRI research has been
studied to determine the relationship between DTI biomarkers and motor behavior following
stroke. However, these studies have focused on the acute phase of stroke due to the time course
of spontaneous neurological recovery (Cho et al., 2007a; Di Lazzaro et al., 2010; Puig et al.,
2011). Only a few studies have shown a significant relationship between brain structure and
motor behavior during the chronic phase (Lindenberg et al., 2012; Zhu et al., 2010). Of specific
Neurological Biomarkers of Post-Stroke Motor Recovery 4

importance is the fact that only a few studies have used DTI biomarkers to predict improvement
in motor behavior or responsiveness to intervention in chronic stroke survivors (Lindenberg et
al., 2012). Again, the majority of studies employed DTI biomarkers to predict motor recovery in
the acute phase (Kim and Winstein, 2017). For chronic stroke, studies have primarily used
multimodal neurological biomarkers in their predictive model, and it has not been well
established if the DTI biomarker can be used alone to predict motor improvement. Importantly,
the DTI biomarker has been more frequently used to predict motor recovery in the moderate to
severe stroke population (Stinear et al., 2007; Stinear et al., 2012). Thus, we need to determine if
the DTI biomarker can be used for prediction of UE motor improvement in chronic stroke
survivors with mild-to-moderate impairment. Given that mild to moderate stroke survivors
represent the population mostly studied in large scale phase III clinical trials, the feasibility of
using the DTI biomarker to estimate improvement in motor behavior should be tested.

No standard method to estimate CST microstructure using DTI. There are three most
common methods to estimate CST structural characteristics using DTI: i) 2-dimensional region
of interest (2-D ROI)-based method; ii) 3-dimensional (3-D) tractography-based method, and iii)
3-D CST template-based method. In a previous study, CST metrics derived from each of these
methods equally correlated with clinical motor scores in chronic stroke survivors (Park et al.,
2013). However, only one research with small sample size was studied to determine the most
accurate DTI-derived estimate of CST microstructure in chronic stroke survivors. As such,
determination of the estimation method that could best correlate with motor outcomes in chronic
stroke survivors with mild-to-moderate motor impairment was not possible. Thus, we need more
Neurological Biomarkers of Post-Stroke Motor Recovery 5

research to determine the DTI-derived estimate of CST structure that best correlates with motor
behavior in a more homogeneous stroke population.

Lack of studies using DTI with research-quality resolution in chronic stroke.
Further, previous studies employed low resolution diffusion MRI scans which lack the spatial
and angular resolution for modeling of brain white matter fibers (Lindenberg et al., 2012). For
instance, diffusion tensor model-based tractography reconstruction cannot distinguish between
white matter fibers that cross at different angles within a single voxel (Jones et al., 2013; Tuch,
2004; Tuch et al., 2002). This may lead to an inaccurate quantification of the CST.(Jones et al.,
2013) Recent studies have shown that either a High Angular Resolution Diffusion Image
(HARDI)(Tuch, 2004; Tuch et al., 2002), Diffusion Spectrum Image (DSI) (Bassett et al., 2011;
Tuch, 2004; Wedeen et al., 2008), or Diffusion Kurtosis Imaging (DKI) (Zhang et al., 2017) can
be used to identify the white matter fibers better in the areas where these fibers are crossing.
Therefore, use of higher resolution diffusion MRI would be more accurate than low resolution
clinical DTI to quantify the structural impairment of the stroke brain.

Lack of clinical evidence to support a relationship between CST microstructural
changes and improvement in motor behavior in the chronic stage. Finally, limited evidence
supports that the microstructural changes in CST are significantly associated with motor
performance improvement in chronic stroke. Only animal studies and clinical studies of post-
stroke aphasia support potential white matter remodeling mechanisms underlying functional
recovery after stroke (Quallo et al., 2009; Ramos-Cejudo et al., 2015; Sato et al., 2009;
Schaechter et al., 2006; Schlaug et al., 2009; Scholz et al., 2009; Steele et al., 2013; Sterr et al.,
Neurological Biomarkers of Post-Stroke Motor Recovery 6

2014). Thus, longitudinal clinical studies using a DTI measure of CST will reveal the
relationship between brain microstructural changes and UE motor behavior improvement in
chronic stroke.

Specific Aims
The overall purpose of this dissertation is to determine the relationship between CST
white matter characteristics using DTI and UE motor behavior in a mild-to-moderate chronic
stroke population. This dissertation has three specific aims, each comprising a separate chapter in
the document (Figure 1-1):
1) To perform a systematic review of current literature to determine if the neurological
biomarkers can be used to predict post-stroke motor recovery.
2) To examine the most accurate DTI-derived estimate of CST structure by comparing
different estimation methods.
3) a. To test the feasibility of a predictive model using DTI-derived CST biomarker to
estimate improvement in motor behavior.
b. To determine if there is an association between CST structural changes and
improvement in motor behavior.
Neurological Biomarkers of Post-Stroke Motor Recovery 7

Figure 1-1. Illustration of specific aims
Neurological Biomarkers of Post-Stroke Motor Recovery 8

Organization of the Dissertation
Three studies were conducted to address each of the specific aims (Chapter 2-4).
Study 1 (Chapter 2) is a systematic review to address specific aim 1. Qualitative and
quantitative assessment of current literature was performed to identify the biomarker or a set of
biomarkers that best predicts motor recovery after stroke. The biomarkers assessed in this study
included DTI, conventional structural MRI, functional MRI, TMS, and multimodal biomarkers.
The methodological quality of previous studies was evaluated to determine the evidence level
associated with each biomarker. Further, as the quantitative assessment, statistical power and
effect size of predictive models in previous studies were evaluated to determine the most robust
predictive model. The results of this systematic review were published in Neurorehabilitation
and Neural Repair in January 2017.
Study 2 (Chapter 3) was done to determine the most accurate DTI-derived estimate of
CST structure that has the strongest relationship with clinical motor outcome measures. The
method identified in Study 2 was used in Study 3.
Study 3 (Chapter 4) has two separate parts: 1) feasibility of DTI biomarker to predict
motor behavior improvement; 2) determine the relationship between DTI-based structural
changes in CST and improvement in motor behavior. For the feasibility test, I compared a model
using the previously identified DTI-derived CST biomarker alone to other models that
incorporate clinical motor outcomes alone. Lasso regression analysis was used to determine the
most predictive model that included both DTI and clinical biomarkers for motor behavior
improvement in this study population. The second part was performed to determine if CST white
matter change can occur during the chronic stage of stroke, and if that change is associated with
improvement in UE motor behavior.
Neurological Biomarkers of Post-Stroke Motor Recovery 9

The neuroimaging and clinical data utilized in studies 2 and 3 are from a phase-I clinical
trial conducted in the Motor behavior and Neurorehabilitation Laboratory at the University of
Southern California Division of Biokinesiology and Physical Therapy. Chapter 2 – 4 are in
manuscript format in preparation for dissemination to appropriate research journals.
The last chapter (Chapter 5) summarizes results, and discusses the clinical significance,
limitations, and future research directions.
Neurological Biomarkers of Post-Stroke Motor Recovery 10

CHAPTER TWO: CAN NEUROLOGICAL BIOMARKERS OF BRAIN
IMPAIRMENT BE USED TO PREDICT POST-STROKE MOTOR RECOVERY? A
SYSTEMATIC REVIEW

Published in Neurorehabilitation and Neural Repair
Kim, B., Winstein, C., 2017. Can Neurological Biomarkers of Brain Impairment Be Used to
Predict Poststroke Motor Recovery? A Systematic Review. Neurorehabil Neural Repair 31, 3–
24. doi:10.1177/1545968316662708

Abstract
Background: There is growing interest to establish recovery biomarkers, especially neurological
biomarkers, in order to develop new therapies and prediction models for the promotion of stroke
rehabilitation and recovery. However, there is no consensus amongst the neurorehabilitation
community about which biomarker(s) have the highest predictive value for motor recovery.
Objective: To review the evidence and determine which neurological biomarker(s) meet the
high evidence quality criteria for use in predicting motor recovery.
Methods: We searched databases for prognostic neuroimaging/neurophysiological studies.
Methodological quality of each study was assessed using a previously employed comprehensive
fifteen-item rating system. Further, we used the GRADE approach and ranked the overall
evidence quality for each category of neurologic biomarkers.
Results: Seventy-one papers met our inclusion criteria; five categories of neurologic biomarkers
were identified: DTI, TMS, functional MRI, conventional structural MRI, and a combination of
these biomarkers. Most studies were conducted with individuals after ischemic stroke in the
Neurological Biomarkers of Post-Stroke Motor Recovery 11

acute and/or subacute stage (~70%). Less than one-third of the studies (21/71) were assessed
with satisfactory methodological quality (80% or more quality score). Conventional structural
MRI and the combination biomarker category ranked ‘High’ in overall evidence quality.
Conclusions: There were three prevalent methodological limitations: 1) lack of cross-validation,
2) lack of minimal clinically important difference (MCID) for motor outcomes, and 3) small
sample size. More high-quality studies are needed to establish which neurological biomarkers are
the best predictors of motor recovery after stroke. Finally, the quarter-century old
methodological quality tool used here should be updated by inclusion of more contemporary
methods and statistical approaches.
Neurological Biomarkers of Post-Stroke Motor Recovery 12

Introduction
There is growing interest in establishing stroke recovery biomarkers. Researchers define
stroke recovery biomarkers as surrogate indicators of disease state that can have predictive value
for recovery or treatment response (Bernhardt et al., 2016). Specifically, previous studies have
suggested that better understanding of neurological biomarkers, derived from brain imaging and
neurophysiological assessments, is likely to move stroke rehabilitation research forward
(Bernhardt et al., 2016; Stinear et al., 2012).
Recovery biomarkers acquired during the acute and subacute phases (acute – within 1
week after onset; subacute – between 1 week and 3 months after onset) may be vital to set
attainable neurorehabilitation goals and to choose proper therapeutic approaches based on the
recovery capacity. Further, motor recovery prediction using neurological biomarkers in the
chronic phase (more than 3 months after onset) can be useful to determine whether an individual
will benefit from specific therapeutic interventions applied after the normal period of
rehabilitation has ended. Hence, use of recovery biomarkers is likely to improve customization
of physical interventions for individual stroke survivors regarding their capacity for recovery,
and to facilitate development in new neurorehabilitation approaches.

There have been fundamental changes in recovery biomarkers from simple clinical
behavioral biomarkers to brain imaging and neurophysiological biomarkers. In particular, a
number of recent studies have shown that neurologic biomarkers (i.e., neuroimaging and/or
neurophysiological measures of brain) are more predictive of motor recovery than clinical
behavioral biomarkers (Borich et al., 2014; Cramer et al., 2007; Feys et al., 2000).
Although there is some evidence that neurological biomarkers are more valuable as
predictors of motor recovery than clinical behavioral biomarkers, there are significant gaps
Neurological Biomarkers of Post-Stroke Motor Recovery 13

between the published evidence and clinical usage. First, there is no consensus on which specific
neurological biomarkers would be best for prediction models (Cramer et al., 2007; Heiss and
Kidwell, 2014; Jang et al., 2010). Viable neurological biomarker of motor recovery have evolved
from lesion size and location, prevalent in the early 1990’s (S K Schiemanck et al., 2006b) to
more contemporary complex brain network analysis variables (Granziera et al., 2012). Despite
this evolution, there is a paucity of high-level evidence for determining the most critical
neurological biomarkers of motor recovery. A number of literature reviews and systematic
reviews of studies published since the 1990’s aimed to identify the most appropriate biomarkers
of motor recovery or functional independence (Chen and Winstein, 2009; Kwakkel and Kollen,
2013; S K Schiemanck et al., 2006b; Stinear et al., 2014). Among these reviews, only one by
Schiemanck and colleagues (S K Schiemanck et al., 2006b) assessed the evidence quality of
neurologic biomarkers, while many focused on clinical measures (i.e., clinical motor and/or
functional measures) (Chen and Winstein, 2009). Their review was limited to only thirteen
studies that employed structural MRI measures of lesion volume as neurologic biomarkers.
Besides lesion volume derived from structural MRI, there are other viable neurological
biomarkers of brain impairment. Therefore, this systematic review includes a broad set of
relevant biomarkers for consideration as critical predictors for inclusion in motor recovery
prediction models.
Further, there is some evidence to suggest that multivariate prediction models which use
neurological biomarkers in addition to clinical outcome measures are more accurate than those
that use clinical outcome measures alone (Stinear et al., 2007; Stinear et al., 2012). However,
there is still no consensus about whether incorporating behavioral and neurological predictors in
Neurological Biomarkers of Post-Stroke Motor Recovery 14

a multimodal prediction model is superior (i.e., more accurate) to a univariate model that
includes either behavioral or neurological predictors alone.
Taken together, this systematic review has two aims. The first is to conduct a critical and
systematic comparison of selected studies to determine which neurological biomarker(s) is likely
to have sufficient high-level evidence in order to render the most accurate prediction of motor
recovery after stroke. The second aim is to identify whether adding clinical measures along with
neurological biomarkers in the model improves the accuracy of the model compared to the
models that use neurological biomarkers alone.

Methods
Inclusion and exclusion criteria. Given the goal to predict motor recovery after stroke,
the inclusion and exclusion criteria were adapted from a recent systematic review of the same
topic (Table 2-1) (Chen and Winstein, 2009). Major differences are that Chen and Winstein
(Chen and Winstein, 2009) used the International Classification of Functioning and Disability as
an organizing framework for dependent measures (i.e., behavioral outcomes), and they included
clinical prognostic studies without neuroimaging/neurophysiological predictors.

Literature search strategy. Research articles published before December 2015 were
searched for in PubMed, ISI Web of Knowledge, and Google Scholar. Search keywords included
[‘stroke’ and ‘motor recovery’ and ‘predict’] and one of the following keywords [‘neuroimaging’
or ‘neurophysiological measure’ or ‘diffusion tensor imaging’ or ‘magnetic resonance imaging’
or ‘transcranial magnetic stimulation’].

Neurological Biomarkers of Post-Stroke Motor Recovery 15

Table 2-1. Inclusion and exclusion criteria.
Categories Inclusion Criteria Exclusion Criteria
Population
characteristics
Post-cerebral stroke with upper/lower
limb deficits
Brainstem or cerebellar stroke
Age older than 18-year-old Non-adult population
Sample size >=5
Study design
Longitudinal study with different time-
points of predictor and outcome
measures
Cross-sectional study
Measurement
variables
Behavioral outcome (dependent)
measures related to upper and/or lower
limb recovery at two different time-
points
Outcome measures not specifically
related to upper/lower limb recovery
(e.g. cognition, language etc.).
Outcome measures for specific
upper/lower limb impairments (e. g.,
pain, sensory deficit, or spasticity).
Predictor (independent) measures using
one or more
neuroimaging/neurophysiological
measures (e. g., MRI, DTI, or TMS, etc)
Predictor measures using behavioral
measures only
Language English only Non-English

Evidence Methodological Quality Evaluation. We evaluated seventy-one studies using
the evidence methodological quality score (EQS). The methodological quality grading criteria
were adapted from previous systematic reviews of a similar topic

(Chen and Winstein, 2009;
Hier and Edelstein, 1991; Kwakkel et al., 1996; S K Schiemanck et al., 2006b). This evaluation
system includes three categories: internal validity, statistical validity, and external validity. Each
Neurological Biomarkers of Post-Stroke Motor Recovery 16

category contains several items, and each item is scored using a binary system: yes (1) or no (0).
A maximum EQS is 15 (Table 2-2, see Appendix A for detail).
This evidence methodological quality scheme was developed based on the
recommendations of the ‘Task Force on Stroke Outcome Research of Impairments, Disabilities
and Handicap’ (Symposium recommendations for methodology in stroke outcome research. Task
Force on Stroke Impairment, Task Force on Stroke Disability, and Task Force on Stroke
Handicap., 1990), and the methodological guidelines for stroke outcome research are consistent
with these criteria (Gresham, 1986; Kwakkel et al., 1996).

Overall Evidence Quality Evaluation. The evidence grading system from the GRADE
Working Group (Atkins et al., 2004) was adapted to evaluate the overall evidence quality for
each neurological biomarker category. We graded the overall evidence quality using a 4-level
system: ‘High’, ‘Moderate’, ‘Low’, and ‘Very Low’. Table 2-3 describes the criteria for each
level.

Prediction Regression Model Evaluation. To evaluate these models, we estimated
effect size and statistical power of each model. We assume that the statistical power of each
prediction model reflects the accuracy and robustness of the model. If the article reported more
than one model, each model’s statistical power and effect size were calculated separately. We
extracted the R-squared value, significance level, and number of participants to calculate
statistical power (Cohen et al., 2013; Soper, 2015). To compare the prediction models with or
without clinical measures or demographic predictors, models were separated into two groups: 1)
regression models using neurological biomarkers alone and 2) regression models using
neurological biomarkers in conjunction with clinical (or demographic) measures.
Neurological Biomarkers of Post-Stroke Motor Recovery 17

Table 2-2. Evidence quality score (EQS) evaluation categories.
Internal Validity
• Appropriate operational definitions of outcome and predictor variables
A o Outcome measures
B o Predictor measures
• Additional tests or citations for validity or reliability of measures
C o Outcome measures
D o Predictor measures
E • Blinded evaluations
F • Appropriate observation time-points
G • Control of dropout
Statistical Validity
H • Control for statistical significance
I • Appropriate sample size
J • Control for multicollinearity
External Validity
K • Stroke pathology identification
L • Specification of inclusion and exclusion criteria
M • Control for additional treatment effects on the outcome measures
N • Cross-validation of the prediction model
O • Discussion of Minimal Clinically Important Differences (MCID)

Neurological Biomarkers of Post-Stroke Motor Recovery 18

Table 2-3. Categories for overall evidence quality for each biomarker type.
Grade Explanation
High
Further research is very unlikely to change our confidence in the estimate of model.
• More than five studies with higher than 12 (80% of maximum) methodological
quality score, and average methodological quality score of the studies that used the
predictor is above 10.
Moderate
Further research is likely to have an important impact on our confidence in the estimate
of model and may change the estimate.
• Fewer than five studies with higher than 12 methodological quality score with
consistent results, and average methodological quality score of the studies that used
the neurological biomarker is above 10.
• More than five studies with higher than 12 methodological quality score, but average
methodological quality score of the studies that used the neurological biomarker is
below 10.
Low
Further research is very likely to have an important impact on our confidence in the
estimate of model and is likely to change the estimate.
• No study with a methodological quality score higher than 12, and average
methodological quality score of the studies that used the neurological biomarker is
above 10.
• Fewer than five studies with higher than 12 methodological quality score, and
average methodological quality score of the studies that used the neurological
biomarker is below 10.
Very Low
Any estimate of model is very uncertain.
• No study with a methodological quality score higher than 12.
• Average methodological quality score of the studies is below 10.

Neurological Biomarkers of Post-Stroke Motor Recovery 19

Results
Literature search. 451 English language articles were found from the three databases.
By screening the title and abstract, 80 articles were selected for review based on inclusion and
exclusion criteria. Of the articles selected from the screening, nine were excluded. Finally,
seventy-one articles were included in the evidence quality evaluation (Figure 2-1). Details of
included studies are summarized in Table 2-4.

Figure 2-1. Evidence search strategy diagram.

Neurological Biomarkers of Post-Stroke Motor Recovery 20
Table 2-4. Summary of included articles
First Author Stroke Phase Stroke Pathology
# of
Participants
Neurological
Biomarker Type
Observation
Period (Month)
Arac et al., 1994 A I, H 27 TMS 6
Borich et al., 2014 C I 13 DTI 0.25
Burke et al., 2014 SA and C I, H 43 Combination 6
Chang et al., 2015 SA I 14 TMS 6
Cho et al., 2007a SA I 55 DTI 6
Cho et al., 2007b SA H 40 DTI 3
Coutts et al., 2009 A I 457 CT 2.5
Cramer et al., 2007 C I 24 fMRI 1
Dawes et al., 2008 C I 18 Combination 6
Di Lazzaro et al., 2010 A and SA I 17 TMS 0.5
Dong et al., 2006 C N/S 8 fMRI 6
Escudero et al., 1998 A I 50 TMS 5
Feng et al., 2015 A I 76 MRI 3
Feydy et al., 2002 SA I 14 fMRI 12
Feys et al., 2000 SA I 64 TMS 6
Granziera et al., 2012 A I 12 Combination 4.5
Grässel et al., 2010 A and SA I 11 DTI 3
Groisser et al., 2014 A I 10 DTI 30

Neurological Biomarkers of Post-Stroke Motor Recovery 21
First Author Stroke Phase Stroke Pathology
# of
Participants
Neurological
Biomarker Type
Observation
Period (Month)
Hand et al., 2006 A I 82 MRI 9
Hendricks et al., 1997 A I 47 TMS 26
Hendricks et al., 2003 A and SA I 43 TMS 6
Holtmannspötter et al., 2005 N/S I 62 Combination 3
Jang et al., 2010 SA H 53 Combination 1
Jang et al., 2005 A and SA I, H 31 DTI 6
Jang et al., 2004 A I, H 17 fMRI 12
Jung et al., 2013 SA I 8 fMRI 1
Kang et al., 2013 A I 284 MRI 2.56
Karibe et al., 2000 A H 28 MRI 1
Kim et al., 2015 SA I, H 49 DTI 3
Koh et al., 2015 SA I, H 140 MRI 1.25
Koski et al., 2004 C N/S 10 TMS 1
Koyama et al., 2012 SA H 15 DTI 1
Koyama et al., 2013 SA H 32 DTI 3
Kusano et al., 2009 A H 18 DTI 6
Kuzu et al., 2012 A H 23 DTI 1.5
Kwon et al., 2012 SA I 71 DTI 3
Lai et al., 2007 A I 28 Combination 0.43

Neurological Biomarkers of Post-Stroke Motor Recovery 22
First Author Stroke Phase Stroke Pathology
# of
Participants
Neurological
Biomarker Type
Observation
Period (Month)
Lai et al., 2015 C I, H 72 TMS 8
Liggins et al., 2013 A I 110 MRI 12
Lindenberg et al., 2012 C I 15 DTI 3
Liu et al., 2012 A I 48 DTI 2
Loubinoux et al., 2003 SA I 9 fMRI 0.5
Ma et al., 2011 A I 50 MRI 4
Ma et al., 2014 A H 25 DTI 1.5
Maeshima et al., 2013 SA H 25 DTI 3
Marumoto et al., 2013 C I 14 Combination 6
Nascimbeni et al., 2006 A I 33 TMS 3
Nouri and Cramer, 2011 C I 60 Combination 24
Nuutinen et al., 2006 A or SA I 48 Combination 6
Piron et al., 2005 SA I 20 TMS 0.5
Puig et al., 2011 A I 60 Combination 4.5
Puig et al., 2013 A or SA I 70 DTI 0.5
Qiu et al., 2008 A H 8 DTI 3
Quinlan et al., 2014 N/S 40 Combination 12
Rapisarda et al., 1996 A I 26 TMS 6.5
Rehme et al., 2011 A I 12 fMRI 1

Neurological Biomarkers of Post-Stroke Motor Recovery 23
First Author Stroke Phase Stroke Pathology
# of
Participants
Neurological
Biomarker Type
Observation
Period (Month)
Rehme et al., 2015 A or SA I 21 Combination 5
Rickards et al., 2012 C I 14 MRI 3
Rosso et al., 2010 A I 79 Combination 6.25
Schiemanck et al., 2006a A and SA I 75 MRI 3
Stinear et al., 2012 SA I 40 Combination 6
Stinear et al., 2007 C N/S 21 Combination 1.5
Tao et al., 2014 A H 32 Combination 3
Timmerhuis et al., 1996 A I 50 Combination 12
van Kuijk et al., 2009 A I 39 TMS 3
Várkuti et al., 2013 SA and C I, H 9 fMRI 1
Wang et al., 2012 A or SA H 27 DTI 6.5
Wöhrle et al., 2004 A N/S 13 Combination 1
Yoshioka et al., 2008 A H 17 DTI 3
Yu et al., 2009 A I 9 DTI 0.5
Zarahn et al., 2011 A I 30 fMRI 3
Neurological Biomarkers of Post-Stroke Motor Recovery 24

Trend in number of publications by year. The earliest study to meet our
Inclusion/Exclusion criteria was published in 1994. Generally, there was an increasing trend in
the number of publications over time (Figure 2-2). On average, every year between 1994 and
2015 there have been approximately three publications per year with a high of nine in 2012.

Figure 2-2. Trend in number of publications per year.

Stroke Pathology and Phases. 3,215 post-cerebral stroke individuals participated in the
71 studies included. The classification of pathology and stroke phase is summarized in Table 2-5.
Among these participants, 1,800 (56%) from thirty-one studies were in the acute phase of
recovery, 649 (20.2%) from sixteen studies were in the subacute phase, and 269 (9.6%) from
twelve studies were in the chronic phase when the neuroimaging/neurophysiological assessments
occurred.
Neurological Biomarkers of Post-Stroke Motor Recovery 25

Most studies were conducted with ischemic stroke patients (2,392 of 3,215, 75%).
Approximately 50% of patients were in an acute phase with ischemic brain damage. (Table 2-5)

Table 2-5. Number of subjects by across studies by stroke pathology type and timing.
Stroke
Phase
Timing
Stroke Pathology
Total
Ischemic Hemorrhagic Both
Not
reported
Acute
1592
(49.52%)
151
(4.7%)
44
(1.37%)
13
(0.40%)
1800
(55.99%)
Subacute
295
(9.18%)
165
(5.13%)
189
(5.88%)
0
649
(20.19%)
Acute
/Subacute
285
(8.86%)
27
(0.84%)
31
(0.96%)
0
343
(10.67%)
Chronic
158
(4.92%)
0
72
(2.24%)
79
(2.46%)
309
(9.61%)
Subacute
/Chronic
0 0
52
(1.62%)
0
52
(1.62%)
Not
Reported
62
(1.93%)
0 0 0
62
(1.93%)
Total
2392
(74.40%)
343
(10.67%)
388
(12.07%)
52
(2.86%)
3215

Neurological Biomarkers of Post-Stroke Motor Recovery 26

Types of Neurological Biomarkers. There were five main categories of neurological
biomarkers: DTI, TMS, fMRI, sMRI, and a combination of these biomarkers. Eighteen studies
utilized these biomarkers and other clinical measures together as predictors. (Table 2-6)
In the acute phase, DTI, TMS, sMRI, and combination biomarker types were similarly
employed for the prediction model. Functional MRI was the least frequently used type (3 of 31
studies with acute participants). In the subacute phase, DTI biomarker type was predominantly
used (7 of 16 studies with subacute participants). In the chronic phase, combination type was the
most frequent type (5 of 12 studies with chronic participants) (Figure 2-3).
As prevalence of CT biomarker was too low to discuss the methodological quality, we
excluded the CT biomarker from the discussion (only one large-scale retrospective observational
study used a CT biomarker).
We described the details of predictor variables for each neurological biomarker type in
Appendix B material.

Figure 2-3. Frequency of each neurological biomarker type for each stroke phase.

Neurological Biomarkers of Post-Stroke Motor Recovery 27

Table 2-6. Summary of five types of neurological biomarkers.
DTI TMS fMRI MRI CT
Combi-
nation
Total
# of studies
(with
behavioral
predictors)
21 (2) 13 (4) 9 (1) 9 (3) 1 (1) 18 (7) 71
# of
participant
612 462 131 859 457 694 3215
% of total
participant
19.04 14.37 4.07 26.72 14.21 21.59

Clinical Endpoints for Motor Recovery. There were approximately thirty-five different
clinical measures to capture motor recovery after stroke (Figure 2-4). The most frequently (six or
more studies used) utilized clinical endpoint measures were: National Institutes of Health Stroke
Scale (NIHSS, including modified version), Rankin Scale (RS, including modified version),
Barthel Index (BI), Fugl-Meyer Assessment (FMA, including upper and lower extremities, or
separate versions), Motricity Index (MI), Medical Research Council (MRC) score, and Wolf
Motor Function Test (WMFT) Time score.
Evidence Methodological Quality. The results of the evidence methodological quality
evaluation are summarized in Table 2-7. The mean EQS (mean ± standard deviation = 9.79 ±
2.13) of all reviewed studies was relatively high (Figure 2-5, range 5-15). There were twenty-one
out of 71 studies (30%) with an EQS score of 12 (80% of total score) or more, and these studies
are highlighted in Table 2-7.

Neurological Biomarkers of Post-Stroke Motor Recovery 28

Figure 2-4. Summary of motor endpoint (dependent) measures.

Neurological Biomarkers of Post-Stroke Motor Recovery 29
Table 2-7 Summary of evidence methodological quality score evaluation. Highlight indicates the studies with high evidence
quality (EQS ≥ 12).
Reference
Neurological
Biomarker
Type
Internal validity
Statistical
Validity
External Validity
Total
Score
A B C D E F G H I J K L M N O
Arac et al., 1994 TMS 1 1 1 1 0 1 1 0 0 0 1 0 0 0 0 7
Borich et al., 2014 DTI 1 1 1 0 0 0 1 1 0 1 1 1 1 0 0 9
Burke et al., 2014 Combination 1 1 0 1 1 1 1 1 0 1 1 1 1 1 0 12
Chang et al., 2015 TMS 1 1 1 1 0 1 1 0 0 0 1 1 0 0 0 8
Cho et al., 2007a DTI 1 1 1 1 0 1 0 1 1 0 1 1 0 0 0 9
Cho et al., 2007b DTI 1 1 1 1 0 1 0 1 0 0 1 1 0 0 0 8
Coutts et al., 2009 CT 0 1 0 1 0 1 0 1 1 1 1 1 0 0 0 8
Cramer et al., 2007 fMRI 1 1 1 1 0 0 1 1 0 1 1 1 1 0 0 10
Dawes et al., 2008 Combination 1 1 1 1 0 0 0 1 0 1 1 1 1 0 0 9
Di Lazzaro et al., 2010 TMS 1 1 1 1 0 1 0 1 0 1 1 1 1 0 0 10
Dong et al., 2006 fMRI 1 1 1 1 0 0 1 1 0 1 0 1 1 1 0 10
Escudero et al., 1998 TMS 1 1 1 1 0 1 1 1 1 1 1 1 0 1 1 13
Feng et al., 2015 MRI 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 13
Feydy et al., 2002 fMRI 1 1 1 1 0 1 0 0 0 0 1 1 0 0 0 7
Feys et al., 2000 TMS 1 1 1 1 0 1 1 1 1 1 1 1 0 0 0 11
Granziera et al., 2012 Combination 1 1 1 1 0 1 1 1 0 1 1 1 0 0 0 10

Neurological Biomarkers of Post-Stroke Motor Recovery 30
Reference
Neurological
Biomarker
Type
Internal validity
Statistical
Validity
External Validity
Total
Score
A B C D E F G H I J K L M N O
Grässel et al., 2010 DTI 1 1 0 1 0 1 1 0 0 0 1 1 0 0 0 7
Groisser et al., 2014 DTI 1 1 1 1 0 1 1 1 0 0 1 1 0 0 0 9
Hand et al., 2006 MRI 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 12
Hendricks et al., 1997 TMS 1 1 1 1 0 1 1 0 0 0 1 1 0 0 0 8
Hendricks et al., 2003 TMS 1 1 1 1 1 1 1 1 0 0 1 1 0 0 0 10
Holtmannspötter et al., 2005 Combination 1 1 0 1 0 0 1 1 1 1 1 0 0 0 0 8
Jang et al., 2010 Combination 1 1 1 1 1 1 1 1 1 0 1 1 0 0 0 11
Jang et al., 2005 DTI 1 1 0 1 1 1 1 1 0 1 1 1 0 0 0 10
Jang et al., 2004 fMRI 1 1 0 1 0 1 1 0 0 0 1 0 0 0 0 6
Jung et al., 2013 fMRI 1 1 0 1 0 1 1 0 0 0 1 1 0 0 0 7
Kang et al., 2013 MRI 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 12
Karibe et al., 2000 MRI 1 1 0 0 0 0 1 0 0 0 1 1 0 0 0 5
Kim et al., 2015 DTI 1 1 0 0 1 1 1 0 0 0 1 1 0 0 0 7
Koh et al., 2015 MRI 1 1 0 0 0 0 1 1 1 1 1 1 0 0 0 8
Koski et al., 2004 TMS 1 1 1 1 0 0 1 1 0 0 0 1 1 0 1 9
Koyama et al., 2012 DTI 1 1 1 1 1 1 1 1 0 1 1 1 1 0 1 13
Koyama et al., 2013 DTI 1 1 1 1 1 1 1 1 0 1 1 1 1 0 1 13
Kusano et al., 2009 DTI 1 1 1 1 0 1 1 1 0 0 1 1 0 0 0 9
Kuzu et al., 2012 DTI 1 1 1 1 0 1 1 1 0 1 1 1 0 0 0 10

Neurological Biomarkers of Post-Stroke Motor Recovery 31
Reference
Neurological
Biomarker
Type
Internal validity
Statistical
Validity
External Validity
Total
Score
A B C D E F G H I J K L M N O
Kwon et al., 2012 DTI 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 12
Lai et al., 2007 Combination 1 1 0 1 0 1 1 0 0 0 1 0 0 0 0 6
Lai et al., 2015 TMS 1 1 1 1 1 1 1 1 1 0 1 1 1 0 0 12
Liggins et al., 2013 MRI 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 13
Lindenberg et al., 2012 DTI 1 1 1 1 1 0 1 1 0 1 1 1 1 0 0 11
Liu et al., 2012 DTI 1 1 0 1 0 1 1 1 0 1 1 1 0 0 0 9
Loubinoux et al., 2003 fMRI 1 1 1 1 0 1 1 1 0 1 1 1 1 1 0 12
Ma et al., 2011 MRI 1 1 1 1 0 1 1 1 1 1 1 1 0 0 0 11
Ma et al., 2014 DTI 1 1 1 0 1 1 1 1 0 1 1 1 1 0 0 11
Maeshima et al., 2013 DTI 1 1 1 1 1 1 1 1 0 1 1 1 1 0 0 12
Marumoto et al., 2013 Combination 1 1 1 1 1 0 1 1 0 1 1 1 1 0 0 11
Nascimbeni et al., 2006 TMS 1 1 0 1 0 1 1 1 0 1 1 1 1 0 0 10
Nouri and Cramer, 2011 Combination 1 1 0 1 1 0 1 1 1 1 1 1 1 0 1 12
Nuutinen et al., 2006 Combination 1 1 1 0 1 1 1 1 0 1 1 1 0 0 0 10
Piron et al., 2005 TMS 1 1 1 1 1 1 1 1 0 0 1 1 0 0 0 10
Puig et al., 2011 Combination 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 12
Puig et al., 2013 DTI 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 12
Qiu et al., 2008 DTI 1 1 0 0 0 1 1 1 0 0 1 0 0 0 0 6
Quinlan et al., 2014 Combination 1 1 1 1 1 0 1 1 0 1 0 1 1 1 1 12

Neurological Biomarkers of Post-Stroke Motor Recovery 32
Reference
Neurological
Biomarker
Type
Internal validity
Statistical
Validity
External Validity
Total
Score
A B C D E F G H I J K L M N O
Rapisarda et al., 1996 TMS 1 1 1 1 0 0 1 1 0 0 1 1 0 0 0 8
Rehme et al., 2011 fMRI 1 1 1 1 0 1 1 1 0 1 1 1 0 0 0 10
Rehme et al., 2015 Combination 1 1 1 1 0 1 1 1 0 0 1 1 1 1 1 12
Rickards et al., 2012 MRI 1 1 1 0 0 0 1 1 0 0 1 1 1 0 0 8
Rosso et al., 2010 Combination 1 1 0 0 0 1 1 1 1 1 1 1 1 0 0 10
Schiemanck et al., 2006a MRI 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 12
Stinear et al., 2012 Combination 1 1 1 1 1 1 1 1 0 0 1 1 1 0 1 12
Stinear et al., 2007 Combination 1 1 1 1 1 0 0 1 0 1 0 1 1 0 0 9
Tao et al., 2014 Combination 1 1 1 1 1 0 1 1 0 0 1 1 0 0 0 9
Timmerhuis et al., 1996 Combination 1 1 1 1 0 1 1 1 1 1 1 1 1 0 0 12
van Kuijk et al., 2009 TMS 1 1 1 1 1 1 1 1 0 0 1 1 1 0 1 12
Várkuti et al., 2013 fMRI 1 1 0 1 0 1 1 1 0 1 0 1 1 1 0 10
Wang et al., 2012 DTI 1 1 1 1 0 1 1 1 0 0 1 1 0 0 0 9
Wöhrle et al., 2004 Combination 1 1 0 1 0 0 1 0 0 0 0 0 0 0 0 4
Yoshioka et al., 2008 DTI 1 1 0 1 0 1 1 0 0 0 1 1 0 0 0 7
Yu et al., 2009 DTI 1 1 1 1 0 1 1 1 0 0 1 1 0 1 0 10
Zarahn et al., 2011 fMRI 1 1 1 0 0 1 1 1 0 1 1 1 0 0 0 9
Total 70 71 52 61 28 54 64 59 20 41 65 65 28 8 9

Neurological Biomarkers of Post-Stroke Motor Recovery 33

Figure 2-5. Distribution of evidence methodological quality scores.

Internal validity. Most studies provided proper operational definitions of predictor and
outcome measure variables. Fifty-two studies (73 %) described validity and/or reliability of
clinical endpoint measures, and sixty-one studies (86 %) explained validity and reliability of
their neurological measures. In 28 studies (39 %), raters of predictor and outcome variables were
blinded to the study purpose and other measures. In 54 studies (76 %), the predictor variables
were measured within 1 month after the index stroke, and the outcome measures were assessed at
least eight weeks after the measure of predictor variables. Among the 59 studies with post-stroke
individuals in the acute and subacute phases, the mean observation period was about 5 months
after stroke (Table 2-8). Sixty-four studies (90 %) described the number of and reasons for
dropouts, or had no dropouts.

Neurological Biomarkers of Post-Stroke Motor Recovery 34

Table 2-8. Summary of observation periods.
Observation period Number of studies (% of total)
Less than 3 months 23 (32.4 %)
More than 3 months and less than 6 months 22 (31 %)
More than 6 months and less than 12 months 18 (25.4 %)
More than 12 months 8 (11.3 %)
Total 71

Statistical validity. Most studies (59 of 71) applied appropriate statistical analyses. For
the appropriate sample size, however, only twenty studies (~ 28 %) met the appropriate sample
size criteria. Further, forty-one studies (~ 58 %) considered multicollinearity to control the
effects of confounding variables on their correlation or regression analyses.

External validity. Most studies (67 of 71) identified stroke pathology. Sixty-five studies
(~ 92%) described inclusion and exclusion criteria. Six studies did not specify inclusion and
exclusion criteria, or their descriptions of the criteria were insufficient to replicate the
recruitment criteria. Only twenty-eight studies (39%) discussed the effects of additional
treatment on outcomes. Most of these studies described the treatments, such as physical therapy
and/or occupational therapy, which were provided during study participation. Only eight studies
(~11%) performed the cross-validation of their prediction model using an independent group of
participants with stroke. Nine studies (~13%) discussed clinically meaningful differences of
predictors or outcome measures.
Neurological Biomarkers of Post-Stroke Motor Recovery 35

Overall Evidence Quality for Each Neurologic Biomarker Category. Table 2-9
summarizes the results of the overall quality grading. Conventional structural MRI and
Combination biomarker types met the ‘high’ grade criteria. DTI biomarker type met the
‘Moderate’ grade criteria. TMS and fMRI met only the ‘Low’ grade. The overall quality grade
for the CT predictor was ‘Very Low’, because there was only one study, which had a limited
EQS score of eight.

Table 2-9 Summary of overall evidence quality of each biomarker type.
Biomarker Type
# of study with EQS≥12
(% of total)
Average ± SD
of EQS
Predictor Quality
Category
Combination 7 (38.89 %) 10.06 ± 2.29 High
sMRI 5 (55.57 %) 10.44 ± 2.80 High
DTI 5 (23.81 %) 9.67 ± 2.03 Moderate
TMS 3 (23.08 %) 9.85 ± 1.82 Low
fMRI 1 (11.11 %) 9.00 ± 1.94 Low
CT 0 (0 %) 8.00 ± 0 Very Low

Prediction Regression Models. Among seventy-one reviewed studies, thirty-two
(~45 %) conducted linear or non-linear regression analyses to develop prediction regression
models that included neurological biomarkers. Thirty studies (~42 %) reported statistically
significant neurological biomarkers as predictors in their regression models. Among these thirty
studies, twenty-two (~31 %) informed statistical details of their regression models.
Neurological Biomarkers of Post-Stroke Motor Recovery 36

From the twenty-two studies that reported details, thirty-nine regression models were
identified that used neurological biomarkers. Among these models, twenty (~50% of reported
models) consisted of neurological biomarkers and clinical measures as statistically significant
predictors (Table 2-10).
The mean statistical power and the mean effect size of the thirty-nine prediction models
was high (i.e., 0.944 and 6.197 respectively). The statistical power and the effect size of
multivariate models (i.e., models using neurological biomarkers and clinical measures as
predictors) were significantly greater than the power of the models using neurological
biomarkers alone (Figure 2-6 & 2.7 and Table 2-11). There were four studies which
overestimated effect size (effect size is greater than 10), but removing these four studies did not
influence on the statistical results for comparing the effect sizes.
Further, six studies reported models with clinical motor behavioral predictors alone, in
addition to the models incorporating behavioral and neurological predictors. Among these six
studies, five reported that the prediction model with clinical behavioral predictors and
neurological biomarkers explained more variance in the outcome variable than the model with
clinical behavioral predictors alone. (Table 2-12)

Neurological Biomarkers of Post-Stroke Motor Recovery 37

Figure 2-6. Comparison of statistical power and effect size between models using
neurological biomarkers alone and models using neurological biomarkers and clinical
measures. (A) Statistical power; (B) effect size. Model group 1 indicates the models using
neurological biomarkers alone, and the model group 2 indicates the models using neurological
biomarkers and clinical measures as predictors. The extremely high effect size for model group 2
(see Table 2-10) has been removed from Figure 2-6B for plotting purposes.

Neurological Biomarkers of Post-Stroke Motor Recovery 38

Figure 2-7. Forest plot for comparing effect sizes between different model groups. Each line
represents 95% of confidence interval of effect size of each model. The size of black box
represents sample size of each model. The diamonds indicate the mean of effect sizes for each
model group.

Neurological Biomarkers of Post-Stroke Motor Recovery 39
Table 2-10 Summary of regression models. Effect size (Cohen’s f2) =
"
#
$%"
#
. Statistical power was calculated using R-squared,
number of predictors, probability level, and sample size.
Reference Predictors Outcome Measures
Predictors
from clinical
measures
R
2

Effect
Size
(f
2
)
Statistical
Power
Borich
Chronicity, Age, CST FA (ipsilesional and
contralesional)
Changes in motor
performance
yes 0.75 2.94 0.97
Chronicity, Age, rFA of CST
Changes in motor
performance
yes 0.57 1.34 0.81
Burke
Leg FM Score, fMRI activation volume within
ipsilesional foot primary sensorimotor cortex
Δ gait velocity from
baseline to week 12
yes 0.63 1.70 0.99
Cramer
Arm motor FM score, Degree of activation in
ipsilesional primary motor cortex
Δ arm motor FM score
baseline to week 10
yes 0.40 0.67 0.93
Dawes
maximal cross-sectional overlap (degree of
overlap of lesion with CST in mm), 10m walk
time, age
Walking Speed yes 0.98 40.67 1.00
maximal cross-sectional overlap (degree of
overlap of lesion with CST in mm), 10m walk
time, age
Normalized Swing Time
Asymmetry
yes 0.99 70.43 1.00
Dong LI of M1
Post-therapeutic Δ the
Mean WMFT Time
no 0.46 0.85 0.59
Escudero
BI, strength, CNS, CCT BI yes 0.29 0.41 0.95
BI, strength, CNS, CCT Strength Index yes 0.54 1.17 0.99
Feng
Weighted CST-lesion load
FM-UE at 3 months after
stroke
no 0.64 1.78 1.00
Weighted CST-lesion load
NIHSS arm motor score at
3 months after stroke
no 0.58 1.38 1.00
Weighted CST-lesion load
NIHSS total score at 3
months after stroke
no 0.45 0.82 1.00

Feys

Motor performance, MEP, overal disability, and
muscle tone within 1-month post-stroke
Brunnstrom-Fugl-Meyer
Score at 6 mo
yes 0.85 5.67 1.00
Motor performance, SSEP at 2-month post-
stroke
Brunnstrom-Fugl-Meyer
Score at 6 mo
yes 0.62 1.53 1.00

Neurological Biomarkers of Post-Stroke Motor Recovery 40
Reference Predictors Outcome Measures
Predictors
from clinical
measures
R
2

Effect
Size
(f
2
)
Statistical
Power

Feys
Motor performance, MEP, overal disability, and
muscle tone within 1-month post-stroke
Brunnstrom-Fugl-Meyer
Score at 12 mo
yes 0.82 4.14 1.00
Motor performance, SSEP at 2-mo post-stroke
Brunnstrom-Fugl-Meyer
Score at 12 mo
yes 0.58 1.38 1.00
Granziera
NIHSS score, GFA M1-SMA, GFA PMd-PMv,
GFA SMA-SMA, Age
NIHSS at 6 mo post-stroke yes 0.98 49.00 1.00
Holtmannspö
tter
MD
Δ Rankin Score about 26
mo post-stroke
no 0.36 0.56 0.99
Jang
rFA (CR group)
Δ MRC score 3 mo post-
stroke
no 0.84 5.29 1.00
rFA (IC group)
Δ MRC score 3 mo post-
stroke
no 0.57 1.31 0.98
Koh
Baseline UE motor score, Stroke pathology
type, Baseline NIHSS score, Lesion location
UE motor score at
discharge
yes 0.59 1.44 1
Stroke pathology type, Baseline NIHSS score,
Lesion location
Δ UE motor score yes 0.47 0.87 1
Koyama
rFA MRC grade (UE) no 0.27 0.37 0.60
rFA MRC grade (LE) no 0.25 0.33 0.55
Koyama
rFA of CP mRS no 0.23 0.31 0.87
rFA of CP MRC grade (UE) no 0.27 0.37 0.92
rFA of CP MRC grade (LE) no 0.19 0.24 0.77
Lindenberg
CST FA & M1-M1 FA
WMFT change (%) post-tx
(12 d from baseline)
no 0.61 1.56 0.98
CST AD & M1-M1 AD WMFT change (%) no 0.64 1.78 0.99
CST RD & M1-M1 AD WMFT change (%) no 0.69 2.23 0.99
Marumoto PLIC rFA FM-UE after tx no 0.65 1.86 0.99
Nascimbeni
NIHSS, MEP amplitude (percentage of
compound motor action potential amplitude)
MRC score yes 0.74 3.00 0.99
NIHSS, MEP amplitude (percentage of
compound motor action potential amplitude),
MI
MRC score yes 0.75 2.82 0.99

Neurological Biomarkers of Post-Stroke Motor Recovery 41
Reference Predictors Outcome Measures
Predictors
from clinical
measures
R
2

Effect
Size
(f
2
)
Statistical
Power
Puig
PLIC damage m-NIHSS 90 d post-stroke no 0.76 6.14 1.00
m-NIHSS & PLIC damage m-NIHSS 90 d post-stroke yes 0.86 9.00 1.00
Quinlan
iM1 activation, iM1-cM1 connectivity,
Percentage injury to CST
Treatment-induced
behavioral gains (from
changes in FMA and
ARAT)
no 0.44 0.79 0.97
Stinear FAAI, MEP presence, clinical score Δ FMA score yes 0.91 10.11 1.00
Timmerhuis MEP, BI, age BI 6 wks post-stroke yes 0.73 2.70 0.99
Zarahn
Weighted CST-lesion load, raw CST-lesion
load, lesion size
Δ FMA score at 3 mo post-
stroke
no 0.73 2.70 1.00
FM, Fugl-Meyer assessment; NIHSS, National Institute of Health Stroke Scale; MRC, Medical Research Council; mRS, Modified Rankin
Score; ARAT, Action Research Arm Test; CST, Corticospinal tract; MEP, Motor-evoked potential; LI, Laterality Index; BI, Barthel Index;
CNS, Canadian Neurological Scale; CCT, Central Conduction Time; GFA, Generalized Fractional Anisotropy; M1, primary motor cortex;
SMA, Supplementary motor area; PMd, Premotor cortex dorsal part; PMv, Premotor cortex ventral part; MD, Mean Diffusivity; AD, Axial
Diffusivity; RD, Radial Diffusivity; FA, Fractional Anisotropy; rFA, ratio of FA between ipsi- and contra-lesional hemispheres; DTI, Diffusion
Tensor Image; DWI, Diffusion Weighted Image; ADC, Apparent Diffusion Coefficient; rADC, ratio of ADC between two hemispheres; iADC,
ipsi-lesional hemisphere ADC; iFA, ipsi-lesional hemisphere FA; iM1, ipsilesional primary motor cortex; cM1, contra-lesional primary motor
cortex; FAAI, FA asymmetry index; PLIC, Posterior Limb of Internal Capsule; m-NIHSS, modified NIHSS; UE, upper extremity [*Highlight
indicates the models with low to moderate effect size (i.e., Cohen’s f
2
<0.8).]
Neurological Biomarkers of Post-Stroke Motor Recovery 42

Table 2-11 Statistical power and effect size comparison between regression models using
neurological biomarkers alone and models using neurological + clinical (demographic)
measures as predictors.
Statistical
Power
Comparison
Mean
Standard
deviation
Degree of
freedom
t-stat P value
Effect size of
t-test
(Cohen’s d)
Models with
neurological
biomarkers
alone
0.9047 0.1557 37 2.0960 .043 0.24
Models with
neurological
biomarker and
clinical
measures
0.9807 0.0439
Effect Size
(Cohen’s f
2
)
Comparison
Mean
Standard
deviation
Degree of
freedom
t-stat P value
Effect size of
t-test
(Cohen’s d)
Models with
neurological
biomarkers
alone
1.6143 1.6188 37 2.0115 .052 2.76
Models with
neurological
biomarker and
clinical
measures
10.5505 19.2872

Neurological Biomarkers of Post-Stroke Motor Recovery 43
Table 2-12 Prediction models using only clinical behavioral predictors.
Author Predictors Outcome measures R-squared
Borich et al.,
2014
Post-stroke duration, age Changes in motor performance 0.008
Cramer et al.,
2007
Arm motor Fugl-Meyer score at baseline Changes in arm motor Fugl-Meyer score 0.21
Feys et al., 2000 Motor performance, muscle tone at baseline Brunnstrom-Fugl-Meyer Score at 2 months 0.74
Koh et al., 2015
Baseline UE motor score Changes in UE motor score 0.04
Baseline NIHSS score Changes in UE motor score 0.05
Baseline finger extension Changes in UE motor score 0.06
Puig et al., 2013 Modified NIHSS within 3 days after stroke Modified NIHSS at 90 days 0.90
Timmerhuis et
al., 1996
Barthel index Barthel index at 6 weeks post-stroke 0.34
Barthel index, Age Barthel index at 6 weeks post-stroke 0.71
* Highlighted model indicates model superiority compared with models using neurologic biomarkers.

Neurological Biomarkers of Post-Stroke Motor Recovery 44

Discussion
To our knowledge, this is the first evidence-based review to have critically and
systematically evaluated the extant literature related to neurological biomarkers to determine the
best predictor variables for motor recovery after stroke.

Evidence Methodological Quality. Numerous methodologically robust clinical studies
provide evidence that structural biomarkers or a combination of different neurological
biomarkers including DTI are useful to predict motor recovery after stroke. Several
methodological weaknesses were found in studies using fMRI or TMS biomarkers, that included:
small sample size, a lack of blinded evaluation of outcome measures, no control for
multicollinearity, or no control for additional treatment effects.
Further, most studies (~90%) provided no cross-validation of the predictive models and
no discussion of the minimal clinically important differences (MCID). A cross-validation of
prediction models on an independent group of participants should be conducted to verify the
validity and accuracy of prediction models (Kwakkel et al., 1996). Therefore, the lack of cross-
validation of the model would be the biggest limitation of the current literature. In recent studies,
a number of statistical methods for model validation have been suggested, such as leave-one-out
cross-validation or k-fold cross-validation (Kirschen et al., 2000; Weimar et al., 2004). Use of
these statistical cross-validation methods is likely to improve the accuracy estimation of the
prediction model (Kirschen et al., 2000). MCID is considered to be an important factor,
particularly for interpretation of the relevance of observed changes in clinical endpoints (Lang et
al., 2008). Consideration of MCDI for neurological biomarkers and clinical motor endpoints will
likely improve the clinical usability of the predictive models for motor recovery.
Neurological Biomarkers of Post-Stroke Motor Recovery 45

Overall Evidence Quality. Only two types of biomarkers (i.e., conventional structural
MRI (sMRI) biomarker type and combination type) were graded as high in overall evidence
quality. Therefore, it is likely that we have sufficient evidence to utilize these two types of
neurological biomarkers for development of prediction model.
Although DTI-derived biomarkers are the most frequently used in reviewed studies, the
evidence methodological quality of those studies was insufficient. DTI is a promising non-
invasive neuroimaging tool that captures orientation and microstructural characteristics of white
matter in the human brain (Soares et al., 2013). The popularity of DTI among stroke
rehabilitation researchers is likely the ease with which one can quantify the structural
characteristics of specific pathways affected by the stroke (Lindenberg et al., 2010). Therefore,
future DTI studies that employ higher methodological quality are likely to have an important
impact on our confidence in the prediction model(s) derived from them.
TMS measures were the second most prevalent biomarker (i.e., 13 of 71 studies).
However, only three of the twelve TMS studies had high methodological quality. The low
prevalence of high-quality prognostic studies in this category resulted in a ‘low’ overall evidence
quality grade.
There was also a low prevalence of high-quality prognostic studies using fMRI.
Specifically, only one fMRI study had an EQS ≥12. Thus, more methodologically robust
prognostic studies using TMS or fMRI will be needed to raise our confidence in the estimate of
the prediction model using these biomarkers.

Neurological Biomarkers of Post-Stroke Motor Recovery 46

Frequently Used Predictor Variables for Each Biomarker Type. In the following
section, we will discuss the most frequently used predictor variables for each biomarker type.
The details of predictor variables are described in the Appendix B.

DTI biomarker type. Ratio and asymmetry index of fractional anisotropy (FA) between
ipsi- and contra-lesional corticospinal tracts (CSTs) were the most popular predictor variables in
DTI studies among many DTI-derived variables in Table 2-13. FA of ipsilesional CST is
associated with microstructural characteristics of white matter fibers (Borich et al., 2014; Scholz
et al., 2009). A Lower FA value of the ipsilesional CST may indicate greater damage of the CST
that can lead to more Wallerian degeneration of CST axons (Puig et al., 2010). However, the FA
value can be influenced by a number of other factors, such as white matter architecture.
Therefore, we need to use the DTI-derived FA values as neurologic biomarkers of brain
impairment with caution.

TMS biomarker type. A number of TMS studies have shown that the presence of an MEP
when stimulating the upper or lower extremity muscle representation areas of ipsilesional
primary motor cortex (M1) is a good indicator of a significant motor recovery. In most cases, a
TMS response in a specific arm/hand or leg/foot muscle was recorded as binary data (i.e., absent
or present) (Stinear et al., 2012). Although the presence of MEP of upper or lower extremity is a
crucial predictor of motor recovery, this variable is insufficient as a predictor alone in the model.
To improve the accuracy of the model, the TMS biomarker should be incorporated with other
neurologic and/or clinical biomarkers (Feys et al., 2000).

Neurological Biomarkers of Post-Stroke Motor Recovery 47

Table 2-13 Summary of DTI-derived predictor variables.
Reference
DTI-derived predictor
variables
DTI method ROIs Data type
Borich iFA
2-D ROI on an
axial slice
PLIC Continuous
Cho CST continuity Tractography CST Categorical
Cho CST continuity Tractography CST Categorical
Granziera Generalized FA Tractography Motor network Continuous
Grässel ADC, rFA, rAD, rRD 3-D VOI PT Continuous
Groisser dFA, dMD, dAD, dRD
Template
3-D VOI
CST Continuous
Jang rFA
2-D ROI on an
axial slice
CR & PLIC Continuous
Jang CST continuity Tractography CST Categorical
Kim iFA, rFA, # of fibers Tractography CST Continuous
Koyama rFA
2-D ROI on an
axial slice
CP Continuous
Koyama rFA
Template
3-D VOI
CST (CR/IC
portion and CR
portion
separately)
Continuous
Kusano rFA
2-D ROI on an
axial slice
CP Continuous
Kuzu iFA
2-D ROI on an
axial slice
CP Continuous
Kwon CST continuity Tractography CST Categorical
Lai CST Topography Tractography CST Categorical
Lindenberg iFA, iAD, iRD Tractography CST & aMF Continuous
Liu
ADC & FA (ipsi, contra, and
ratio)
2-D ROI on an
axial slice
CP Continuous
Ma Generalized FA
2-D ROI on an
axial slice
CP Continuous
Maeshima iFA
2-D ROI on an
axial slice
CP Continuous
Marumoto rFA
Template
3-D VOI
PLIC Continuous
Puig rFA Tractography
CST at the
pons
Continuous
Puig
Damage to specific CST
regions
Tractography
MC, PMC, CS,
CR, PLIC
Categorical
Qiu
Pyramidal fiber tract drawn
lines
Tractography CST (PT) Continuous
Neurological Biomarkers of Post-Stroke Motor Recovery 48

Reference
DTI-derived predictor
variables
DTI method ROIs Data type
Quinlan iFA
2-D ROI on an
axial slice
CP Continuous
Rosso ADC
Template
3-D VOI
CST Continuous
Stinear FA asymmetry index
Template
3-D VOI
PLIC
Continuous &
Categorical
Stinear FA & MD asymmetry index
2-D ROI on an
axial slice
PLIC
Continuous &
Categorical
Tao rFA
2-D ROI on an
axial slice
CP Continuous
Wang rFA
2-D ROI on an
axial slice
CP Continuous
Yoshioka rFA & rADC Tractography CST Continuous
Yu rFA, rMD, rAD, rRD Tractography CST & CC Continuous
2-D ROI, two-dimensional region of interest; 3-D VOI, three-dimensional volume of interest; FA, fractional
anisotropy; AD, axial diffusivity; RD, radial diffusivity; MD, medial diffusivity; ADC, apparent diffusion
coefficient; i-, ipsilesional; r-, ratio between ipsilesional and contralesional; d-, difference between ipsilesional
and contralesional; CST, corticospinal tract; CR, corona radiata; PT, pyramidal tract; PLIC, posterior limb of
internal capsule; CP, cerebral peduncle; aMF, alternative motor fibers; CC, corpus callosum; MC, motor cortex;
PMC, premotor cortex; CS, centrum semiovale

sMRI biomarker type. Conventional sMRI studies usually used lesion location and
volume information as predictors. Specifically, CST-lesion overlap volume (CST-lesion load)
was calculated to quantify how much CST is damaged due to stroke (Burke et al., 2014). In this
review, we carefully used the term ‘CST structural integrity’ separating it from ‘CST
microstructural characteristics’. In previous literature, there was no distinction between ‘CST
structural integrity’ and ‘CST structural characteristics’. Investigators refer to DTI-derived
metrics, such as fractional anisotropy (FA) and mean diffusivity (MD) of CST as ‘CST structural
integrity’, but these DTI-derived metrics represent the water molecules’ diffusion directions and
patterns along the axon (Soares et al., 2013; Sterr et al., 2014). Thus, using ‘structural integrity’
for these DTI-derived variables lacks precision. We refer to ‘CST structural integrity’ as the
amount of damage to CST, which represents the overlap volume between the stroke lesion and
the CST.
Neurological Biomarkers of Post-Stroke Motor Recovery 49

Functional MRI biomarker type. In functional imaging studies, the laterality index of
ipsilesional M1 and functional connectivity between bilateral M1s during ipsilateral motor task
performance were the most common predictor variables. Further, recent studies have shown that
functional connectivity among sensori-motor regions after stroke in resting-state can be a
significant biomarker of brain functional impairment (Jung et al., 2013; Rehme et al., 2011;
Várkuti et al., 2013).
Combination type. In studies using multiple neurological biomarkers, a combination of
DTI and conventional sMRI biomarkers was the most common case. Further, a combination of
DTI and TMS biomarkers was the next most frequently used. All combinations of biomarkers
from included studies are listed in Table 2-14.

Table 2-14 Summary of combinations of neurological biomarkers.
Reference Combinations of Neurological Biomarkers
Burke DTI + fMRI
Dawes DTI + sMRI
Granziera DTI + sMRI
Holtmannspötter DTI + sMRI
Jang DTI + TMS
Lai DTI + sMRI
Marumoto DTI + sMRI
Nouri sMRI + TMS
Nuutinen DWI + PWI + CT
Puig DTI + sMRI
Quinlan DTI + sMRI + fMRI
Rehme fMRI + sMRI
Rosso DTI + sMRI
Stinear DTI + TMS
Stinear DTI + TMS + fMRI
Tao DTI + sMRI
Timmerhuis TMS + sMRI
Wöhrle DWI + PWI + TMS
DTI, diffusion tensor image; fMRI, function magnetic resonance image; sMRI, conventional
structural magnetic resonance image (including T1-weighted and T2-weighted images); TMS,
transcranial magnetic stimulation; DWI, diffusion weighted image; PWI, perfusion weighted image;
CT, computed tomography
Neurological Biomarkers of Post-Stroke Motor Recovery 50

Dependent Variables (Clinical Endpoints) of Prediction Models. The most commonly
employed clinical outcomes were the Rankin Scale (RS or modified RS), NIH Stroke Scale (or
modified NIHSS) motor portion, and Barthel Index (BI). Although these clinical outcome
measures have been proven to be highly reliable and valid (Kasner, 2006), they lack specificity
for motor impairment or performance in individuals post-stroke (Bonita and Beaglehole, 1988).
Further, these low-level categorical scales lack sensitivity and resolution for detecting motor
recovery (Kasner, 2006).
There were several other clinical motor outcome variables that were utilized in reviewed
studies, such as the Fugl-Meyer Assessment (FMA), Wolf Motor Function Test (WMFT) time
score, Motricity Index (MI), Action Research Arm Test (ARAT), Nine Hole Peg Test (NHPT),
Grip force assessment, and Walking performance measures. These measures represent the ‘Body
Structure’ and/or ‘Activity’ levels of the International Classification of Functioning (ICF) and
are more specific to motor impairment/performance (Chen and Winstein, 2009). As such, these
motor-specific measures are more sensitive and may be more appropriate than other more
generic and broad-based clinical outcome measures when the goal is to capture changes in motor
behavior.
Recent studies have utilized a composite measure statistically derived from multiple
clinical outcome measures, as a single outcome measure cannot capture all dimensions of motor
recovery (Kasner, 2006). Quinlan and colleagues (Quinlan et al., 2014) utilized principal
component analysis (PCA) of two different motor outcome scores to specify changes in motor
behavior. Incorporating several different motor outcome measures will improve the accuracy of
model estimation by reducing the measurement errors in dependent variable. However, although
using these data reduction methods including PCA are attractive from a statistical standpoint, the
Neurological Biomarkers of Post-Stroke Motor Recovery 51

clinical meaningfulness of the derived composite measure is not that transparent. Therefore,
researchers should consider the pros and cons of when composite measures motor recovery are
used as dependent variables of prediction models pertaining to changes in motor behavior.

Difference in Prediction models among the acute, subacute and the chronic phases
of stroke. Most participants in the included studies were in the acute stroke phase rather than
subacute or chronic phase (~56%). It has been suggested that prediction of motor recovery in the
early phase of stroke may play an important role in tailoring neurorehabilitation therapies for
each individual (Groisser et al., 2014; Puig et al., 2011; Stinear and Stinear, 2010). The initial
assessment of brain impairment within 1 week after stroke including lesion volume or location
was the most common predictor of motor recovery in previous studies. This might be because of
the predominance of retrospective prognostic studies using diagnostic structural MRI. Further,
TMS was also frequently used to predict motor recovery during the acute phase. Researchers
have reported that the presence of an MEP for hand muscles on the affected side during the acute
phase is a strong predictor of upper extremity motor recovery (Arac et al., 1994; Stinear et al.,
2012, 2007). DTI biomarkers were also frequently used as a predictor in the acute phase, but this
is a controversial topic. There is evidence that DTI measures of CST cannot capture the
structural impairment within two weeks after stroke (Puig et al., 2013). This might be associated
with the time course of Wallerian degeneration of white matter fibers after stroke (Puig et al.,
2010). Puig and colleagues also showed that the FA ratio between ipsi- and contra-lesional CSTs
acquired at admission and at day 3 after stroke is not a significant predictor of motor recovery,
while these metrics acquired at day 30 is a strong predictor of motor recovery at 3 months (Puig
et al., 2013). Further, DTI biomarkers were predominantly used for the prediction model in the
Neurological Biomarkers of Post-Stroke Motor Recovery 52

subacute phase. This suggests that a DTI measure of sensorimotor pathways taken during the
subacute phase would be a viable predictor for motor recovery, while those captured during the
acute phase would not.
Prediction of motor recovery at the chronic stroke phase may be much more complex.
Individuals at the chronic stage may be near to their motor recovery potential (Van Kordelaar et
al., 2014; Winters et al., 2014), and there are likely a number of additional secondary factors that
can influence motor recovery including: psycho-social factors (Winstein et al., 2014),
biomechanical factors (Gao et al., 2009), motor learning (Winstein et al., 1999), and changes in
brain structural and/or functional connectivity (Crofts et al., 2011). As such, it makes sense that
prediction of motor recovery during this later phase should include multiple neurologic and
clinical biomarkers to account for these additional secondary factors. A number of studies of
individuals in the chronic stage after stroke have developed prediction models for motor
improvement using neurological biomarkers (Lindenberg et al., 2012). Prediction models for
motor recovery at the chronic stage are focused on whether the individual can benefit from
specific treatments, such as transcranial direct current stimulation (t-DCS) (O’Shea et al., 2014),
constraint-induced movement therapy (CIMT) (Rickards et al., 2014), or behavioral
neurorehabilitation therapies including task-specific motor practice (Stinear et al., 2007).
However, we lack high-quality prospective longitudinal clinical trials to develop a multimodal
prediction model for motor recovery in the chronic phase (Stinear and Byblow, 2014).

Limitations. Our findings are limited by the prevalent use of broad-based clinical
endpoints that lack sensitivity, specificity, and resolution for motor recovery, particularly that
along the restitution-substitution continuum (Levin et al., 2009).
Neurological Biomarkers of Post-Stroke Motor Recovery 53

Further, several methodological weaknesses limited the impact of studies using
functional MRI, conventional structural MRI, or TMS biomarkers. The reader is cautioned,
however, that the low prevalence of methodologically robust prognostic studies that used these
biomarkers as opposed to the more prevalent ones may have skewed our results.
Another limitation is that the evidence methodological evaluation tool used here is likely
outdated and in need of revision. As the evidence evaluation tool was developed in 1990 when
the imaging technology was in its infancy, several criteria may be inappropriate for recent
developments in neuroimaging/neurophysiologic methods. Further, although a number of items
are critical to determining methodological quality of the evidence, all items carry the same
weight. We attempted to minimize this limitation by redefining the criteria for several items,
however is likely not a complete fix. The development of newer methodological quality
evaluation tools would likely improve the systematic review process in this area.

Future Research. To improve methodological quality, we recommend that future
studies: 1) perform cross-validation of the model, 2) consider the minimal clinically important
differences (MCID) of motor recovery outcome measures, and 3) recruit a large enough sample
to provide sufficient statistical power. Further, future studies that employ more sensitive and
specific clinical endpoints coupled with valid neurological biomarkers are more likely to advance
our understanding of the motor recovery process after stroke. Such an approach may lead to the
development of more accurate prediction models than that achieved with the more traditional
broad-based clinical endpoints that have dominated the literature, thus far.

Neurological Biomarkers of Post-Stroke Motor Recovery 54

Conclusion
Heterogeneity of post-stroke brain pathology and motor impairment is a considerable
challenge for the development of accurate prediction models. Accurate prediction of recovery is
critical for determining the best neurorehabilitation protocol that will promote motor recovery
and maximize meaningful outcomes. This focused systematic review found that conventional
structural MRI and combination biomarker types possess the most methodologically robust
evidence to be used for predicting motor recovery after stroke. Further, it is not surprising that
prediction models that used neurological biomarkers along with clinical measures (e.g., Fugl-
Meyer score, age, or chronicity) were more accurate than models that used neurological
biomarkers alone.

Neurological Biomarkers of Post-Stroke Motor Recovery 55

CHAPTER THREE: A COMPARISON OF SEVEN DIFFERENT DTI-DERIVED
ESTIMATES OF CORTICOSPINAL TRACT STRUCTURAL CHARACTERISTICS IN
CHRONIC STROKE SURVIVORS.

Abstract
Background. Different diffusion tensor imaging (DTI)-derived methods have been used to
estimate corticospinal tract (CST) structure in the context of stroke rehabilitation research.
However, there is no gold standard that has consistently been shown to provide the most accurate
estimate of CST structure in chronic stroke survivors. We compared DTI-derived approaches
using three criteria: 1) the method can capture a significant decrease in fractional anisotropy
(FA) of ipsilesional CST compared to FA of contralesional CST; 2) CST FA asymmetry falls
within the normative range between -0.03 and 0.25; 3) significant relationship between CST FA
asymmetry and clinical motor outcome.
Methods. Imaging and behavioral data were obtained from a single-site randomized clinical trial
of stroke rehabilitation. Participants were chronic stroke survivors with mild-to-moderate arm
and hand motor impairment (N=37, average chronicity=3 years). Imaging data were processed
using BrainSuite16a (http://brainsuite.org/), and the Wolf Motor Function Test log mean time
score for distal control items (WMFT-distal) was chosen as the primary motor outcome. We
calculated mean FA for CST of each tract (L, R) using 7 different methods: 1) 3-D individual
tractography-based CST volume, 2) 3-D Template CST volume from a standard white matter
atlas, 3) 3-D Template CST volume generated from the participants’ contralesional CST
tractography, 4) 3-D posterior limb of the internal capsule (PLIC) template volume from a Johns
Hopkins University (JHU) white matter atlas, 5) 3-D cerebral peduncle (CP) template volume
Neurological Biomarkers of Post-Stroke Motor Recovery 56

from a JHU white matter atlas, 6) manually drawn 2-D PLIC region, and 7) manually drawn 2-D
CP region. Separate paired t-tests were used to compare ipsi- and contralesional CST FA for
each method. One-way ANOVA was used to test if any of the seven methods derives different
FA asymmetry. Partial correlation analyses were conducted between each CST FA asymmetry
index and WMFT-distal time score, controlling for age and chronicity. Bonferroni correction was
used for multiple paired T-test comparisons, Post-hoc comparison of ANOVA, and correlation
analyses (corrected significance level = 0.05/7 = 0.0071)
Results. The mean ipsilesional CST FA was significantly lower than the mean contralesional
CST FA for each of the 7 methods. Only two methods (the 3-D individual CST tractography and
the the 3-D template CST from participants) met the normative range criterion for CST FA
asymmetry. Further, CST FA asymmetry from the 3-D individual CST tractography-based
method showed a significant partial correlation with the primary motor outcome (r=0.46,
p=0.005), while CST FA from the other six methods did not.
Discussion. These findings suggest that compared to the six other methods, 3-D individual
tractography-based CST FA asymmetry provides the best estimate of CST structural
characteristics in chronic stroke survivors with mild-to-moderate motor impairment. We
recommend this method for future research seeking to understand brain-behavior mechanisms of
motor recovery in chronic stroke survivors.

Neurological Biomarkers of Post-Stroke Motor Recovery 57

Introduction
The corticospinal tract (CST) delivers motor commands from primary motor and sensory
cortices to the spinal cord through direct (lateral and medial corticospinal tract) and indirect
pathways. As such, the amount of post-stroke damage affecting this descending motor pathway
is a crucial factor that determines the level of motor impairment following stroke (Corbetta et al.,
2015; Jang, 2011). Diffusion tensor imaging (DTI)-derived estimate of CST structural damage to
the CST has been used to predict motor recovery in stroke survivors. DTI-derived metrics of
CST (Koyama et al., 2013; Lindenberg et al., 2012, 2010, Puig et al., 2013, 2011), specifically
fractional anisotropy (FA) asymmetry and the ratio between ipsi- and contra-lesional CST, are
the most frequently used predictor variables in prognostic studies (Lindenberg et al., 2010; Puig
et al., 2013; Stinear et al., 2012). In general, FA of ipsilesional CST is decreased after stroke,
leading to an increase in FA asymmetry (Stinear et al., 2012).
Although DTI-derived estimates of ipsilesional CST is widely used, the specific
estimation method varies considerably (Lindenberg et al., 2010); as such, comparison across
studies is challenging. The four most common regions/volumes of interest that investigators have
used to estimate CST structure are: 1) a 2-dimensional region of interest (2-D ROI) at the
posterior limb of the internal capsule (PLIC); 2) a 2-D ROI at the cerebral peduncle (CP); 3) a 3-
dimensional (3-D) volume of interest (3-D VOI) of CST from individual stroke patient’s CST
tractography; and 4) a 3-D VOI of CST from non-disabled adults’ CST template (Kim and
Winstein, 2017).
For 2-D ROI-based methods, a PLIC or CP ROI is manually drawn on structural images
(Koyama et al., 2012). To perform CST structural estimation using 3-D CST VOI, one can
reconstruct the individual CST of each individual using diffusion tensor-based tractography
Neurological Biomarkers of Post-Stroke Motor Recovery 58

(Lindenberg et al., 2012, 2010; Yu et al., 2009). Template CST tractography is reconstructed
from age-matched non-disabled participants’ DTI data, or an established template standard CST
atlas can be used (Park et al., 2013).
There are advantages and disadvantages of each estimate of CST structure. The benefits
of using 2-D PLIC or CP ROIs are that both ROIs are easily identified on structural images, and
the FA value of these regions has been shown to provide an accurate estimation of CST
structural damage (Kuzu et al., 2012). A decreased FA at these regions is thought to reflect the
Wallerian degeneration that occurs across the entire CST after stroke (Koyama et al., 2012; Kuzu
et al., 2012; Lindenberg et al., 2012). Thus, determination of the DTI-derived metrics at a remote
CST section, such as PLIC or CP, is assumed to represent the degree of degeneration of the
entire CST without underestimation. However, the ROI-based methods may be biased,
particularly when PLIC or CP ROIs are manually drawn (Lindenberg et al., 2010). To control for
possible bias introduced by these manual methods, some researchers have instead used PLIC or
CP sections of CST tractography or established subcortical white matter atlases (Park et al.,
2013).
Compared with 2-D ROI-based methods, individual tractography-based CST estimation
methods can represent the entire CST structure using the subject’s own reconstructed fiber tracts.
However, the individual CST method may not be appropriate for cases in which some if not most
CST fibers cannot be traced using diffusion tensor-based tractography (Cho et al., 2007a,
2007b).
In general, template-based CST estimates are considered a more objective method
compared to ROI-based methods, as it relies on automated processes known to reduce operator-
dependent bias (Park et al. 2003). However, the brain of chronic stroke survivors may present
Neurological Biomarkers of Post-Stroke Motor Recovery 59

with an aberrant ipsilesional CST trajectory due in part to significant subcortical white matter
atrophy; in this case, an estimate of CST structure that relies on template CST volume will likely
be incorrect (Jang, 2011).
Although these different DTI-derived estimates for CST structure are widely used in
stroke rehabilitation research, there is no gold standard which most accurately represents CST
structural characteristics in chronic stroke survivors. Thus, this study aims 1) to determine which
method most accurately estimates CST structural damage, and 2) to examine the degree to which
the measure of CST structural damage correlates with a well-known clinical motor outcome in
chronic stroke survivors with mild-to-moderate motor impairment.
To determine the most accurate estimate of CST structure, each method was evaluated
based on three criteria: 1) the method can capture a significant decrease in ipsilesional CST
fractional anisotropy (FA) compared to the FA of contralesional CST; 2) CST FA asymmetry
range falls within the normative range between -0.03 and 0.25; and 3) a significant relationship
between CST FA asymmetry and clinical motor outcome.
Our expected findings are that the most accurate method for this cohort will be the 3-D
individual tractography-based CST method, and the CST FA asymmetry derived from this
method will result in the strongest correlation with our primary motor outcome. Our hypotheses
are based on two primary assumptions: 2-D ROIs may not represent the entire CST structure, and
3-D template CST VOI may not represent the distorted ipsilesional CST trajectory commonly
seen in the chronic stroke brain.

Neurological Biomarkers of Post-Stroke Motor Recovery 60

Methods
Participants. The clinical and neuroimaging data were from a single-site randomized
trial of stroke rehabilitation conducted in the Motor Behavior and Neurorehabilitation Laboratory
at Division of Biokinesiology and Physical Therapy, University of Southern California
(ClinicalTrials.gov ID: NCT 01749358). The purpose of this clinical trial was to determine the
effect of different doses of therapy on motor outcomes in chronic stroke. Predefined inclusion
and exclusion criteria for the RCT are described in Appendix C. For this project, a total of 37 out
of 42 trial participants’ data met inclusion criteria that required a complete set of baseline clinical
motor outcome scores and DTI neuroimaging. Five data sets were excluded because of missing
neuroimaging or artifacts present on imaging.

Table 3-1. Participants’ characteristics
Characteristics Mean (Standard deviation)
Age [years] 59.43 (12.57)
Sex [Male/Female] 27/20
Chronicity [years] 3.01 (3.10)
Hand dominance [Rt/Lt] 34/3
Affected hemisphere [Rt/Lt] 18/19
Lesion volume [mm
3
] 19,840.89 (37,868.69)
CST-lesion overlap volume [%] 4.59 (5.88)
Fugl Meyer Upper Extremity 43.49 (8.99)
WMFT time score – distal items [log(sec)] 2.20 (1.08)

Clinical motor outcome measure. We utilized an arm-specific clinical motor outcome
measure. Specifically, we employed a subset of laboratory-based Wolf Motor Function Test
(WMFT). WMFT time score is a reliable and valid method to evaluate UE motor performance
Neurological Biomarkers of Post-Stroke Motor Recovery 61

after stroke, particularly in the research environment (J.-H. Lin et al., 2009; Wolf et al., 2001).
This measure includes fifteen timed motor tasks. The timed tasks can be distributed into two task
categories: 1) tasks related to joint-segment movements, and 2) tasks related to integrative
functional movements (Wolf et al., 2001). The joint-segment movement tasks are primarily those
with proximal joint control (e.g., shoulder and elbow), while the integrative functional tasks
require some level of hand dexterity for object manipulation. Given our sample of those with
mild-to-moderate motor impairment, we employed the WMFT-distal time score (WMFT-distal),
the log-transformed average time score for the integrative functional task items (i.e., distal arm
control task items).

MRI acquisition. We used a 3 Tesla GE Signa Excite MRI scanner to acquire high
resolution T1-weighted and diffusion-weighted images with the following acquisition
parameters:

T1-weighted image. A set of coronal scout scans covering the entire brain was acquired
first to define the field-of-view, and then 124 sagittal slices covering the entire brain were
acquired. Next a sagittal anatomic 3-D volumetric study designed to increase spatial resolution
and tissue contrast was acquired using a gradient-echo (SPGR) T-1 weighted series with TR=24,
TE=3.5ms, flip angle=20°, field of view (FoV)=24 cm, and slice thickness=1.2 mm with no
gaps. This procedure was completed in approximately 10 minutes.

Diffusion MRI. The diffusion MRI scan used a single shot spin echo EPI pulse sequence
using the following parameters: TR=10,000 ms, TE=88 ms, 75 axial slices, FOV=256 mm, slice
Neurological Biomarkers of Post-Stroke Motor Recovery 62

thickness=2.0 mm, Matrix = 128 x 128, b-value=1000, 64 diffusion gradient directions. The
diffusion MRI sequence was completed in approximately 10 minutes.

MRI data analysis. The raw DICOM files were converted to NIFTI format using the
MRIcron DICOM to NIFTI function (http://people.cas.sc.edu/rorden/mricron/dcm2nii.html).
BrainSuite software (http://brainsuite.org) was used to process T1-weighted images and DTI.
The MRI process sequence for T1-weighted images includes cortical surface extraction and
surface-volume registration (SVREG). The cortical surface extraction includes skull striping
(Sandor and Leahy, 1997), intensity non-uniformity correction (Shattuck et al., 2001), tissue
classification (Shattuck et al., 2001), inner cortex masking, surface generation, and hemisphere
labeling. This process was automatically performed with pre-determined parameters. A quality
check was performed at each step, and if the result of the step was incorrect due to presence of
the stroke lesion, manual correction was performed. Specifically, results from the tissue
classification and inner cortex masking were frequently incorrect due to the presence of the
lesion. To improve the tissue classification, intensity non-uniformity correction was re-
performed using manual bias field correction software and/or Brainsuite’s bias field correction
function in iterative mode. In addition, manual mask correction was conducted if the inner cortex
mask did not accurately identify the gray/white boundary. This manual correction procedure
improved the result of surface-volume registration in Brainsuite. For quality control of the
cortical extraction process, images were visually checked by the first author (BK) at each step.
Surface-volume registration (the SVReg function in Brainsuite) is an automated process
for co-registration of the T1-weighted image to an atlas brain in which a nonrigid alignment
between the subject and atlas is performed, constrained by mapping of the surface of the cerebral
Neurological Biomarkers of Post-Stroke Motor Recovery 63

cortex of the atlas onto that of the subject (Joshi et al., 2012). We used BrainSuiteAtlas1
(http://brainsuite.org/atlases/), a labeled version of the Colin27 Average Brain (Holmes et al.,
1998). The result of applying SVReg to each volume is to automatically label subcortical and
cortical structures in the subject space. A visual quality check of the anatomical brain label
outputs was performed to assess accuracy of the registration.
For the diffusion MRI preprocessing, we used FSL’s eddy tool
(https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/eddy) to correct eddy currents and movements in diffusion
MRI data (Andersson and Sotiropoulos, 2016). Then, the BrainSuite diffusion pipeline (BDP)
was used to process diffusion MRI data. We co-registered the diffusion MRI data to the
anatomical T1-weighted images using the INVERSION method (Bhushan et al., 2014); this
method includes correction for susceptibility-induced distortions using non-rigid registration.
Diffusion tensors were estimated using a weighted linear least squares method, and we computed
the scalar diffusion parameters: fractional anisotropy (FA), mean diffusivity (MD), radial
diffusivity (RD), and axial diffusivity (AD) based on an eigen-decomposition of the tensors, as
described by Kim et al., 2009. After processing of the diffusion MRI data, all images were
visually checked for quality control.
After the BDP process, deterministic whole brain diffusion tensor-based tractography
was performed using the BrainSuite Diffusion Toolbox. Then we identified the corticospinal
tracts (CST) tractography for each hemisphere. CST was identified as streamlines passing
through cerebral peduncle (CP), pons on the same side, and originating from primary motor
cortex (M1), primary sensory cortex (S1), and supplementary-motor cortex (SMA). Any fibers
passing through the corpus callosum, midline of the pons, and cerebellum were excluded. We
visually inspected each CST tractography for accuracy (Figure 3-1). In detail, we applied three
Neurological Biomarkers of Post-Stroke Motor Recovery 64

steps of filtering to the CST tractography: 1) filter out streamlines to contralateral hemisphere
and to cerebellum; 2) select streamlines passing through CP and pons in the same side; 3) select
streamlines passing through precentral gyrus (M1), postcentral gyrus (S1), and paracentral lobule
(SMA) labels in the same side. If the CST tractography included non-CST streamlines, we
manually removed the non-CST streamlines using ‘TRACK FILTERING’ function in the
BrainSuite graphic user interface. Briefly, we selected a portion of the non-CST streamlines, and
generated a spherical volume of interest in that portion. Streamlines passing through the
spherical volume of interest were then excluded (Figure 3-1-C). If there were multiple non-CST
streamlines, we repeated track filtering as necessary.
DTI-derived estimates of CST structural characteristics. We employed five 3-D VOIs
and two 2-D ROIs to determine the most accurate estimate of CST structural characteristics. The
mean FA for each 2-D ROI or 3-D VOI was calculated.

3-D individual CST tractography. Using the CST tractography of each participant, 3-D
tractography-based CST VOI was used to calculate mean FA for the entire CST of each side.

3-D template CST from standard white matter atlas. Using a 3-D template CST from
natbrainlab (http://www.natbrainlab.co.uk/atlas-maps ) (Thiebaut de Schotten et al., 2011), the
mean FA for template CST VOI was calculated for each side.

3-D template CST from participants’ contralesional CST tractography. To control for
the age factor affecting CST trajectory, we also generated a template CST from the
contralesional CST tractography of each participant. The CST template from the established
Neurological Biomarkers of Post-Stroke Motor Recovery 65

white matter atlas was from non-disabled young adults, and therefore may not be appropriate for
our stroke cohort. For the left CST template, we used left CST tractography from participants
with right hemisphere damage, and vice versa. We determined each CST template using a group-
level overlap of each side CST binary masks at a threshold of 50% of the number of participants
(Park et al., 2013).

3-D PLIC and CP VOIs from standard white matter atlas. To eliminate operator-
dependent bias of manual ROI delineation, we computed the mean FA for both 3-D PLIC and
CP VOIs on each side from the Johns Hopkins University (JHU) DTI-based white matter atlas
(http://neurovault.org/collections/264/ ). The white matter atlas was co-registered to the
BrainSuite atlas space, and transformed to each participant’s space using the 3-D deformation
field generated from SVReg’s mapping between the atlas and the individual T1-weighted image.

2-D manually drawn PLIC and CP ROIs. 2-D PLIC and CP ROIs on each side were
manually drawn on an axial slice of standard T1-weighted BrainSuite atlas image (72
nd
axial
slice [MNI Z coordinate: 0] for PLIC and 54
th
axial slice [MNI Z coordinate: -18] for CP among
181 axial slices). Then the ROIs were transformed to each participant’s space using the 3-D
deformation field generated from SVReg’s mapping between the atlas and the T1-weighted
image. The mean FA for each ROI on each side was calculated.

Neurological Biomarkers of Post-Stroke Motor Recovery 66

Figure 3-1. Visualization of a reconstructed CST tractography from a single participant.
CST tractography is a reconstruction of streamlines passing through sensorimotor cortices
(primary motor, primary sensory, and supplementary motor cortices), cerebral peduncle, and
pons in the same side, and excluding interhemispheric fibers and cerebellum fibers. The different
colors of tractography indicate varying fiber directions. (Red: right ↔ left, Green: anterior ↔
posterior, Blue: caudal ↔ rostral). The color labels on the superior axial slice indicate brain
cortical regions (aqua – left paracentral lobule; purple – left pre-central gyrus; corn flower blue –
left post-central gyrus). (A) A representative illustration of CST tractography in the
Neurological Biomarkers of Post-Stroke Motor Recovery 67

contralesional hemisphere of a chronic stroke survivor with two axial slices showing the
sensorimotor cortices and cerebral peduncle regions of interest. (B) An example of inaccurate
CST tractography in the lesioned hemisphere. a, b, and c indicate non-CST fibers that need to be
excluded manually to improve the accuracy of CST tractography. (C) Manually corrected CST
tractography of Figure 3-1-B. Three spherical volumes of interest (a’, b’, and c’) were used to
exclude fibers passing through these volumes. Detailed description of tractography manual
correction procedure is presented in the Methods section.

Statistical analysis. Separate paired t-tests were used to test for differences between
ipsilesional and contralesional CST FA, with the ipsilesional CST FA expected to be smaller
than contralesional CST FA, for each method. In order to test if the FA of ipsilesional CST is
decreased in the CST area remote from the lesion, we compared the CST FA between
ipsilesional and contralesional sides for each axial slice. This analysis was done for each of the
3-D CST analysis methods, including individual CST tractography, template CST from an atlas,
and template CST from the participants’ contralesional CST tractography. For the slice-by-slice
comparison, the FA images and CST masks were transformed to BrainSuite atlas standard space.
Following this transformation, mean FA for each axial slice for each side was then calculated.
Separate paired t-tests were used to test for differences in the FA between ipsilesional and
contralesional sides for each axial slice.
To test if CST FA asymmetry is different among different methods, a one-way ANOVA
was used with method being the repeated measure. We computed FA asymmetry index
[(contralesional CST FA – ipsilesional CST FA) / (contralesional CST FA + ipsilesional CST
FA)] in order to control for inter-individual variability in FA. Further, we examined if the range
Neurological Biomarkers of Post-Stroke Motor Recovery 68

of CST FA asymmetry from each method is within the normative range for this cohort of stroke
survivors (between -0.03 and 0.25). An estimate of the normative range of CST FA asymmetry
was based on previous studies that reported the CST FA values for a similar stroke population as
well as age-matched non-disabled controls (Lindenberg et al., 2012; Puig et al., 2013; Stinear et
al., 2007)
Lastly, a partial correlation analyses was used to determine which method quantified the
CST that resulted in the strongest relationship with our primary motor outcome score—the
WMFT-distal time score. Partial correlation analyses were used to control for age and stroke
chronicity.
Bonferroni correction was used to account for the multiple t-tests, the ANOVA post-hoc
multiple comparison test, and the multiple correlation analyses. The corrected significance level
was set at 0.0071 (alpha = 0.05/7). For the slice by slice analysis for the 3-D CST methods, the
corrected significance level was set at 0.000276 (alpha = 0.05/181) as we had 181 axial slices.

Neurological Biomarkers of Post-Stroke Motor Recovery 69

Results
Characteristics of participants. Demographic characteristics and clinical motor
outcome measures of our stroke participants indicate that they were chronic stroke survivors with
mild-to-moderate motor impairment. Our stroke participants’ baseline demographics and clinical
motor outcomes are summarized in Table 3-1. Mean age of the participants was 59.43 ± 12.57
years old, ranged from 32 to 80 years old. Participants’ mean chronicity was 3.01 ± 3.10 years
after onset, ranged from 6 months to 14 years after onset. Further, participants had mild to
moderate structural damage on the ipsilesional CST: the range of CST-lesion overlap volume
was from 0% to 25% of the entire CST volume. The range of Upper Extremity Fugl-Meyer was
from 25 to 58 out of 66, and the range of modified WMFT time score was from 0.82 to 4.71
second[log]).

Difference in CST FA between Ipsilesional and Contralesional tracts. All seven
methods showed that the ipsilesional CST FA was lesser than the contralesional CST FA (Table
3-2 & Figure 3-2). The slice by slice analysis of 3-D individual CST tractography showed that in
group level, the FA of ipsilesional CST was significantly decreased compared to the FA of
contralesional CST only in the most damaged axial slices. On the other hand, the slice by slice
analysis of two 3-D template CST methods showed that the FA of ipsilesional CST was
significantly decreased in a majority of axial slices (Figure 3-3 A – C). The individual level slice
by slice analysis of individual CST tractography revealed that only 15 of 37 (40%) participants
had lower FA in the remote area to the CST-lesion junction, while the other ~60% showed lower
FA of CST near the CST-lesion intersection only (Figure 3-3 C and D).

Neurological Biomarkers of Post-Stroke Motor Recovery 70

Table 3-2. Within-method comparison of ipsilesional and contralesional CST FA.

FA of
ipsilesional CST
FA of
contralesional CST
t-stat P-value
Individual CST
Tractography
0.41 0.47 -10.50 <.001***
Template CST
from atlas
0.44 0.53 -7.01 <.001***
Template CST from
participants’ contra-
lesional hemisphere
0.30 0.33 -10.96 <.001***
3-D PLIC 0.38 0.44 -6.25 <.001***
3-D CP 0.40 0.42 -3.70 <.001***
2-D PLIC 0.41 0.47 -5.87 <.001***
2-D CP 0.42 0.47 -6.24 <.001***
(***p < .001 with Bonferroni correction)
Neurological Biomarkers of Post-Stroke Motor Recovery 71

Figure 3-2. Difference in FA between ipsi- and contra-lesional CSTs from different
methods. (A) 3D individual tractography. (B) 3D atlas template. (C) 3D subject template. (D)
3D PLIC. (E) 3D CP. (F) 2D PLIC. (G) 2D CP (***P < .001).

Neurological Biomarkers of Post-Stroke Motor Recovery 72

Figure 3-3. Comparison of FA between ipsi- and contra-lesional CST in axial slices. (A)
Group difference in FA between ipsi- and contra-lesional CST in axial slices from 3-D individual
CST tractography-based method. Only most damaged axial slices showed significant difference
in FA. (B) Group difference in FA between ipsi- and contra-lesional CST in axial slices from 3-
D template atlas CST-based method. (C) Group difference in FA between ipsi- and contra-
lesional CST in axial slices from 3-D template CST from participants’ contralesional CST
tractography method. (D) An example from a participant who showed lower FA of ipsilesional
CST than contralesional CST only near the CST-lesion intersection. (E) An example from a
participant who showed lower FA of ipsilesional than contralesional CST also in remote areas to
CST-lesion intersection. Blue line – contralesional CST FA; Red line – ipsilesional CST FA;
Black line – number of lesion voxels (right y-axis). Whiskers indicate standard error. Green
vertical dashed lines indicate the Cerebral Peduncle (CP) and Posterior Limb of the Internal
Capsule (PLIC) sections of CST. The black dots on red line indicate the significant difference in
FA between ipsi- and contra-lesional sides (p < .000276).

Neurological Biomarkers of Post-Stroke Motor Recovery 73

Difference in CST FA asymmetry among different methods. There was a statistically
significant difference in CST FA asymmetry among the different methods (F stat = 4.88, p
<0.001) (Table 3-3, Figure 3-4). Post-hoc analysis revealed that FA asymmetry from the 3-D CP
method was significantly smaller than that from the CST atlas template, 3-D PLIC, and 2-D
PLIC (Figure 3-4). Only the CST FA asymmetry for the 3-D individual CST tractography and 3-
D subject Template CST methods ranged within the normative CST FA asymmetry range for this
population. Maximal CST FA asymmetry from three of the seven methods (3-D CST atlas
template, 3-D PLIC, and 2-D PLIC) was equal to or greater than 0.25. Additionally, the
minimum CST FA asymmetry from the 2-D CP and 3-D CP methods was equal to or below -
0.03 (Figure 3-4, Table 3-4).

Table 3-3. ANOVA of FA Asymmetry.
Source SS df MS F-stat P value
Methods 0.11 6 0.018 4.88 <.001***
Error 0.94 252 0.004
Total 1.05 258
(***p < .001 with Bonferroni correction)

Neurological Biomarkers of Post-Stroke Motor Recovery 74

Table 3-4. Summary of Statistics of FA Asymmetry. Bold indicates the values that are
considered out of normative FA asymmetry boundary for our population.
Methods Min Max Mean SD
3-D Tractography
CST VOI
-0.02 0.14 0.06 0.04
3-D Template CST
VOI
-0.03 0.29 0.09 0.08
3-D Subject
Template CST
VOI
-0.01 0.11 0.05 0.03
3-D PLIC VOI -0.06 0.28 0.08 0.08
3-D CP VOI -0.07 0.11 0.02 0.04
2-D PLIC ROI -0.05 0.25 0.07 0.08
2-D CP ROI -0.13 0.21 0.06 0.06

Figure 3-4. Comparison of FA asymmetry from different methods. Red horizontal dash line
indicates the normative range of CST FA asymmetry in this cohort of chronic stroke survivors
with mild to moderate motor impairment (*p < .05, **p < .01, ***p < .001).

Neurological Biomarkers of Post-Stroke Motor Recovery 75

Relationships between neuroimaging CST variables and WMFT-distal time score.
There was a statistically significant partial correlation between 3-D individual CST tractography-
based FA asymmetry and modified WMFT log-transformed mean time score for distal control
tasks. No other significant correlations were found for any of the other DTI-derived CST
methods (Figure 3-5, Table 3-5).

Table 3-5. Partial correlation coefficients between DTI-derived CST FA asymmetries and
mWMFT time score.

3-D
Tractography
CST VOI
3-D Template
CST VOI
3-D Subject
Template CST
VOI
3-D
PLIC
VOI
3-D
CP
VOI
2-D
PLIC
ROI
2-D
CP
ROI
mWMFT 0.46* 0.05 0.18 0.14 -0.02 0.02 -0.19
(*p < .05 with Bonferroni correction)

Figure 3-5. Partial correlation between individual tractography-based CST FA asymmetry
and mWMFT time score, controlled for age and chronicity.
Neurological Biomarkers of Post-Stroke Motor Recovery 76

Discussion
Our results provide robust evidence that among seven different DTI-derived estimates of
CST, 3-D individual CST tractography-based FA asymmetry is the best measure for
characterizing CST structure in this cohort of chronic stroke survivors with mild-to-moderate
impairment. In the following discussion, we provide an explanation for this finding in light of the
three criteria we used to evaluate the seven different methods:
First, all seven methods were able to detect the decreased fractional anisotropy in the
ipsilesional CST compared to the contralesional CST. This result indicates that all seven methods
met the first criteria for the appropriate CST structural estimates.
Second, based on previous reports (Lindenberg et al., 2012; Puig et al., 2013; Stinear et
al., 2007), participants with motor impairment in the mild to moderate range, are expected to
have CST FA asymmetry that is less than 0.25 and greater than -0.03. This lower limit is an
estimation from previous studies that reported the FA of left and right CST in non-disabled
adults (Lindenberg et al., 2012) and CST FA ratio between hemispheres in mildly impaired
stroke survivors (Puig et al., 2013). CST FA asymmetry greater than 0.25 indicates that the
individual has no recovery potential, which implies severe stroke damage on the descending
motor pathways (Stinear et al., 2007). Only two methods, 3-D individual CST tractography and
3-D subject Template CST met our expected range boundary for this stroke cohort (Table 3-4).
The other five methods either under- or overestimated CST FA asymmetry. Therefore, with
respect to FA asymmetry boundary range, the 3-D individual CST tractography-based method
and the 3-D subject Template CST method appear to be the most accurate methods for
quantification of CST structural characteristics in our population.
Neurological Biomarkers of Post-Stroke Motor Recovery 77

Third, only the 3-D individual CST tractography-based FA asymmetry was significantly
correlated with our clinical motor behavior measure – the modified WMFT time score. We used
partial correlation analysis to control for age and chronicity. It has been shown that age and
chronicity has an impact on the relationship between brain biomarker and clinical motor
outcomes (Cramer, 2008; Lindenberg et al., 2012; Stinear et al., 2007). We also conducted the
correlation analysis between CST FA asymmetry and the modified WMFT time score (data not
shown). The result from correlation analysis was similar to the result from partial correlation
analysis. This indicates that the age and chronicity did not affect the relationship between the
CST FA asymmetry and motor outcome measure in this cohort of stroke survivors. It has been
shown that the amount of CST structural damage is a major factor associated with motor deficits
after stroke (Corbetta et al., 2015). Therefore, we expected the FA asymmetry index from an
accurate quantification of CST damage would be significantly correlated with motor
performance. We believe that the FA asymmetry index from the other six methods were not the
most accurate representation of CST structural damage and therefore were not found to correlate
significantly with WMFT-distal time score. Further, the fact that our results showed that a
measure of CST structural impairment is associated with motor performance may be due to the
vital role of CST in delivering precise motor commands for integrative functional upper
extremity movements, especially those for distal arm and hand manipulation. Disruption of CST
structure can disharmonize the motor commands leading to poor distal arm motor control (Dong
et al., 2006; van Kuijk et al., 2009). Therefore, clinical motor outcome measures related to distal
arm and hand function may be more appropriate than general motor impairment measures, such
as the Upper Extremity Fugl-Meyer, to determine the relationship between CST structure and
motor behavior in chronic stroke survivors with mild-to-moderate motor impairment.
Neurological Biomarkers of Post-Stroke Motor Recovery 78

In addition to the results that pertain to the three criteria, there are several reasons why
we consider the individual tractography-based CST quantification to be the most appropriate
estimate of CST structural characteristics.
Individual tractography-based CST quantification accounts for a distorted ipsilesional
CST trajectory, whereas template-based CST quantification is unable to do so (Figure 3-6). The
distorted ipsilesional CST trajectory is easily observed in chronic stroke survivors in large part
due to significant subcortical white matter atrophy (Jang, 2011). Our tensor-based tractography
method was able to accurately trace the CST streamlines in the ipsi-lesional hemisphere for all
participants. However, this was not the case using the template method. Primarily due to white
matter atrophy and enlarged ipsilesional lateral ventricle, the template-based method inaccurately
included the lateral ventricle in 12 of 37 participants (i.e. over 1/3
rd
of the sample). Given that the
FA value is relatively low in the ventricle (i.e., more isotropic diffusion of water molecules in the
ventricle), this may cause an overestimation of the CST structural damage. For this reason,
individual tractography-based CST quantification more accurately represents the ipsilesional
CST structural characteristics compared with template-based methods.
Further, 3-D individual tractography-based methods represent the entire CST. The 2-D
ROI-based methods, frequently used in the literature, do not represent the entire CST. The
assumption behind these 2-D ROI-based methods is that Wallerian degeneration causes the
decreased FA in the remote area to the CST-lesion junction, such as PLIC or CP. The FA of
ipsilesional PLIC or CP is assumed to reflect the structural integrity of the entire CST. The slice-
by-slice analysis of 3-D individual CST tractography demonstrated that the decreased FA of
ipsilesional CST is observed only in the most damaged axial slices (Figure 3-3A). Further, we
found that only 15 of 37 (40%) participants had decreased FA in the remote area to the CST-
Neurological Biomarkers of Post-Stroke Motor Recovery 79

lesion junction, such as cerebral peduncle and posterior limb of the internal capsule (Figure 3-
3D). The other ~60% showed decreased FA of CST near the lesional area only (Figure 3-3E).
Thus, use of 3-D VOI or 2-D ROI of PLIC or CP may not capture the structural integrity of the
entire CST for those who have no decrease in FA in these remote areas (e.g., Figure 3-3D).

Figure 3-6. 3-D Template CST VOI and 3-D individual CST VOI from tractography on the
T1-weighted image. Green indicates CST mask from diffusion tensor-based tractography from
one participant, and red indicates template CST mask registered to subject space. (A) A coronal
view showing the overall trajectories of CST tractography and CST atlas template. (B) An axial
slice showing trajectories of the tractography-based CST mask and the atlas template-based CST
mask.

Finally, individual CST tractography processing was automated with minimal user bias
for computing the DTI-derived metrics. Although previous studies reported operator-dependent
bias of CST tractography of stroke patients (Jang, 2011; Kwon et al., 2011), our CST
tractography method using BrainSuite software was automated to a large extent with minimal to
no operator manual correction necessary. In addition, the cortical regions for identifying CST
Neurological Biomarkers of Post-Stroke Motor Recovery 80

were automatically labeled by the software, and subcortical structures, such as PLIC, CP and
Pons, were manually drawn on a standard atlas and transformed to subject space to reduce inter-
subject variability in ROIs. Thus, individual tractography-based CST quantification using
BrainSuite is an objective method for determining CST structural characteristics with minimal
operator-dependent bias.
There has been only one previous work that compared different standard DTI-derived
estimates of CST structure. Our results are inconsistent with the previous work by Park et al.,
(2013). Park and colleagues showed that different methods had similar correlation with motor
ability, while our study showed only the individual tractography-bsed method has significant
correlation with a motor outcome measure. There are several factors that may result in the
inconcistent results between our work and the previous work: 1) difference in cohort of stroke
survivors, 2) difference in DTI analysis software, 3) difference in CST template, and 4)
difference in motor outcome measures.
First, our study recruited chornic stroke survivors with mild-to-moderate severity, while
previous work was conducted with stroke surviors with broader range of severity and chronicity.
The difference in cohort of stroke survivors may be a main reason of the inconsistent results
between our work and previous work. Second, we used BrainSuite software to process MRI data,
while Park and colleagues used Statistical Parmetric Mapping software (SPM, Wellcome Dept.
of Cognitive Neurology, London, UK). BrainSuite software uses different co-registration
algorithms and different tractography methods than does SPM. This difference in software may
account for the inconsistent results between the two studies. Third, the previous work used the
age-matched control participant’s CST tractography as their template, while our study used a
template from standard atlas and another template generated from the contralesional hemisphere
Neurological Biomarkers of Post-Stroke Motor Recovery 81

of our study participants. We assume that the contralesional CST structure is intact, and thus the
template CST volume from contralesional CST tractography may be a better alternate for the
template CST than that from age-matched non-disabled controls. Lastly, we utilized the WMFT-
distal time score to represent the motor behavior of our participants. On the other hand, the
previous work employed a composite value of different motor outcome measures derived from
principal component analysis. Differences in clinical motor outcome measures between the two
studies may have influenced the correlations between DTI-derived CST metrics and motor
ability scores.
One limitation concerns the possibility of error in the spatial registration of stroke brain
structural images. Distortions in the registered structural images could be introduced in case with
enlarged ipsilesional lateral ventricle and image intensity changes in lesional and surrounding
areas (Crinion et al., 2007; Feng et al., 2015). The consequence of the inaccurate spatial
registration is that the template CST-derived estimates are inaccurate. Therefore, four of the
seven estimate methods (3-D template CST from atlas, 3-D template CST from subjects, 3-D
template PLIC and 3-D template PC) are directly impacted by this limitation. Another limitation
is that our findings may not generalize to the wider stroke population. Given that we included
chronic stroke survivors with mild-to-moderate severity, our findings are not applicable to stroke
survivors at any stage of recovery with severe motor impairment, as tractography is limited in the
presence of severe CST damage (Jang, 2011). Lastly, while our sample size is relatively large
compared with previous DTI studies, ideally, we would need a larger sample to replicate and
further validate our findings.

Neurological Biomarkers of Post-Stroke Motor Recovery 82

Conclusion
A systematic comparison of seven different DTI-derived estimates of CST structure
found that 3-D individual tractography-based CST FA asymmetry is the most accurate estimate.
The process of individual tractography-based CST FA computation using BrainSuite software is
semi-automatic with minimal manual correction needed. As such, we recommend this method to
highlight critical brain-behavior mechanisms for future stroke rehabilitation and recovery
research.

Neurological Biomarkers of Post-Stroke Motor Recovery 83

CHAPTER FOUR: CAN A DTI-DERIVED CST BIOMARKER BE USED TO PREDICT
MOTOR IMPROVEMENT IN CHRONIC STROKE SURVIVORS?

Abstract
Background. Prediction of motor recovery in chronic stroke survivors is essential to improve the
efficacy of stroke rehabilitation. Corticospinal tract (CST) microstructure derived from diffusion
tensor imaging (DTI) is a significant predictor of motor recovery in acute and subacute post-
stroke individuals. However, use of this measure to predict motor improvement in chronic stroke
survivors has not been fully investigated. Conversely, it is unknown whether motor improvement
in chronic stroke survivors during rehabilitation is associated with brain structural changes.
Purpose. This study aims to determine the feasibility of prediction models using DTI-derived
CST fractional anisotropy (FA) asymmetry to estimate motor improvement in chronic stroke
survivors, and to examine if brain structural changes are correlated with motor improvement in
this population.
Methods. This study is part of a longitudinal phase-I stroke rehabilitation clinical trial in chronic
individuals (ClinicalTrials.gov ID: NCT 01749358). Chronic stroke survivors with mild-to-
moderate motor impairment participated (N=37, average chronicity = 3.01±3.08 years, upper
extremity Fugl-Meyer score range [UEFM] 19 to 60). They completed clinical motor behavior
and neuroimaging assessments at two different time-points 3 months apart. Some participants
received upper extremity motor therapy during the 3-month period with different doses (15, 30,
and 60 hours), while others received no therapy. Upper extremity motor behavior was assessed
using the Wolf Motor Function Test (WMFT). We calculated log mean time score for both all 15
items and 8 distal control items (WMFT-distal). Clinically important difference in WMFT was
Neurological Biomarkers of Post-Stroke Motor Recovery 84

defined as 1.5 second change. Neuroimaging assessments were completed with T1-weighted
imaging and DTI. Three different brain structural biomarkers were employed to estimate the
brain structural characteristics: 1) individual tractography-based CST FA asymmetry; 2) lateral
ventricle volume asymmetry; 3) mean FA for stroke lesion. We tested for changes in brain
structural characteristics and motor behavior following a 3-month period. Additionally,
regression analyses were used to examine the predictive value of CST FA asymmetry for
changes in motor behavior. Lasso regression analyses were used to test if additive predictor
variables improved the accuracy of the model. Dependent variables of regression models were 1)
Δ WMFT-distal as percent of baseline WMFT-distal, and 2) binary CID of WMFT.
Results. There was no dose effect on motor improvement. There was a significant decrease in
the WMFT-distal time score for all participants regardless of group assignment (p < .001), with
no significant brain structural changes. Baseline CST FA asymmetry explained 19% of the
variance in Δ WMFT-distal. The combination of CST FA asymmetry, age, and UEFM increased
the variance explained to 23%. Further, the baseline CST FA asymmetry also predicted the CID
in the WMFT (concordance index = 0.62, p < .01). A model with Age, UEFM, and CST FA
asymmetry from a lasso logistic regression also better predicted the CID in WMFT (concordance
index = 0.68, p < .01).
Discussion. Baseline CST FA asymmetry was a significant predictor of improvement in motor
behavior, even though brain structural changes were not associated with motor improvement in
this population. Further, lasso regression with cross-validation can be used to develop a more
accurate and robust prediction model with significant imaging, demographic, and clinical motor
behavior variables.

Neurological Biomarkers of Post-Stroke Motor Recovery 85

Introduction
Stroke is a pathologic cerebrovascular condition that results in long-term disability
(Benjamin et al., 2017). An accurate prognosis of motor recovery after stroke is essential to
improve the efficacy of rehabilitation interventions, and to reduce the indirect and direct costs of
stroke rehabilitation (Stinear et al., 2017). Heterogeneity of stroke pathology is the major barrier
to estimate the potential for recovery, to set achievable goals, and to plan an ideal treatment
accordingly (Boyd et al., 2017).
Brain neurological biomarkers are defined as surrogate imaging/neurophysiological
indicators of brain pathology (Bernhardt et al., 2016; Boyd et al., 2017). They are significant
predictors of motor recovery as well as response to therapy during both early and late stages of
stroke (Bernhardt et al., 2016; Boyd et al., 2017). These biomarkers have higher predictive
values for motor recovery than clinical behavioral measures (Kim and Winstein, 2017). Further,
these biomarkers could potentially be used to uncover neural mechanisms underlying motor
recovery after stroke (Bernhardt et al., 2016). Numerous brain imaging biomarkers have been
suggested since the advance in non-invasive in vivo neuroimaging techniques. Ipsilesional
primary motor cortex activation level, structural and functional integrity of the corticospinal tract
(CST), and brain structural/functional connectivity have been widely used as essential brain
biomarkers to predict acute post-stroke motor recovery (Boyd et al., 2017; Kim and Winstein,
2017). While lesion size and location were considered the most important brain structural
biomarkers for stroke prognosis in the past, recent studies have supported that the CST
microstructural characteristics derived from diffusion tensor imaging (DTI) are more accurate
than lesion size and location (Kim and Winstein, 2017; Puig et al., 2017; Stinear et al., 2014),
Given that damage to the descending cortico-spinal motor pathways is the most crucial factor
Neurological Biomarkers of Post-Stroke Motor Recovery 86

that determines motor impairment in stroke survivors (Corbetta et al., 2015), it follows that the
DTI-derived CST biomarker would be a significant predictor of motor recovery after stroke.
The majority of previous studies have employed diffusion tensor imaging (DTI)-derived
CST structural biomarkers to predict motor recovery in the early phases post-stroke (Kim and
Winstein, 2017). Only a few underpowered studies have been conducted to determine the
predictive values of DTI-derived CST biomarkers on motor improvement or treatment response
in chronic stroke survivors. Importantly, CST biomarkers have been more frequently used to
predict motor recovery in individuals with moderate to severe stroke (Stinear et al., 2007; Stinear
et al., 2012), while most clinical trials for stroke rehabilitation have targeted those with mild-to-
moderate motor impairment. Further, recent systematic reviews have suggested that combining
DTI-derived CST biomarkers and clinical scores better predict motor improvement and treatment
response in chronic survivors (Kim and Winstein, 2017; Puig et al., 2017). However, we lack
high-quality prognostic studies to validate if a multimodal prediction model is more accurate
than a model using CST biomarker alone to predict motor improvement in chronic stroke
survivors with mild-to-moderate severity.
In addition, it has yet to be determined whether brain structural changes even occur as a
result of therapy during the chronic stage of stroke. As such, it is not well studied whether brain
structural changes are associated with improvement in motor behavior. As mentioned above,
CST microstructural changes would be an essential factor that determines motor improvement
with therapy. In fact, this is well established in animal research and clinical studies targeting
early phase of stroke (Hermann and Chopp, 2012; Ramos-Cejudo et al., 2015; Sterr et al., 2014),
while it is unknown if changes in CST structure is associated with motor improvement in chronic
stroke survivors.
Neurological Biomarkers of Post-Stroke Motor Recovery 87

This study primarily aims to determine if a DTI-derived CST biomarker can predict the
improvement in motor behavior after a 3-month period in chronic stroke survivors with mild-to-
moderate motor impairment. Secondly, we examine if combining the CST biomarker and clinical
variables, such as the baseline upper extremity Fugl-Meyer score, will improve the accuracy of a
prediction model. Lastly, we investigate if brain structural changes are occurring in chronic
stroke survivors, and if CST structural changes are associated with upper extremity motor
improvement.
We hypothesized that the baseline DTI-derived CST estimate will predict changes in a
motor outcome measure. Further, we also hypothesized that a mulitimodal prediction model
using baseline CST biomarker, a clinical motor impairment score, and demographic variables
will be more accurate than a model using the CST biomarker alone. In addition to the changes in
motor outcome measures, we investigated predictive values of CST biomarker for a clinically
important difference (CID) in a motor outcome measure. We hypothesized that the CST structure
will predict if the stroke survivor will have CID in a motor outcome after a 3-month period, and
whether combining the CST structural biomarker and clinical score will improve the accuracy of
prediction of the CID. Lastly, we expect CST microstructural changes after a 3-month period,
without changes in lesion area and subcortical white matter volumes, given that previous studies
have shown no structural changes in either lesion and white matter volume in chronic stroke.
Further, individuals with greater improvement in CST microstructure will show greater motor
improvement after a 3-month period.

Neurological Biomarkers of Post-Stroke Motor Recovery 88

Methods
Participants. The clinical and neuroimaging data were from a single-site randomized
trial of stroke rehabilitation conducted in the Motor Behavior and Neurorehabilitation Laboratory
at Division of Biokinesiology and Physical Therapy, University of Southern California
(ClinicalTrials.gov ID: NCT 01749358). A total of 37 participants who completed baseline and
post-intervention clinical and neuroimaging assessments were included in this study. As one
participant’s second time-point imaging data were missing, only 36 participants were included in
the comparison of brain structures between two time-points. Details of participant inclusion and
exclusion criteria are described in Appendix C.

The experimental procedure of Dose Optimization for Stroke Rehabilitation
(DOSE) clinical trial. Baseline (time point 1, T1) and post-therapy (time point 2, T2)
evaluations included the clinical tests and MRI scans. There was a 3-month period between T1
and T2. Participants were randomly assigned into one of four study groups that varied in total
number of hours of UE motor therapy (1 active monitoring control group and three treatment
groups). However, for the current study, the four dose groups were combined, given that our
purpose was not to test the effect of therapy dose on brain structural changes and motor
improvement.
Details of the upper extremity (UE) motor therapy provided to participants in the
treatment groups is described in the primary outcome paper for the DOSE clinical trial (paper in
preparation). Briefly, the therapy program used was the Accelerated Skill Acquisition Program
(ASAP). ASAP is a patient-centered task-oriented upper limb motor training program (Winstein
Neurological Biomarkers of Post-Stroke Motor Recovery 89

et al., 2014, 2016). For the active-monitoring control group, participants underwent the same
evaluations by an equivalent time interval (about 3 months) since no therapy was provided.

Clinical motor outcome measure. As described in Study 2 (Chapter 3), we utilized an
arm-specific clinical motor outcome measure: Wolf Motor Function Test (WMFT). Given our
sample of those with mild-to-moderate motor impairment, we employed the WMFT-distal time
score (WMFT-distal). WMFT-distal is the log-transformed average time score for the integrative
functional task items (i.e., distal arm control task items).

MRI acquisition. MRI acquisition procedure is described in Study 2 (Chapter 3).

MRI data analysis. Details of MRI data analysis is described in Study 2 (Chapter 3).
Briefly, the T1-weighted and diffusion imaging data were processed using BrainSuite. The
imaging data process sequence includes data format conversion, cortical surface extraction and
surface-volume registration (SVREG) of T1-weighted images, eddy current correction and co-
registration of diffusion-weighted images, and deterministic whole brain diffusion tensor-based
tractography. For quality control of the brain imaging process, images were visually checked by
the first author (BK) at each step.
We identified the corticospinal tracts (CST) tractography for each hemisphere. CST was
identified as streamlines passing through the cerebral peduncle (CP), pons on the same side, and
originating from primary motor cortex (M1), primary sensory cortex (S1), or supplementary
motor area (SMA). Any fibers passing through the corpus callosum, the midline of the pons, and
Neurological Biomarkers of Post-Stroke Motor Recovery 90

cerebellum were excluded. We visually inspected each CST tractography for accuracy (Figure 3-
1).

DTI-based CST microstructure quantification. We utilized a 3-dimensional individual
CST tractography-based quantification method. This method was chosen based on results from
Study 2 (Chapter 3). Using the CST tractography of each participant, 3-D tractography-based
CST volume of interest (VOI) was used to calculate mean FA for the entire CST of each side.
We computed the FA asymmetry index [(contralesional CST FA – ipsilesional CST FA) /
(contralesional CST FA + ipsilesional CST FA)] in order to control for inter-individual
variability in FA.

CST-lesion overlap volume. We calculated the CST-lesion overlap volume using a 3-D
template CST from natbrainlab (http://www.natbrainlab.co.uk/atlas-maps ) (Thiebaut de
Schotten et al., 2011). Binary lesion masks were drawn on each participant’s T1-weighted image.
Then the 3-D template CST images were transformed to each participant’s T1-weighted image
space. The number of overlap voxels between binary template CST mask and lesion mask was
calculated (Zhu et al., 2010). The unit of CST-lesion overlap volume was mm
3
, given that the
voxel size was 1×1×1 mm
3
.

Lateral Ventricle Volume Asymmetry (LVA). We calculated the LVA between
hemispheres, which represents the degree of subcortical white matter atrophy. Previous studies
have shown that the ventricle-to-brain ratio (VBR) is a valid measure of subcortical white matter
atrophy in people with neurologic disorders, such as Alzheimer’s disease, schizophrenia, stroke,
Neurological Biomarkers of Post-Stroke Motor Recovery 91

etc (Breteler et al., 1994; Sachdev, 2004). In this study, we utilized the LVA instead of VBR, as
stroke affects mainly one hemisphere leading to asymmetry of subcortical white matter volume
between hemispheres (Rickards et al., 2012). Further, a previous study has shown a significant
linear relationship between lateral ventricle volume and clinical motor scores (Liew et al., 2017).
A mask for each left and right lateral ventricle was manually drawn on the participant’s T1-
weighted image in the subject space using MRIcron Draw toolbox
(http://people.cas.sc.edu/rorden/mricron/). LVA was then calculated [(ipsilesional Ventricle –
contralesional Ventricle) / (ipsilesional Ventricle + contralesional Ventricle)].

Lesion FA. To confirm that there was no structural change in stroke lesion area, we
computed mean FA for the lesion. The binary lesion mask was overlapped on the participant’s
FA image, then the mean FA for the lesion area was calculated. We utilized the BrainSuite Data
Processing Matlab toolbox (http://neuroimage.usc.edu/neuro/Resources/BST_SVReg_Utilities).

Statistical analysis.
Change in motor behavior. A Wilcoxon signed rank test was used to test if there was a
statistically significant change in WMFT-distal time score after a 3-month period. We used a
nonparametric statistical test given that the WMFT-distal time scores in both time-points were
not normally distributed based on one-sample Kolmogorov-Smirnov test for normal distribution
(Massey, 1951). Further, we examined if each participant had a clinically important difference
(CID) in WMFT.

Neurological Biomarkers of Post-Stroke Motor Recovery 92

Linear relationship between baseline CST biomarker and change in motor behavior.
Simple linear regression analysis was used to test if the baseline CST FA asymmetry can predict
changes in motor behavior. Predictor variable was CST FA asymmetry, and the dependent
variable was Δ WMFT-distal (ΔWMFT-distal).

Regularization and selection of best set of variables for predicting change in motor
behavior. We utilized LASSO (least absolute shrinkage and selection operator) linear regression
with leave-one-out cross-validation to regulate and select the best set of variables, and to
enhance prediction accuracy. Predictor variables included age, chronicity, baseline upper
extremity Fugl-Meyer score (UEFM), CST FA asymmetry, lesion FA, lateral ventricle volume
asymmetry (LVA), and CST-lesion overlap volume. The dependent variable was ΔWMFT-distal.
We utilized bootstrapping to calculate mean and 95% confidence interval of beta for each
selected predictor variable. We generated 1,000 bootstrap samples of randomly selected from N
= 37 participants’ data. A model that generates the minimal mean square error of LASSO fit was
chosen for each bootstrap sample. Using mean beta for each predictor variable, we calculated
predicted ΔWMFT-distal. Then we calculated the explained variance in actual ΔWMFT-distal by
the predicted ΔWMFT-distal. Specifically, we identified the most frequently selected lambda
from bootstrapping, and models with the most frequently selected lambda were used to calculate
mean and 95% confidence interval of beta for each surviving variable. Among 1,000 best lasso
regression models, there were 62 models with the most frequently selected lambda (Appendix
D).

Neurological Biomarkers of Post-Stroke Motor Recovery 93

Baseline CST biomarker as a predictor of clinically important difference in motor
behavior. We utilized simple logistic regression analysis to examine if the baseline CST FA
asymmetry can predict whether a clinically important difference will occur over a 3-month
period. We also used an independent t-test to compare baseline CST FA asymmetry between
participants with CID and participants without CID.
Regularization and selection of best set of variables for predicting clinically important
difference in motor behavior. We utilized LASSO logistic regression analysis with leave-one-
out cross-validation to select the best set of variables and to enhance prediction accuracy. A
model that generates either the minimal deviance or minimal deviance + 1 standard error was
chosen. Among those two models, the one with the least number of predictors was selected.
We also compared the multimodal model from LASSO logistic regression analysis with
the simple logistic model (Baseline CST FA asymmetry alone) to determine if the multimodal
model is more accurate than the simple model. The two models were compared using the C-
index (concordance index). The C-index is between 0 and 1, which indicates how well the model
classifies between different responses (in this case, the different responses were achieving CID
or not). A value below 0.5 indicates a very poor model, a value equal to 0.5 implies the model is
no better than random guessing. A value greater than 0.7 indicates the model is good (Hosmer
and Lemeshow, 2000).

Changes in brain structural biomarkers (CST FA asymmetry, lateral ventricle volume
asymmetry, and lesion FA). Separate Wilcoxon signed rank tests were utilized to compare all
three brain structural biomarkers between two different time-points. Bonferroni correction was
Neurological Biomarkers of Post-Stroke Motor Recovery 94

used for multiple comparisons. The corrected significance level was set at 0.0167 (alpha =
0.05/3).
We utilized 2×2 repeated measures analysis of the variance (2×2 RM ANOVA) to
examine if there was a change in either ipsilesional or contralesional CST FA between two time-
points (2 Hemispheres × 2 time-points).
Further, we utilized the 2×2 RM ANOVA to test if there was a change in either
ipsilesional or contralesional lateral ventricle volume (2 hemispheres × 2 time-points). The
corrected significance level was set at 0.025 (alpha = 0.05/2).

Relationship between brain structural changes and motor behavior improvement. We
used linear regression analyses to test if there was a significant linear relationship between
changes in brain structural characteristics and motor behavior. The corrected significance for
three brain structural variables was set at 0.0167 (alpha = 0.05/3).

Neurological Biomarkers of Post-Stroke Motor Recovery 95

Results
Change in motor behavior. There was a significant decrease in WMFT-distal time score
across all participants (Z = 3.38, p < .001, Figure 4-1). Only eleven out of thirty-seven
participants demonstrated a clinically important difference in WMFT time score for all 15 items
(Figure 4-2).

Figure 4-1. Difference in WMFT-distal between two time-points. T1 – the baseline time
point; T2 – 3-month post baseline. The central blue box is the interquartile range (IQR), a red
horizontal line is the median; the whiskers above and below the box are the locations of the
minimum and maximum values. Each individual line indicates each participant’s change in
WMFT-distal time score (Dark green lines – participants who had clinically important difference
(CID) in WMFT; red lines – participants who did not show CID).
Neurological Biomarkers of Post-Stroke Motor Recovery 96

Figure 4-2. The clinically important difference (CID) in WMFT. Green bars indicate those
participants who demonstrated CID in WMFT time score change. Note that we used the CID of
WMFT mean time score for all fifteen items, as there is no CID defined for WMFT-distal.

Neurological Biomarkers of Post-Stroke Motor Recovery 97

Linear relationship between baseline CST biomarker and change in motor
behavior. Baseline CST FA asymmetry significantly predicted changes in motor behavior.
Baseline CST FA asymmetry alone explained 22% of the variance in ΔWMFT-distal [%] (F (1,
36) =10.1, p < .01, Figure 4-3).

Figure 4-3. Linear regression between baseline CST FA asymmetry and Δ WMFT-distal.

Neurological Biomarkers of Post-Stroke Motor Recovery 98

Regularization and selection of best set of variables for predicting change in motor
behavior. The lasso linear regression analysis with leave-one-out cross-validation showed that a
model including age, baseline UEFM, and CST FA asymmetry most accurately predicted
ΔWMFT-distal. The predicted ΔWMFT-distal from the lasso regression model explained 25% of
the variance in the actual ΔWMFT-distal (95% confidence interval of R
2
: 0.23 – 0.26, p < .001,
Figure 4-4 and 4-5 (A)). Given that two participants were determined as outliers based on
statistical rules (including Cook’s distance and residuals), we also performed the regression
between the predicted ΔWMFT-distal and actual ΔWMFT-distal without the two participants.
Explained variance was increased to 39% (p < .001, Figure 4-5 [B]). Predicted ΔWMFT-distal
from LASSO regression was calculated as following:
Predicted ΔWMFT
/01234
= 16.0+6.9×<=> ?@
ABCD
−3.1×GH?I−3.5×@KL

Figure 4-4. Cross-validated mean square error of lasso fit. Red dots – the mean square error
of a model with a corresponding lambda value (x-axis). Whiskers – the standard error of mean
square error of leave-one-out cross-validation. The green vertical dash line – the lambda of the
model that showed the minimal mean square error.
Neurological Biomarkers of Post-Stroke Motor Recovery 99

Figure 4-5. Regression between actual Δ WMFT-distal and predicted Δ WMFT-distal from
the lasso regression model. (A) Regression model with all 37 participants. Two participants in
dashed circle were determined as outliers. (B) Regression model with 35 participants, 2 outliers
marked in (A) were excluded.

Table 4-1. 95% Confidence interval of beta values for predictor variables from
bootstrapped samples.
Predictor Variable Mean Beta 95% CI
Constant 16.0 15.1 – 16.9
CST FA Asym 6.9 5.7 – 8.0
UEFM -3.1 -4.0 – -2.3
Age -3.5 -4.5 – -2.5

Neurological Biomarkers of Post-Stroke Motor Recovery 100

Figure 4-6 Diagnostics of lasso regression model. (A) Case order plot of Cook's distance. Red
circles indicate each participant's Cook's distance, and horizontal dashed line shows mean Cook's
distance plus 2 standard deviation. (B) Normal probability plot of residuals. Two outliers are
marked with red dashed circle.

Neurological Biomarkers of Post-Stroke Motor Recovery 101

Baseline CST biomarker as a predictor of clinically important difference in motor
behavior. A logistic regression analysis showed that the baseline CST FA asymmetry alone can
predict if a participant will have CID in WMFT-total (C-index=0.6231, Chi
2
statistics vs.
constant model = 7.81, p < .01, Figure 4-7). Further, an independent t-test on the baseline CST
FA asymmetry between people who showed CID and people who did not revealed that people
with CID had significantly greater baseline CST FA asymmetry (Figure 4-8).

Figure 4-7. Logistic regression between baseline CST FA asymmetry and CID in WMFT-
total. Black dots indicate data points of participants, and the red line is the fit of the logistic
regression between baseline CST FA asymmetry and CID in WMFT.
Neurological Biomarkers of Post-Stroke Motor Recovery 102

Figure 4-8. An independent t-test of baseline CST FA asymmetry between participants
with CID and without CID. Red circles indicate data points of participants who did not show
CID (N=26), and blue circles indicate data points of participants who had CID (N=11). Bars
indicate group means, and whiskers show standard error of group mean.

Neurological Biomarkers of Post-Stroke Motor Recovery 103

Regularization and selection of best set of variables for predicting clinically
important difference in motor behavior. We chose a cross-validated lasso regression model
with a lambda that showed minimal deviance plus one standard error, instead of a model that
showed minimal deviance. This is because the model with minimal deviance plus one standard
error has less predictor variables, and therefore would prevent overfitting. The model including
baseline CST FA asymmetry, UEFM, and age was the most accurate model to predict the CID in
WMFT (Figure 4-9 and 4-10). Further, this multimodal model was slightly more accurate than
the simple logistic model that used CST FA asymmetry alone (C-index=0.6802).

Figure 4-9. Cross-validated deviance of logistic lasso fit. Red dots indicate the mean square
error of a model with a corresponding lambda value. Whiskers indicate the standard error of
deviance of leave-one-out cross-validation. The green vertical line indicates the model that
showed the minimal deviance, and the blue vertical line indicates the model with minimal
deviance plus 1 standard error.

Neurological Biomarkers of Post-Stroke Motor Recovery 104

Figure 4-10. Logistic regression fit between the predicted CID from lasso logistic regression
and actual CID in WMFT. Black dots indicate data points of participants, and red line indicates
the logistic regression model fitting between baseline CST FA asymmetry and CID in WMFT.

Neurological Biomarkers of Post-Stroke Motor Recovery 105

Changes in brain structural biomarkers (CST FA asymmetry, lateral ventricle
volume asymmetry, and lesion FA). Separate Wilcoxon signed rank tests showed that there
were no statistically significant changes in brain structural biomarkers, including CST FA
asymmetry, lesion FA, and lateral ventricle volume asymmetry, between the baseline and 3-
month post (p>0.05 Table 4-1).

Table 4-2. Separate Wilcoxon signed rank tests for brain structural variables.
Variable
Baseline
(Mean ± SD)
3-month post
(Mean ± SD)
Z statistics df p
CST FA Asym. 0.065 ± 0.04 0.059 ± 0.04 1.41 35 0.16
Lesion FA 0.165 ± 0.08 0.164 ± 0.08 1.04 35 0.30
Ventricle Asym. 0.118 ± 0.20 0.101 ± 0.20 1.12 35 0.27

A 2×2 repeated measures ANOVA on CST FA showed a difference in FA between ipsi
and contra-lesional CSTs at both time-points, but no changes between two time-points and no
interaction between hemisphere and time-points (Table 4-2 and Figure 4-11). Further, a 2×2
repeated measures ANOVA on lateral ventricle volumes showed there was an increase in both
ipsi- and contra-lesional lateral ventricle volumes at the second time-point, and also a difference
in the lateral ventricle volume between hemispheres, but no interaction between the hemisphere
and time-points (Table 4-3 and Figure 4-12).

Neurological Biomarkers of Post-Stroke Motor Recovery 106

Table 4-3. 2×2 repeated measures ANOVA on CST FA.
Source SS df MS F p
Hemisphere 0.0971 1 0.0971 89.89 <.001
Time-points 0.0003 1 0.0003 3.55 .0680
Hemisphere
x Time-
points
0.0001 1 0.0001 3.36 .0752

Figure 4-11. 2×2 repeated measures ANOVA on CST FA. Blue line indicates group mean
change in FA of the contralesional CST, and red dash line indicates group mean change in FA of
the ipsilesional CST. Whiskers indicate standard error of group mean.
Neurological Biomarkers of Post-Stroke Motor Recovery 107

Table 4-4. 2×2 repeated measures ANOVA on lateral ventricle volumes.
Source SS df MS F p
Hemisphere 759406620 1 759406620 8.08 .007
Time-points 7479313.36 1 7479313.36 8.01 .008
Hemisphere
x Time-
points
710930.028 1 710930.028 0.87 .359

Figure 4-12. 2×2 repeated measures ANOVA on lateral ventricle volumes. Blue line
indicates group mean change in the contralesional lateral ventricle volume, and red dash line
indicates group mean change in the ipsilesional lateral ventricle volume. Whiskers indicate
standard error of group mean.
Neurological Biomarkers of Post-Stroke Motor Recovery 108

A linear relationship between brain structural changes and motor improvement.
Separate linear regression analyses showed no significant linear relationship between changes in
any brain structural biomarkers and changes in WMFT-distal (Table 4-4 and Figure4-13).

Table 4-5. Linear regression analyses between changes in brain structural variables and
changes in WMFT-distal.
Model R
2
Coefficients SE t-stat p value
ΔWMFT-distal ~ ΔCST FA asymmetry 0.02
Constant 17.0 4.3 3.9 < .001
ΔCST FA asym. 216.2 312.7 0.7 .49
ΔWMFT-distal ~ ΔLVA 0.05
Constant 17.2 4.1 4.2 < .001
ΔLVA 66.9 45.2 1.5 .20
ΔWMFT-distal ~ ΔLesion FA 0.04
Constant 16.8 4.0 4.2 < .001
ΔLesion FA 432.7 425.1 1.0 .32
(WMFT – Wolf motor function test, LVA – lateral ventricle volume asymmetry, CST –
corticospinal tract, FA – fractional anisotropy.)
Neurological Biomarkers of Post-Stroke Motor Recovery 109

Figure 4-13. Linear regression between ΔCST FA asymmetry and ΔWMFT-distal.

Discussion
This study provides evidence that the DTI-derived CST biomarker can predict motor
improvement in chronic stroke survivors with mild-to-moderate motor impairment. This study
utilizes cross-validated lasso regression analyses to regularize and select the most essential
predictors for post-stroke motor improvement. A recent systematic review of current prognostic
studies using biomarkers derived from neuroimaging/neurophysiology (DTI, structural and
function MRI, TMS, etc.) found that these studies did not cross-validate their prediction model
for post-stroke motor recovery (Kim and Winstein, 2017). Thus, our study provides a more
accurate and robust prediction model for motor improvement in chronic stroke survivors by
utilizing a machine learning-based statistical method.
Neurological Biomarkers of Post-Stroke Motor Recovery 110

Our findings indicate that chronic stroke survivors with greater CST microstructural
impairment at baseline (i.e., greater CST FA asymmetry) had greater motor improvement after a
3-month period. This result is inconsistent with previous studies which showed less CST
microstructural impairment (i.e., less FA asymmetry) predicted greater improvement in motor
behavior after therapy (Borich et al., 2014; Lindenberg et al., 2012). For example, Lindenberg
and colleagues reported that chronic stroke survivors with greater fractional anisotropy of the
ipsilesional pyramidal tract at baseline showed greater change in the WMFT after a five-day non-
invasive brain stimulation and rehabilitation therapy program. Further, Borich and colleagues
found that chronic stroke individuals with greater FA asymmetry of the posterior limb of the
internal capsule (PLIC) demonstrated less motor learning capability following UE training. The
inconsistency in our results compared to previous studies may be due to differences in sample
size, severity of stroke participants, DTI-derived metrics, and/or intervention duration.
Importantly, previous studies recruited stroke participants with a wider range of severity with
smaller sample size. On the other hand, our study recruited a relatively large sample size with a
much narrower range of severity. We believe that people with moderate motor impairment
demonstrated greater motor improvement than people with mild motor impairment, given that
they had not reoched their maximal recovery potential. While a ceiling effect for motor recovery
may have occurred in people with milder impairment, thus no more improvement. By combining
two opposing results, those from our study and Lindenberg et al., we expect that the potential
relationship between CST structural damage and motor improvement in chronic stroke survivors
would be an inverted “U” shape as depicted in Figure 4-14.

Neurological Biomarkers of Post-Stroke Motor Recovery 111

Figure 4-14. A hypothesized inverted “U” shape relationship between CST FA asymmetry
and motor improvement in chronic stroke survivors.

This study employed cross-validated lasso linear and logistic regression analyses. To our
best knowledge, this is the first study that utilized the lasso regression to develop a prediction
model for UE motor improvement in chronic stroke survivors. Several studies have utilized the
lasso regression for prediction of motor recovery in subacute stroke survivors (Quinlan et al.,
2014), and for prediction of apraxia of speech (Ballard et al., 2016). Using the cross-validated
lasso regression, we can avoid the multicollinearity among different predictors, and improve the
accuracy and robustness of the model (Tibshirani, 1996). The lasso linear and logistic regression
analyses indicate that the combination of baseline CST FA asymmetry, UEFM, and age better
predict changes and a clinically important difference (CID) in motor performance in this
population. This is consistent with previous studies that also used multimodal prediction model
Neurological Biomarkers of Post-Stroke Motor Recovery 112

by combining neurological biomarkers, clinical motor scores, and demographic information to
improve prediction of motor recovery after stroke (Burke et al., 2014; Kim and Winstein, 2017).
In addition to that, the lasso regression analyses revealed that the baseline CST FA asymmetry is
the most crucial factor to predict motor improvement in chronic stroke survivors, given that the
beta of baseline CST FA asymmetry was as twice as greater than betas of baseline UEFM and
age in the multimodal model. It is consistent with previous findings that CST structural damage
is the most crucial factor of motor impairment after stroke.
Further, this study tested if the baseline CST FA asymmetry can predict the clinically
important difference (CID) of WMFT. A previous study defined the WMFT CID in chronic
stroke survivors (K. C. Lin et al., 2009). A two or more second decrease in mean time score for
fifteen timed items was determined as a minimal CID of WMFT. Among 37 participants, 11
showed a greater decrease in WMFT-total time score than the minimal CID after a 3-month
period. In general, people with WMFT CID had significantly greater baseline CST FA
asymmetry than people did not achieve WMFT CID. Further, the logistic regression and lasso
logistic regression showed that people who had greater CST FA asymmetry would have a greater
chance to achieve WMFT CID. This finding may indicate that the CST FA asymmetry can be
used to predict whether a chronic stroke individual will have clinically meaningful improvement
in their motor behavior. Further, future studies should be conducted to determine if the CST FA
asymmetry can predict response to motor therapy in chronic stroke survivors.
While this research did not factor in the dose effect on motor behavior, we did
hypothesize that there would be a significant increase in FA of ipsilesional CST following the 3-
month UE therapy. Our results did not support any changes in ipsilesional CST microstructure,
and no linear relationship between changes in CST FA and motor improvement. Further, we did
Neurological Biomarkers of Post-Stroke Motor Recovery 113

not observe any changes in other DTI-derived metrics of CST, such as mean diffusivity, axial
diffusivity, and radial diffusivity (not reported). Previous studies with non-disabled healthy
adults have shown that motor skill practice can induce white matter microstructural changes, and
these changes are associated with motor learning (Boyke et al., 2008; Fields, 2008; Gregg et al.,
2007; May, 2011; Scholz et al., 2009; Steele et al., 2013; Taubert et al., 2011, 2010). Further,
studies with animal stroke models have demonstrated that CST microstructural improvements,
such as re-myelination of damaged axons or increase in axonal diameter, are associated with
functional recovery (Gregg et al., 2007; Ramos-Cejudo et al., 2015; Sato et al., 2009; Song et al.,
2003). Recent clinical trials for chronic post-stroke aphasia also supported that microstructural
changes in white matter fibers that connect essential language-related brain regions after
intensive language therapy are associated with improvement in speech (Schlaug et al., 2009;
Wan et al., 2014). It has been suggested that plasticity in white matter structural morphology,
such as axonal diameter, myelin thickness, and the number of myelinated axons, may contribute
to functional motor performance improvement with learning by optimizing the speed or
synchrony of brain signal delivery (Zatorre et al., 2012). However, our findings did not show any
microstructural changes in CST or concomitant motor improvement. Given that motor
improvement, as determined by movement speed for example, can be a function of
compensatory movement strategies (Levin et al., 2009, 2002; Michaelsen et al., 2004), it is
possible that CST microstructural changes do not play a crucial role in motor improvement in
this case. A number of studies have shown that experience-dependent neuroplasticity, such as the
training-induced reorganization of the ipsilesional primary motor cortex, can be an essential
neural mechanism underlying motor improvement during the chronic stage (Kitago and
Krakauer, 2013; Krakauer, 2006, 2005, 2004; Krakauer et al., 2012; Zarahn et al., 2011).
Neurological Biomarkers of Post-Stroke Motor Recovery 114

Possibly, a functional change such as an increase in ipsilesional primary motor cortex activity
and not a structural change is the putative neural mechanism underlying motor improvement in
this population. Other reasons why changes in CST microstructure were not observed may be
associated with: 1) the cohort of stroke participants we examined; 2) the limited resolution of
DTI used here; and 3) the timing of CST microstructural changes. As our participants were mild-
to-moderately impaired chronic stroke survivors, only 11 of 37 participants showed substantial
changes in their motor behavior (i.e., CID of WMFT). This might have limited the observation of
any CST microstructural changes. Further, although we used relatively high angular resolution
DTI (64 gradient directions) compared to previous DTI studies, we did not use a sufficient b-
value that would enable use of more accurate tractography reconstruction methods. It has been
shown that orientation distribution function (ODF)-based tractography is more accurate in
distinguishing crossing fibers. Further, more recent diffusion MRI methods, such as diffusion
kurtosis image (DKI), can better detect changes in white matter microstructural characteristics
(Spampinato et al., 2017). Thus, the limitation in DTI method may prevent observation of actual
changes in CST microstructure, especially if those changes are subtle. Lastly, we had a 3-month
period between measurement time-points. The time course associated with white matter
microstructural change is unclear. Thus we do not know if a 3-month period is sufficient to
observe CST microstructural changes (Han et al., 2009; Hermann and Chopp, 2012; Song et al.,
2003; Wan et al., 2014). The fact is that the reorganization of primary motor cortex after
intensive training may induce microstructural changes in CST (May, 2011). We need more high-
quality larger scale clinical trials to determine if CST microstructural changes can be a crucial
neural mechanism underlying motor improvement in chronic stroke survivors.
Neurological Biomarkers of Post-Stroke Motor Recovery 115

As we expected, there were no changes in lesion FA. It is well known that the actual
stroke lesion does not show any structural/functional changes during the chronic stage due to the
limited neuronal repair and regeneration in the CNS lesion (Schwamm et al., 1998). Further, we
also expected that there would be a therapy-induced decrease in ipsilesional lateral ventricle
volume, indicative of increased subcortical white matter (Breteler et al., 1994; Sachdev, 2004). A
previous cross-sectional study demonstrated that the white matter volume is associated with
clinical motor scores in chronic stroke survivors (Rickards et al., 2012). Previously, the
ventricle-to-brain ratio (VBR) has been used to estimate subcortical white matter volume. Our
study employed a new method, the lateral ventricle volume asymmetry (LVA), to represent the
white matter volume in the ipsilesional hemisphere. Our study, however, demonstrated that there
was no significant relationship between changes in white matter volume and motor improvement
in this population. In fact, there was a significant increase in lateral ventricle volume in both
hemispheres, indicative of ongoing white matter atrophy. This may be associated with a known
age effect on the brain white matter volume (Walhovd et al., 2005).
This study has several limitations. As mentioned in the previous chapter, the small
sample size and a limited cohort of stroke participants are the biggest concerns. Further, as we
mentioned above, there are higher resolution diffusion MRI, such as DKI, that can better capture
the microstructural changes in CST (Spampinato et al., 2017). Thus, use of these newer diffusion
MRI methods should be considered in future studies to examine CST microstructural changes.
Another limitation of this study is that we employed a time-based clinical motor score that lacks
knowledge of how motor improvement is achieved. Biomechanical assessment of the UE, such
as kinematic and kinetic assessments of goal-directed UE movements, should be included in
future studies to better determine the relationship between brain structure and motor behavior
Neurological Biomarkers of Post-Stroke Motor Recovery 116

(Bernhardt et al., 2017; Cramer et al., 2017; Langhorne et al., 2009). Lastly, there are several
other potential predictors of motor improvement in chronic stroke survivors. Recent studies have
shown that brain-derived neurotrophic factor (BDNF) genotype is associated with motor skill
learning in non-disabled adults and stroke survivors (Chang et al., 2017; Kleim et al., 2006).
Specifically, Chang and colleagues reported that BDNF genotype is a significant predictor of UE
motor recovery in subacute stroke survivors. Thus, including BDNF genotype may improve the
accuracy of prediction for motor improvement in chronic stroke survivors. Further, numerous
studies have reported the importance of social-cognitive psychological factors on motor recovery
after stroke (Glass et al., 1993; Livingston-Thomas et al., 2016; Winstein et al., 2014, 2016). As
the social-cognitive psychological factors, including self-efficacy, autonomy support, social-
relatedness, depression, family support, etc., influence motor performance and learning in stroke
survivors, these factors should be included in the prediction model for motor improvement
(Winstein et al., 2014).

Conclusion
Baseline DTI-derived CST FA asymmetry is a significant predictor of motor
improvement in chronic stroke survivors with mild-to-moderate motor impairment. Further,
combining baseline CST FA asymmetry with clinical and demographic variables slightly
improves the accuracy of prediction. However, microstructural changes in CST were not
observed, nor any parallel changes in motor behavior. Future studies with high-resolution
diffusion MRI should be conducted to better understand the relationship between CST
microstructural changes and motor improvement in chronic stroke survivors.
Neurological Biomarkers of Post-Stroke Motor Recovery 117

CHAPTER FIVE: SUMMARY AND GENERAL DISCUSSION

The overall purpose of this dissertation was to determine the relationship between CST
white matter characteristics derived from diffusion tensor imaging (DTI) and upper extremity
(UE) motor behavior in chronic stroke survivors with mild-to-moderate severity. DTI-derived
CST biomarker has been commonly used in previous studies to estimate stroke-related brain
structural damage that is associated with motor behavior. However, there is a lack of high-quality
evidence to utilize the DTI-derived CST biomarker in the chronic stroke population. Thus, we
conducted three studies to identify the methodological limitations in current evidence, and to
determine the relationship between CST structural characteristics and motor behavior in chronic
stroke survivors. This chapter summarizes main results of these three studies, interprets results,
and discusses future implications of our results. Lastly, limitations of dissertation studies are
discussed.

Summary of Main Results
The first study (Chapter 2) was a systematic review of longitudinal prognostic clinical
studies aimed to identify prediction models for post-stroke motor recovery using neurological
biomarkers. We performed qualitative and quantitative assessments of 71 clinical studies. We
found a lack of evidence supporting the use of the DTI-derived CST biomarker to predict motor
improvement in chronic stroke survivors. Only a few clinical studies with methodological
limitations reported significant predictive values of CST biomarker. Most prominent
methodological limitations are small sample size, lack of cross-validation of their model, and no
consideration of the clinically important difference in motor outcome measures. Most DTI
Neurological Biomarkers of Post-Stroke Motor Recovery 118

studies were conducted with stroke survivors in early phases. Further, the quantitative analyses
showed that the model utilizing the neurological biomarkers along with clinical and demographic
information has higher prediction accuracy than the model utilizing the neurological biomarkers
alone.
The second study (Chapter 3) was performed to determine the most accurate estimate of
CST microstructure in chronic stroke survivors with mild-to-moderate severity. We
systematically compared seven different DTI-derived CST microstructure estimates. We found
that the 3-dimensional (3-D) individual tractography-based CST estimate was the most accurate
estimate. Only this method was significantly correlated with a motor outcome measure. This
result suggests that the individual tractography-based CST estimate is appropriate to determine
the relationship between CST microstructure and motor behavior in this stroke population.
The third study (Chapter 4) aimed to determine if the DTI-derived CST biomarker can be
used to predict motor improvement, and to examine if brain structural changes are associated
with motor improvement in this stroke population. We found that the DTI-derived CST
biomarker is a significant predictor of motor improvement in chronic stroke survivors, and
combining CST biomarker and clinical and demographic variables can improve the prediction
accuracy. On the other hand, we were unable to observe any brain structural changes in a 3-
month period, and there was no significant association between brain structural changes and
motor improvement.

The relationship between Brain Structure and Motor Behavior.
This dissertation investigated the relationship between brain structure and motor behavior
in chronic stroke survivors with mild-to-moderate motor impairment. This study utilized the
Neurological Biomarkers of Post-Stroke Motor Recovery 119

DTI-derived CST biomarker to identify the relationship between brain structure and motor
behavior. Although previous have shown the feasibility of DTI-derived CST biomarker to be
used to predict motor recovery during an early phase of stroke, only a few studies investigated
the feasibility of CST biomarker to be utilized for prediction of motor improvement in chronic
stroke survivors. Further, there is no evidence supporting that if CST microstructural changes are
associated with motor improvement in chronic stroke survivors.
In this section, I will discuss how the brain structural characteristics are associated with
motor behavior in chronic stroke survivors.

The 3-D individual tractography-based CST estimate is the most accurate
estimation of CST microstructure. We found that DTI-derived CST biomarker is a valid
surrogate of CST microstructure that is associated with motor behavior in chronic stroke
survivors with mild-to-moderate motor impairment. Specifically, 3-D individual tractography-
based CST FA asymmetry is an accurate estimate of CST microstructural characteristics.
Although previous studies have shown that DTI-derived fractional anisotropy (FA) of a portion
of CST, such as posterior limb of the internal capsule (PLIC) or cerebral peduncle (CP),
represents the entire CST microstructure (Lindenberg et al., 2012; Park et al., 2013), our findings
opposed to these results. An underlying assumption of previous studies was that there is
Wallerian degeneration along with the entire CST, and thus FA of PLIC or CP can accurately
estimate the degree of CST structural damage (Lindenberg et al., 2012). Our results did not
support this assumption. We observed most participants had decreased FA of ipsilesional CST
only near the lesion. Thus, 3-D individual CST tractography most accurately estimates the CST
Neurological Biomarkers of Post-Stroke Motor Recovery 120

structure. This might be associated with the stroke population recruited in this study, given that
we included only the chronic stroke survivors with mild-to-moderate motor impairment.

Prediction of motor improvement using DTI-derived CST biomarker in chronic
stroke survivors with mild-to-moderate motor impairment. Our findings support that the
CST biomarker can be used to predict motor improvement in this population. We found that
people with greater CST structural impairment had more improvement in UE motor behavior.
This result is inconsistent with a previous work conducted by Lindenberg and colleagues
(Lindenberg et al., 2012). They reported that chronic stroke survivors with greater ipsilesional
CST FA showed less improvement in UE motor behavior. This might be related to a different
cohort of chronic survivors. Our participants’ ipsilesional CST FA ranged from 0.32 to 0.51
(mean ± standard deviation = 0.41 ± 0.04), while the previous study participants’ ipsilesional
CST FA ranged from 0.2 to 0.45 (mean ± standard deviation = 0.32 ± 0.06). Combining our and
their results together, it is possible that the relationship between baseline ipsilesional CST FA
and motor improvement may be inversed “U” shape.

Feasibility of multimodal prediction model for motor improvement using lasso
regression. Adding clinical and demographic variables to the CST biomarker may improve the
accuracy of prediction model for motor improvement. Results from a systematic review and
cross-validated lasso regression analyses supported that the model including CST FA asymmetry
and other clinical and demographic variables may be more accurate than a simple regression
model including CST FA asymmetry alone. Motor improvement during the chronic stage is
influenced by several factors, while the motor recovery in the early phase is mainly driven by the
Neurological Biomarkers of Post-Stroke Motor Recovery 121

spontaneous neurological recovery (Bernhardt et al., 2017, 2016; Boyd et al., 2017; Cramer et
al., 2017). Thus, multimodal prediction model is more accurate than the model utilizes the CST
biomarker alone (Kim and Winstein, 2017). This study showed a feasibility of cross-validated
lasso regression to regularize and select the essential predictor variables for motor improvement
in chronic stroke survivors to improve prediction model accuracy.

Brain Structural Changes in Chronic Stroke Survivors. We utilized three
neuroimaging-derived brain structural measures. There was no significant structural change
neither in lesional area nor corticospinal descending motor pathways in both hemispheres.
Although a subgroup of participants who had clinically important difference in WMFT, there
was no significant brain structural changes. These results did not support our hypothesis that
there will be significant decrease in CST FA asymmetry after a 3-month period. This result may
support previous findings that experience-dependent functional neuroplasticity in the ipsilesional
primary motor cortex plays an essential role in motor improvement in chronic stroke survivors.
However, there are several potential reasons why the results did not support our hypothesis.
First, the cohort of stroke participants of this study would be a matter. As they were mild-to-
moderately impaired chronic stroke survivors, only 11 of 37 participants presented substantial
changes in their motor behavior. As majority of our participants did not show significant motor
behavioral changes, there would be no brain structural changes in a 3-month period. Another
reason why we could not observe the notable change in CST microstructure would be related to
the limited resolution of DTI used in this study. Although we used relatively high angular
resolution DTI (64 gradient directions) compared to previous DTI studies (15-30 directions), we
did not use a sufficient b-value to utilize higher level tractography reconstruction methods. It has
Neurological Biomarkers of Post-Stroke Motor Recovery 122

been shown that the probabilistic q-ball fiber tracking (orientation distribution function [ODF]-
based tractography) is more accurate to track white matter fibers, and to estimate the CST
microstructure (Bucci et al., 2013). Further, more recent diffusion MRI methods, such as
diffusion kurtosis imaging (DKI), can better detect the changes in white matter microstructural
characteristics (Spampinato et al., 2017). Thus, the limitation in DTI method can prevent to
observe the actual changes in CST microstructure. Another reason is the timing of CST
microstructural change. Lastly, we had a 3-month period between two time-points. It is not clear
the timing of white matter microstructural change, and we do not know if a 3-month period is
sufficient to observe CST structural changes (Han et al., 2009; Hermann and Chopp, 2012; Song
et al., 2003; Wan et al., 2014).
We observed the increased lateral ventricle volume in both hemispheres. This may
indicate the subcortical white matter atrophy in both hemispheres, given that the lateral ventricle
volume is negatively correlated with with the volume of the subcortical white matter (Breteler et
al., 1994; Sachdev, 2004). Our study employed a new method, the lateral ventricle volume
asymmetry (LVA), to represent the white matter atrophy in the ipsilesional hemisphere.
Although we did not observe the significant increase in ipsilesional hemisphere white matter
volume (i.e., decrease in lateral ventricle volume asymmetry), there was significant increase in
lateral ventricle volumes in both hemispheres. This may indicate that there was an ongoing white
matter atrophy in both hemispheres, and it may be associated with age effect on the brain white
matter volume (Walhovd et al., 2005).

No Relationship between Brain Structural Changes and Motor Improvement. The
brain structural changes did not explain the motor improvement. It is possible that motor
Neurological Biomarkers of Post-Stroke Motor Recovery 123

improvement in this chronic stroke population may be not associated with CST structural
changes. Experience-dependent functional reorganization of ipsilesional primary motor cortex
and functional connectivity among sensorimotor cortices can be an essential neural mechanism
underlying motor improvement during the chronic stage (Cramer, 2008; Cramer and Riley, 2008;
Grefkes and Fink, 2011; Kleim et al., 2002; Krakauer, 2006). However, it is also possible that the
reorganization of primary motor cortex after an intensive training may induce microstructural
changes in CST (May, 2011). We need more high-quality larger scale clinical trials to determine
if CST microstructural changes can be a crucial neural mechanism underlying motor
improvement in chronic stroke survivors.

Clinical Implications and Future Research Directions
Our findings provide an evidence supporting that the DTI-derived CST biomarker is
feasible to be used for prediction of motor improvement in chronic stroke survivors with mild-to-
moderate motor impairment. This cohort of stroke survivors has been frequently targeted by
numerous clinical trials to test the efficacy of rehabilitation therapies, given this population
would be more responsive to the therapy. Future clinical research needs to include diffusion MRI
assessment of brain structure in order to assess potential therapeutic responses (Boyd et al.,
2017). Developing an accurate prediction model for motor recovery in chronic stroke survivors
will improve the efficacy of stroke rehabilitation in the chronic phase by customizing
rehabilitation strategies and determining optimal therapeutic approaches and dose (Boyd et al.,
2017; Lindenberg et al., 2012; Puig et al., 2017; Stinear et al., 2017). Thus, this study may
indirectly contribute to improve the prediction modeling for motor improvement in post-stroke
individuals in chronic stage. Although direct clinical application of this study to clinical practice
Neurological Biomarkers of Post-Stroke Motor Recovery 124

is limited, DTI-derived CST biomarker can be used for participant selection and categorization in
clinical trials examining efficacy of stroke rehabilitation interventions (Stinear, 2017). Thus, this
dissertation will contribute to improve the efficacy of stroke rehabilitation by providing an
evidence for DTI-derived CST biomarkers that can be used in future clinical trials.
Future research utilizing higher resolution diffusion MRI should be conducted to
determine if brain structural changes are associated with motor improvement in chronic stroke
survivors. Further future research utilizing multimodal imaging methods, including DTI,
functional MRI, and transcranial magnetic stimulation (TMS), etc., should also be performed to
better understand the brain-behavior relationship in chronic stroke survivors.

Limitations
The small sample size and limited cohort of stroke participants are the biggest concerns.
Further, DTI resolution used in this study is another limitation. There are higher resolution
diffusion MRI, such as DKI, that can better capture the microstructural changes in CST
(Spampinato et al., 2017). Thus, use of these newer diffusion MRI methods should be used in
future studies to examine the CST microstructural changes. Another limitation of this study is
that we employed a time-based clinical motor score that lacks the restitution-substitution
continuum of motor improvement in stroke survivors. Biomechanical assessment of UE, such as
kinematic and kinetic assessments of goal-directed UE movements, should be included in future
studies to better determine the relationship between brain structure and motor behavior
(Bernhardt et al., 2017; Cramer et al., 2017; Langhorne et al., 2009). Lastly, there are several
other potential predictors of motor improvement in chronic stroke survivors, such as BDNF
genotype (Chang et al., 2017; Kleim et al., 2006) and social-cognitive psychological factors
Neurological Biomarkers of Post-Stroke Motor Recovery 125

(Glass et al., 1993; Livingston-Thomas et al., 2016; Winstein et al., 2014, 2016). We did not
include these potential predictors in the model. Thus, future research should test if adding these
variables can improve the prediction accuracy of motor improvement in chronic stroke survivors.

Neurological Biomarkers of Post-Stroke Motor Recovery 126

APPENDIX A. EVIDENCE METHODOLOGICAL STRENGTH EVALUATION

1. Internal Validity
A. Appropriate operational definitions of dependent (outcome) variables. If a study
provided sufficient operational definitions of outcome variables, we gave a score of ‘1’.
B. Appropriate operational definitions of predictor variables. If a study provided
sufficient operational definitions of predictor variables, we gave a score of ‘1’.
C. Reliable or valid measures of dependent variables. If the study conducted
additional experiments demonstrating the validity or reliability of outcome measures for
dependent variables, a score of ‘1’ was given. Proper citation of validity or reliability of these
measures was also considered valid or reliable.
D. Reliable or valid measures of predictor variables. If the study conducted additional
experiments demonstrating the validity or reliability of the measures of predictor variables, a
score of ‘1’ was given. Proper citation of validity or reliability of these measures was also
considered valid or reliable.
E. Blinded or masked evaluations. The evaluators of outcome or predictor measures
should be blinded to the study design and purposes to minimize evaluator bias. Further, the
evaluators of outcome measures also should be blinded to the results of predictor measures, vice
versa.
F. Appropriate observation time-points. We gave a score of ‘1’ if the predictor
variable measure was occurred within 1 month after stroke, and the dependent outcome variable
measure was done at least 8 weeks after the predictor variable measure. This is due in part to the
knowledge that natural neurologic recovery (i.e., recovery of penumbral tissues and mitigation of
Neurological Biomarkers of Post-Stroke Motor Recovery 127

diaschisis) along with significant improvements in motor performance is usually observed within
8 weeks after stroke onset.(Van Kordelaar et al., 2014)
G. Control of dropout. If the author provided information about the dropout of
participants, a score of ‘1’ was given. In addition, ‘1’ was given if there was no dropout of any
participants.

2. Statistical Validity
H. Control for statistical significance. If the author provided information about the
statistical analysis chosen to identify the relationship between predictor and outcome measures,
control for statistical significance was considered sufficient. Specifically, regression analyses
(including linear regression, non-linear regression, and logistic regression), correlation analyses,
and a mixed model would be considered as appropriate statistical models to explore or to test the
relationship between predictor and outcome measures.(Cohen et al., 2013) In addition if the
author provided an appropriate rationale for the statistical analyses, the control for statistical
significance was considered qualified.
I. Appropriate sample size. At least 50 participants were considered an appropriate
sample size. In instances where there were fewer than 50 participants, the sample size was
considered adequate if the author calculated the sample size with appropriate statistical methods.
J. Control for multicollinearity. If there were more than two predictors, studies
including a multicollinearity test, using multiple regression, partial correlation, or semi-partial
correlation, were given a score of ‘1’. In Chen and Winstein
6
, ‘not applicable (NA)’ was used
instead of a score of ‘0’ for studies that did not use a regression model for outcome prediction. In
Neurological Biomarkers of Post-Stroke Motor Recovery 128

this review, a score of ‘0’ was given if the studies did not use a regression model, or if there was
only one predictor variable in the regression model. This was done because the contribution of
participants’ demographics or other factors was not controlled.(Cohen et al., 2013)

3. External Validity
K. Identification of stroke pathology. Because ischemic and hemorrhagic strokes show
different trajectories of motor recovery(Kelly et al., 2003), if results were stratified by pathology,
we gave a score of ‘1’. Further, if a study recruited either hemorrhagic or ischemic stroke, we
gave a score of ‘1’.
L. Specification of inclusion and exclusion criteria. If a study provided the inclusion
and exclusion criteria to specify the study participants’ characteristics, we gave a score of ‘1’.
M. Control for additional treatment effects. In the case that the author described any
medical or rehabilitative interventions during the study to control the effects of additional
treatments on predictor and outcome measures, a score of ‘1’ was given.
N. Cross-validation of the predictor. A score of ‘1’ was given if the study included a
cross-validation of the predictor with an additional independent group of participants with stroke
to validate prediction models. The predicted outcome measures from the model should be
compared to the actual outcome measures of participants in the independent group. This analysis
would provide independent validity and accuracy estimation of the prediction model.(Cohen et
al., 2013)
O. Discussion of minimal clinically important differences (MCID). MCID values are
vital to understand the clinical implications of any change in an outcome measure of motor
Neurological Biomarkers of Post-Stroke Motor Recovery 129

impairment and/or function.(Lang et al., 2008) Thus, studies that included the MCID value of
their outcome measures were given a score of ‘1’.

Neurological Biomarkers of Post-Stroke Motor Recovery 130

APPENDIX B. MOST COMMON PREDICTOR VARIABLES OF EACH
NEUROLOGICAL BIOMARKER TYPE

DTI biomarker type
DTI was the most common neurological biomarker type. Twenty-one studies with 612
participants (~ 19%) used DTI measures as a major biomarker. The most popular DTI-derived
variable was corticospinal tract (CST) fractional anisotropy (FA) ratio between ipsilesional and
contralesional hemispheres measured at the cerebral peduncle region of interest (ROI).
DTI measures dominate the literature by quantification of the structural characteristics or
integrity of the white matter surrounding specific motor related pathways, especially that for the
corticospinal tract. There are a number of DTI-derived variables that have been used to quantify
the structural characteristics of the corticospinal tract (CST). Two distinct approaches were the
most frequently used: 1) DTI-derived metric calculations at a specific two-dimensional (2-D)
ROI, such as the posterior limb of the internal capsule (PLIC) or cerebral peduncle (CP); 2) DTI-
derived metric calculations that involve averaging DTI-derived metrics of reconstructed three-
dimensional (3-D) CST tractography. A number of studies use the 2-D ROI, such as PLIC or CP
to quantify the CST microstructure. The underlying assumption of this approach is that Wallerian
degeneration is occurring across the entire CST after stroke damage to the CST.(Koyama et al.,
2012; Kuzu et al., 2012; Lindenberg et al., 2012) So calculating the DTI-derived metrics at a
remote CST section, such as PLIC or CP, can represent the degree of degeneration of the entire
CST.(Kuzu et al., 2012) It has been shown that the FA is decreased across the entire ipsilesional
CST compared to the FA of contralesional CST in chronic stroke.(Lindenberg et al., 2012)
Further, Park et al(Park et al., 2013) assessed these different approaches in chronic stroke, and
Neurological Biomarkers of Post-Stroke Motor Recovery 131

they found that there were no significant differences in DTI-derived metrics between 3-D CST
tractography and 2-D ROIs. They defined the 2-D PLIC ROI on an axial slice within the 3-D
CST volume of interest (VOI). It is reasonable to use 2-D ROI to represent CST damage, but
concerns pertaining to the 2-D ROI approach are several: First, there are only a few studies to
support the notion that DTI-derived metrics calculated from a 2-D ROI are representative of the
entire CST microstructure. Second, it is possible that if the lesion is small, the DTI may not be
able to capture the subtle post-stroke changes in white matter microstructure across the entire
CST. Further, most studies included in this review used manually drawn 2-D ROIs on an axial
slice without inter- and/or intra-rater reliability tests. In this case, computation of DTI-derived
metrics of CST at PLIC or CP ROI may not be an accurate method to represent the entire CST.
As the PLIC or CP ROI is manually drawn, and it may also include not only CST but also fibers
from other tracts (e. g. sensory pathways). We believe higher resolution diffusion imaging, such
as high angular resolution diffusion imaging (HARDI), and 3-D CST VOI will be more accurate
for identification of post-stroke changes in CST microstructure. Therefore, we need studies that
employ higher resolution diffusion MRI and that compare various DTI CST microstructure
quantification approaches.
A number of diffusion MRI analysis software packages, such as BrainSuite
(http://brainsuite.org/), FSL (www.fmrib.ox.ac.uk/fsl), or FreeSurfer (http://freesurfer.net/), can
be useful to reconstruct diffusion tensor-based CST tractography. These software packages
provide automated image registration, brain region labeling, and fiber tracking processes. It is
likely that the computation of metrics using the automated imaging data process will provide
more robust inter- and intra-rater reliability of DTI-derived biomarkers.(Park et al., 2013)
Neurological Biomarkers of Post-Stroke Motor Recovery 132

As CST is not the only motor pathway that influences motor ability, some researchers
have focused on other sensorimotor-related pathways, such as the interhemispheric M1-M1
connection(Lindenberg et al., 2012), the cortico-cortical connection between motor-related
cortices(Granziera et al., 2012), or alternative motor pathways.(Lindenberg et al., 2012, 2010)
Recent studies also have shown that functional and structural connectivity of remote areas from
the lesion site can be affected after stroke.(Crofts et al., 2011; Grefkes and Fink, 2011; Johansen-
Berg et al., 2010; Nudo, 2013)

TMS biomarker type
TMS was the second most frequently used neurological biomarker. Thirteen studies with
462 participants (~ 14%) used TMS-derived variables as predictors.
TMS has been used to examine the functional integrity of the cortico-spinal connection
(i.e., corticospinal tract) by measuring the presence of MEP of upper or lower extremity muscles.
In most cases, a TMS response in a specific arm/hand or leg/foot muscle was recorded as binary
data (i.e., absent or present).(Stinear et al., 2012) Sometimes, conduction velocity or time has
been used to predict motor recovery after stroke, but these variables did not turn out to be
significant biomarkers of motor recovery.

Structural MRI biomarker type
Nine studies with 859 participants (~ 26%) used conventional structural MRI measures to
identify lesion size and location as their predictors. The studies using diffusion-weighted image
(DWI) were classified as a sMRI category rather than as a DTI category, as their primary
purpose of DWI assessment was to calculate the lesion volume and to identify the lesion
Neurological Biomarkers of Post-Stroke Motor Recovery 133

location. Each study selected different variables from sMRI. Usually lesion volume, lesion
location, or white matter (including CST) lesions were used as predictors. Because a few large-
scale retrospective research studies included sMRI, the number of participants from those nine
studies comprised about a quarter of the total participants included in this systematic review.
Conventional structural MRI measures of stroke brains are usually used to identify lesion
location and volume information(S K Schiemanck et al., 2006b), or sometimes to calculate grey
matter thickness of the primary motor cortex.(Muñoz Maniega et al., 2004) Further, CST-lesion
overlap volume (CST-lesion load) is calculated to quantify how much CST is damaged due to
stroke.(Burke et al., 2014) For this method, the lesion is drawn on the T1-weighted, T2-
weighted, FLAIR, or diffusion images, and the overlap volume of the lesion and template CST is
calculated. The template CST can be acquired from the standard brain atlas or from age-matched
participants. This procedure can be used to quantify CST structural integrity.

Functional MRI biomarker type
There were nine fMRI studies with 131 participants (~ 4.1%). The fMRI variables varied
across studies. Most studies analyzed the activation patterns of ipsilesional and/or contra-lesional
motor-related ROIs during specific motor task performance (5 of 9 studies). Three studies used
the resting-state functional connectivity among sensorimotor areas, and one study used the
sensorimotor cortex activation pattern during passive movement.
In most functional imaging studies, fMRI has been used to quantify the level of activation
of the ipsilesional M1. The laterality index of ipsilesional M1 activation or functional
connectivity between bilateral M1s during ipsilateral motor task performance was the most
common fMRI-derived biomarker of motor improvement. In addition, resting-state functional
Neurological Biomarkers of Post-Stroke Motor Recovery 134

connectivity among sensorimotor areas was also used in few studies. In particular, these fMRI
biomarkers were used in studies with chronic stroke individuals. (Table A-1)
Combination type of different neurological biomarkers
Eighteen studies with 694 participants (~ 21%) utilized a combination of neurological
biomarkers to develop multimodal prediction models. In most cases, two structural neurological
biomarkers were combined or structural and functional neurological biomarkers were combined.
The most common case was the combination of DTI biomarker and sMRI biomarker (8 of 18
studies). In addition, a combination of DTI biomarker and TMS biomarker was the next most
frequently used (3 of 18 studies).
A combination of DTI measures and conventional structural MRI measures provides the
microstructural characteristics of corticospinal motor pathways in the brain (DTI metrics of
CST), and quantified measures of CST damage (CST-lesion overlap volume).
TMS measures provide functional integrity of CST, and DTI measures are used to
quantify structural characteristics of CST. Thus, a combination of DTI and TMS measures can be
used to examine the structural impairment of CST, and to determine whether the residual CST
can functionally deliver the motor commands from ipsilesional M1 to paretic limbs muscles.

Neurological Biomarkers of Post-Stroke Motor Recovery 135

Table A-1. fMRI-derived predictor variables
Reference fMRI-derived predictor variables
Cramer Degree of activation in iM1
Dong Laterality index of M1 & Voxel counts for iM1
Feydy fMRI activation patterns
Jang iSM1 BOLD signal during passive movement
Jung Resting state M1-M1 functional connectivity
Loubinoux fMRI activation patterns of ipsilesional SMA and inferior BA 40
Rehme
Functional connectivity among M1, SMA, and ventral PMC during
motor task
Várkuti
Resting state functional connectivity of iM1, cM1, visuospatial
system, SMA
Zarahn Task-related activation pattern
SM1, primary sensori-motor cortex; M1, primary motor cortex; SMA, supplementary motor
area; PMC, premotor cortex; i-, ipsilesional; c-, contralesional
Neurological Biomarkers of Post-Stroke Motor Recovery 136

APPENDIX C. INCLUSION AND EXCLUSION CRITERIA FOR DOSE RANDOMIZED
CONTROLLED TRIAL

Inclusion Criteria Exclusion Criteria
1. Diagnosis of stroke: ischemic or
intraparenchymal hemorrhagic stroke
without intraventricular extension with
confirmatory neuroimaging
2. Stroke onset is more than 180 days (6
months) before
3. Age >= 21 and no upper limit
4. Persistent hemiparesis leading to impaired
UE motor function
a. UE FM motor score >= 19 out of 66
b. At least a score of 1 on the hand item
for finger mass extention/grasp release
(Fritz et al., 2005)
5. Evidence of preserved cognitive function
a. MMSE: 24 or higher
6. No UE musculoskeletal injury or
conditions that limited use prior to the
stroke
7. Pre-stroke independence
a. Barthel index no less than 95
8. Judged medically stable to participate as
indicated by the patient’s primary care
physician
9. Expressed desire and ability by participant
with confirmation by the family/caregiver
to attend all testing evaluations for a
duration of approximately 10 months
10. Mostly resolved UE hemiparesis
11. UE FM motor and coordination score is
greater than 60 of 66
12. Severe UE sensory impairment
13. Anesthesia to light touch on the UE FM
sensation and proprioception
14. Neglect determined by unstructed
Mesulam test
15. Inability to give informed consent for
study participation
16. Current major depressive disorder
a. PHQ2 (depression screening survey)
greater than 3
17. Severe arthritis or orthopedic problems
that limit passive ROM of UE joints
a. Shoulder flexion & abduction < 90
b. Shoulder external rotation < 45
c. Elbow extension >20 from full
extension
d. Forearm supination and pronation >
45 from neutral
e. Wrist extension < neutral
f. Metacarpophalangeal and
interphalangeal joints extension > 30
from full extension

Neurological Biomarkers of Post-Stroke Motor Recovery 137

APPENDIX D. HISTOGRAM OF BEST LASSO LINEAR REGRESSION MODELS
FROM BOOTSTRAP SAMPLES

Neurological Biomarkers of Post-Stroke Motor Recovery 138

REFERENCES

Andersson, J.L.R., Sotiropoulos, S.N., 2016. An integrated approach to correction for off-
resonance effects and subject movement in diffusion MR imaging. Neuroimage 125, 1063–
78. doi:10.1016/j.neuroimage.2015.10.019
Arac, N., Sagduyu, A., Binai, S., Ertekin, C., 1994. Prognostic value of transcranial magnetic
stimulation in acute stroke [published erratum appears in Stroke 1995 Feb;26(2):336].
Stroke 25, 2183–2186. doi:10.1161/01.STR.25.11.2183
Assaf, Y., Pasternak, O., 2008. Diffusion tensor imaging (DTI)-based white matter mapping in
brain research: a review. J. Mol. Neurosci. 34, 51–61. doi:10.1007/s12031-007-0029-0
Atkins, D., Best, D., Briss, P.A., Eccles, M., Falck-Ytter, Y., Flottorp, S., Guyatt, G.H., Harbour,
R.T., Haugh, M.C., Henry, D., Hill, S., Jaeschke, R., Leng, G., Liberati, A., Magrini, N.,
Mason, J., Middleton, P., Mrukowicz, J., O’Connell, D., Oxman, A.D., Phillips, B.,
Schünemann, H.J., Edejer, T.T.-T., Varonen, H., Vist, G.E., Williams, J.W., Zaza, S., 2004.
Grading quality of evidence and strength of recommendations. BMJ 328, 1490.
doi:10.1136/bmj.328.7454.1490
Ballard, K.J., Azizi, L., Duffy, J.R., McNeil, M.R., Halaki, M., O’Dwyer, N., Layfield, C.,
Scholl, D.I., Vogel, A.P., Robin, D.A., 2016. A predictive model for diagnosing stroke-
related apraxia of speech. Neuropsychologia 81, 129–139.
doi:10.1016/j.neuropsychologia.2015.12.010
Bassett, D.S., Brown, J. a., Deshpande, V., Carlson, J.M., Grafton, S.T., 2011. Conserved and
variable architecture of human white matter connectivity. Neuroimage 54, 1262–1279.
doi:10.1016/j.neuroimage.2010.09.006
Neurological Biomarkers of Post-Stroke Motor Recovery 139

Benjamin, E.J., Blaha, M.J., Chiuve, S.E., Cushman, M., Das, S.R., Deo, R., de Ferranti, S.D.,
Floyd, J., Fornage, M., Gillespie, C., Isasi, C.R., Jiménez, M.C., Jordan, L.C., Judd, S.E.,
Lackland, D., Lichtman, J.H., Lisabeth, L., Liu, S., Longenecker, C.T., Mackey, R.H.,
Matsushita, K., Mozaffarian, D., Mussolino, M.E., Nasir, K., Neumar, R.W., Palaniappan,
L., Pandey, D.K., Thiagarajan, R.R., Reeves, M.J., Ritchey, M., Rodriguez, C.J., Roth,
G.A., Rosamond, W.D., Sasson, C., Towfighi, A., Tsao, C.W., Turner, M.B., Virani, S.S.,
Voeks, J.H., Willey, J.Z., Wilkins, J.T., Wu, J.H., Alger, H.M., Wong, S.S., Muntner, P.,
2017. Heart Disease and Stroke Statistics—2017 Update: A Report From the American
Heart Association. Circulation 135, e146–e603. doi:10.1161/CIR.0000000000000485
Bernhardt, J., Borschmann, K., Boyd, L., Thomas Carmichael, S., Corbett, D., Cramer, S.C.,
Hoffmann, T., Kwakkel, G., Savitz, S.I., Saposnik, G., Walker, M., Ward, N., 2016.
Moving rehabilitation research forward: Developing consensus statements for rehabilitation
and recovery research. Int. J. Stroke 11, 454–8. doi:10.1177/1747493016643851
Bernhardt, J., Hayward, K.S., Kwakkel, G., Ward, N.S., Wolf, S.L., Borschmann, K., Krakauer,
J.W., Boyd, L.A., Carmichael, S.T., Corbett, D., Cramer, S.C., 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
Bhushan, C., Haldar, J.P., Choi, S., Joshi, A.A., Shattuck, D.W., Leahy, R.M., 2014. Co-
registration and distortion correction of diffusion and anatomical images based on inverse
contrast normalization. Neuroimage 115, 30–33. doi:10.1016/j.neuroimage.2015.03.050
Bonita, R., Beaglehole, R., 1988. Recovery of motor function after stroke. Stroke. 19, 1497–
1500. doi:10.1002/dev.20508
Neurological Biomarkers of Post-Stroke Motor Recovery 140

Borich, M.R., Brown, K.E., Boyd, L. a, 2014. Motor skill learning is associated with diffusion
characteristics of white matter in individuals with chronic stroke. J. Neurol. Phys. Ther. 38,
151–60. doi:10.1097/NPT.0b013e3182a3d353
Boyd, L.A., Hayward, K.S., Ward, N.S., Stinear, C.M., Rosso, C., Fisher, R.J., Carter, A.R.,
Leff, A.P., Copland, D.A., Carey, L.M., Cohen, L.G., Basso, D.M., Maguire, J.M., Cramer,
S.C., 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
Boyke, J., Driemeyer, J., Gaser, C., Büchel, C., May, A., 2008. Training-induced brain structure
changes in the elderly. J. Neurosci. 28, 7031–7035. doi:10.1523/JNEUROSCI.0742-
08.2008
Breteler, M.M., van Amerongen, N.M., van Swieten, J.C., Claus, J.J., Grobbee, D.E., van Gijn,
J., Hofman, A., van Harskamp, F., 1994. Cognitive correlates of ventricular enlargement
and cerebral white matter lesions on magnetic resonance imaging. The Rotterdam Study.
Stroke 25, 1109–1115. doi:10.1161/01.STR.25.6.1109
Bucci, M., Mandelli, M.L., Berman, J.I., Amirbekian, B., Nguyen, C., Berger, M.S., Henry,
R.G., 2013. Quantifying diffusion MRI tractography of the corticospinal tract in brain
tumors with deterministic and probabilistic methods. NeuroImage Clin. 3, 361–368.
doi:10.1016/j.nicl.2013.08.008
Burke, E., Dobkin, B.H., Noser, E.A., Enney, L.A., Cramer, S.C., 2014. Predictors and
biomarkers of treatment gains in a clinical stroke trial targeting the lower extremity. Stroke.
45, 2379–84. doi:10.1161/STROKEAHA.114.005436
Chang, M.C., Do, K.H., Chun, M.H., 2015. Prediction of lower limb motor outcomes based on
Neurological Biomarkers of Post-Stroke Motor Recovery 141

transcranial magnetic stimulation findings in patients with an infarct of the anterior cerebral
artery. Somatosens. Mot. Res. 32, 249–253. doi:10.3109/08990220.2015.1091769
Chang, W.H., Park, E., Lee, J., Lee, A., Kim, Y.-H., 2017. Association Between Brain-Derived
Neurotrophic Factor Genotype and Upper Extremity Motor Outcome After Stroke. Stroke
48, 1457–1462. doi:10.1161/STROKEAHA.116.015264
Chen, S.-Y.Y., Winstein, C.J., 2009. A systematic review of voluntary arm recovery in
hemiparetic stroke: critical predictors for meaningful outcomes using the international
classification of functioning, disability, and health. J. Neurol. Phys. Ther. 33, 2–13.
doi:10.1097/NPT.0b013e318198a010
Cho, S.-H., Kim, D.G., Kim, D.-S., Kim, Y.-H., Lee, C.-H., Jang, S.H., 2007a. Motor outcome
according to the integrity of the corticospinal tract determined by diffusion tensor
tractography in the early stage of corona radiata infarct. Neurosci. Lett. 426, 123–7.
doi:10.1016/j.neulet.2007.08.049
Cho, S.-H., Kim, S.H., Choi, B.Y., Cho, S.H., Kang, J.H., Lee, C.-H., Byun, W.M., Jang, S.H.,
2007b. Motor outcome according to diffusion tensor tractography findings in the early stage
of intracerebral hemorrhage. Neurosci. Lett. 421, 142–6. doi:10.1016/j.neulet.2007.04.052
Cohen, J., Cohen, P., West, S.G., Aiken, L.S., 2013. Applied Multiple Regression/Correlation
Analysis for the Behavioral Sciences. Routledge.
Concha, L., Gross, D.W., Wheatley, B.M., Beaulieu, C., 2006. Diffusion tensor imaging of time-
dependent axonal and myelin degradation after corpus callosotomy in epilepsy patients.
Neuroimage 32, 1090–1099. doi:10.1016/j.neuroimage.2006.04.187
Corbetta, M., Ramsey, L., Callejas, A., Baldassarre, A., Hacker, C.D., Siegel, J.S., Astafiev, S.
V., Rengachary, J., Zinn, K., Lang, C.E., Connor, L.T., Fucetola, R., Strube, M., Carter,
Neurological Biomarkers of Post-Stroke Motor Recovery 142

A.R., Shulman, G.L., 2015. Common Behavioral Clusters and Subcortical Anatomy in
Stroke. Neuron 85, 927–941. doi:10.1016/j.neuron.2015.02.027
Coutts, S.B., O’Reilly, C., Hill, M.D., Steffenhagen, N., Poppe, A.Y., Boyko, M.J., Puetz, V.,
Demchuk, A.M., 2009. Computed tomography and computed tomography angiography
findings predict functional impairment in patients with minor stroke and transient ischaemic
attack. Int. J. Stroke 4, 448–53. doi:10.1111/j.1747-4949.2009.00346.x
Cramer, S.C., 2008. Repairing the human brain after stroke: I. Mechanisms of spontaneous
recovery. Ann. Neurol. 63, 272–287. doi:10.1002/ana.21393
Cramer, S.C., Parrish, T.B., Levy, R.M., Stebbins, G.T., Ruland, S.D., Lowry, D.W., Trouard,
T.P., Squire, S.W., Weinand, M.E., Savage, C.R., Wilkinson, S.B., Juranek, J., Leu, S.-
Y.Y., Himes, D.M., 2007. Predicting functional gains in a stroke trial. Stroke. 38, 2108–14.
doi:10.1161/STROKEAHA.107.485631
Cramer, S.C., Riley, J.D., 2008. Neuroplasticity and brain repair after stroke. Curr. Opin. Neurol.
21, 76–82. doi:10.1097/WCO.0b013e3282f36cb6
Cramer, S.C., Wolf, S.L., Adams, H.P., Chen, D., Dromerick, A.W., Dunning, K., Ellerbe, C.,
Grande, A., Janis, S., Lansberg, M.G., Lazar, R.M., Palesch, Y.Y., Richards, L., Roth, E.,
Savitz, S.I., Wechsler, L.R., Wintermark, M., Broderick, J.P., 2017. Stroke Recovery and
Rehabilitation Research. Stroke 48, 813–819. doi:10.1161/STROKEAHA.116.015501
Crinion, J., Ashburner, J., Leff, A., Brett, M., Price, C., Friston, K., 2007. Spatial normalization
of lesioned brains: Performance evaluation and impact on fMRI analyses. Neuroimage 37,
866–875. doi:10.1016/j.neuroimage.2007.04.065
Crofts, J.J., Higham, D.J., Bosnell, R., Jbabdi, S., Matthews, P.M., Behrens, T.E.J., Johansen-
Berg, H., 2011. Network analysis detects changes in the contralesional hemisphere
Neurological Biomarkers of Post-Stroke Motor Recovery 143

following stroke. Neuroimage 54, 161–169. doi:10.1016/j.neuroimage.2010.08.032
Dawes, H., Enzinger, C., Johansen-Berg, H., Bogdanovic, M., Guy, C., Collett, J., Izadi, H.,
Stagg, C., Wade, D., Matthews, P.M., 2008. Walking performance and its recovery in
chronic stroke in relation to extent of lesion overlap with the descending motor tract. Exp.
brain Res. 186, 325–33. doi:10.1007/s00221-007-1237-0
Dayan, E., Cohen, L.G., 2011. Neuroplasticity subserving motor skill learning. Neuron.
doi:10.1016/j.neuron.2011.10.008
Di Lazzaro, V., Profice, P., Pilato, F., Capone, F., Ranieri, F., Pasqualetti, P., Colosimo, C.,
Pravatà, E., Cianfoni, A., Dileone, M., 2010. Motor cortex plasticity predicts recovery in
acute stroke. Cereb. Cortex 20, 1523–8. doi:10.1093/cercor/bhp216
Dong, Y., Dobkin, B.H., Cen, S.Y., Wu, A.D., Winstein, C.J., 2006. Motor cortex activation
during treatment may predict therapeutic gains in paretic hand function after stroke. Stroke
37, 1552–1555. doi:10.1161/01.STR.0000221281.69373.4e
Escudero, J. V., Sancho, J., Bautista, D., Escudero, M., Lopez-Trigo, J., 1998. Prognostic Value
of Motor Evoked Potential Obtained by Transcranial Magnetic Brain Stimulation in Motor
Function Recovery in Patients With Acute Ischemic Stroke. Stroke 29, 1854–1859.
doi:10.1161/01.STR.29.9.1854
Feng, W., Wang, J., Chhatbar, P.Y., Doughty, C., Landsittel, D., Lioutas, V.-A., Kautz, S.A.,
Schlaug, G., 2015. Corticospinal tract lesion load: An imaging biomarker for stroke motor
outcomes. Ann. Neurol. n/a-n/a. doi:10.1002/ana.24510
Feydy, a., Carlier, R., Roby-Brami, A., Bussel, B., Cazalis, F., Pierot, L., Burnod, Y., Maier, M.
a. A., 2002. Longitudinal Study of Motor Recovery After Stroke: Recruitment and Focusing
of Brain Activation. Stroke 33, 1610–1617. doi:10.1161/01.STR.0000017100.68294.52
Neurological Biomarkers of Post-Stroke Motor Recovery 144

Feys, H., Van Hees, J., Bruyninckx, F., Mercelis, R., De Weerdt, W., 2000. Value of
somatosensory and motor evoked potentials in predicting arm recovery after a stroke. J.
Neurol. Neurosurg. Psychiatry 68, 323–31. doi:10.1136/jnnp.68.3.323
Fields, R.D., 2008. White matter in learning, cognition and psychiatric disorders. Trends
Neurosci. 31, 361–70. doi:10.1016/j.tins.2008.04.001
Gao, F., Grant, T.H., Roth, E.J., Zhang, L.Q., 2009. Changes in passive mechanical properties of
the gastrocnemius muscle at the muscle fascicle and joint levels in stroke survivors. Arch
Phys Med Rehabil 90, 819–826. doi:10.1016/j.apmr.2008.11.004
Glass, T.A., Matchar, D.B., Belyea, M., Feussner, J.R., 1993. Impact of social support on
outcome in first stroke. Stroke 24, 64–70. doi:10.1161/01.STR.24.1.64
Granziera, C., Daducci, A., Meskaldji, D.E., Roche, A., Maeder, P., Michel, P., Hadjikhani, N.,
Sorensen, A.G., Frackowiak, R.S., Thiran, J.-P., Meuli, R., Krueger, G., 2012. A new early
and automated MRI-based predictor of motor improvement after stroke. Neurology 79, 39–
46. doi:10.1212/WNL.0b013e31825f25e7
Grässel, D., Ringer, T.M., Fitzek, C., Fitzek, S., Kohl, M., Kaiser, W.A., Witte, O.W., Axer, H.,
2010. Wallerian degeneration of pyramidal tract after paramedian pons infarct.
Cerebrovasc. Dis. 30, 380–8. doi:10.1159/000319573
Grefkes, C., Fink, G.R., 2011. Reorganization of cerebral networks after stroke: new insights
from neuroimaging with connectivity approaches. Brain 134, 1264–76.
doi:10.1093/brain/awr033
Gregg, C., Shikar, V., Larsen, P., Mak, G., Chojnacki, A., Yong, V.W., Weiss, S., 2007. White
matter plasticity and enhanced remyelination in the maternal CNS. J. Neurosci. 27, 1812–
1823. doi:10.1523/JNEUROSCI.4441-06.2007
Neurological Biomarkers of Post-Stroke Motor Recovery 145

Gresham, G.E., 1986. Stroke outcome research. Stroke. 17, 358–60.
Groisser, B.N., Copen, W. a, Singhal, A.B., Hirai, K.K., Schaechter, J.D., 2014. Corticospinal
Tract Diffusion Abnormalities Early After Stroke Predict Motor Outcome. Neurorehabil.
Neural Repair 28, 1545968314521896-. doi:10.1177/1545968314521896
Gupta, R.K., Saksena, S., Hasan, K.M., Agarwal, A., Haris, M., Pandey, C.M., Narayana, P. a,
2006. Focal Wallerian degeneration of the corpus callosum in large middle cerebral artery
stroke: serial diffusion tensor imaging. J. Magn. Reson. Imaging 24, 549–55.
doi:10.1002/jmri.20677
Han, Y., Yang, H., Lv, Y.T., Zhu, C.Z., He, Y., Tang, H.H., Gong, Q.Y., Luo, Y.J., Zang, Y.F.,
Dong, Q., 2009. Gray matter density and white matter integrity in pianists’ brain: A
combined structural and diffusion tensor MRI study. Neurosci. Lett. 459, 3–6.
doi:10.1016/j.neulet.2008.07.056
Hand, P.J., Wardlaw, J.M., Rivers, C.S., Armitage, P.A., Bastin, M.E., Lindley, R.I., Dennis,
M.S., 2006. MR diffusion-weighted imaging and outcome prediction after ischemic stroke.
Neurology 66, 1159–63. doi:10.1212/01.wnl.0000202524.43850.81
Heiss, W.-D., Kidwell, C.S., 2014. Imaging for prediction of functional outcome and assessment
of recovery in ischemic stroke. Stroke. doi:10.1161/STROKEAHA.113.003611
Hendricks, H.T., Hageman, G., van Limbeek, J., 1997. Prediction of recovery from upper
extremity paralysis after stroke by measuring evoked potentials. Scand. J. Rehabil. Med. 29,
155–9.
Hendricks, H.T., Pasman, J.W., van Limbeek, J., Zwarts, M.J., 2003. Motor evoked potentials of
the lower extremity in predicting motor recovery and ambulation after stroke: a cohort
study. Arch. Phys. Med. Rehabil. 84, 1373–9. doi:10.1016/S0003-9993(03)00237-5
Neurological Biomarkers of Post-Stroke Motor Recovery 146

Hermann, D.M., Chopp, M., 2012. Promoting brain remodelling and plasticity for stroke
recovery: Therapeutic promise and potential pitfalls of clinical translation. Lancet Neurol.
doi:10.1016/S1474-4422(12)70039-X
Hier, D.B., Edelstein, G., 1991. Deriving clinical prediction rules from stroke outcome research.
Stroke 22, 1431–1436. doi:10.1161/01.STR.22.11.1431
Holmes, C.J., Hoge, R., Collins, L., Woods, R., Toga, A.W., Evans, A.C., 1998. Enhancement of
MR images using registration for signal averaging. J. Comput. Assist. Tomogr. 22, 324–33.
Holtmannspötter, M., Peters, N., Opherk, C., Martin, D., Herzog, J., Brückmann, H., Sämann, P.,
Gschwendtner, A., Dichgans, M., 2005. Diffusion magnetic resonance histograms as a
surrogate marker and predictor of disease progression in CADASIL: a two-year follow-up
study. Stroke. 36, 2559–65. doi:10.1161/01.STR.0000189696.70989.a4
Hosmer, D., Lemeshow, S., 2000. Applied Logistic Regression, 2nd Editio. ed. John Wiley &
Sons, New York, NY, USA.
Jang, S.H., 2011. A review of diffusion tensor imaging studies on motor recovery mechanisms in
stroke patients. NeuroRehabilitation 28, 345–52. doi:10.3233/NRE-2011-0662
Jang, S.H., Ahn, S.H., Sakong, J., Byun, W.M., Choi, B.Y., Chang, C.H., Bai, D., Son, S.M.,
2010. Comparison of TMS and DTT for predicting motor outcome in intracerebral
hemorrhage. J. Neurol. Sci. 290, 107–11. doi:10.1016/j.jns.2009.10.019
Jang, S.H., Cho, S.-H., Kim, Y.-H., Han, B.S., Byun, W.M., Son, S.-M., Kim, S.H., Lee, S.J.,
2005. Diffusion anisotrophy in the early stages of stroke can predict motor outcome. Restor.
Neurol. Neurosci. 23, 11–7.
Jang, S.H., Kim, Y.-H., Chang, Y., Han, B.S., Byun, W.M., Chang, C.H., 2004. The predictive
value of cortical activation by passive movement for motor recovery in stroke patients.
Neurological Biomarkers of Post-Stroke Motor Recovery 147

Restor. Neurol. Neurosci. 22, 59–63.
Johansen-Berg, H., Scholz, J., Stagg, C.J., 2010. Relevance of structural brain connectivity to
learning and recovery from stroke. Front. Syst. Neurosci. 4, 146.
doi:10.3389/fnsys.2010.00146
Jones, D.K., Knösche, T.R., Turner, R., 2013. NeuroImage White matter integrity , fi ber count ,
and other fallacies : The do s and don ts of diffusion MRI. Neuroimage 73, 239–254.
doi:10.1016/j.neuroimage.2012.06.081
Joshi, A.A., Shattuck, D.W., Leahy, R.M., 2012. A Method for Automated Cortical Surface
Registration and Labeling. Biomed. image Regist. ... proceedings. WBIR (Workshop
2006- ) 7359, 180–189. doi:10.1007/978-3-642-31340-0_19
Jung, T.-D., Kim, J.-Y., Seo, J.-H., Jin, S.-U., Lee, H.J., Lee, S.-H., Lee, Y.-S., Chang, Y., 2013.
Combined information from resting-state functional connectivity and passive movements
with functional magnetic resonance imaging differentiates fast late-onset motor recovery
from progressive recovery in hemiplegic stroke patients: a pilot study. J. Rehabil. Med. 45,
546–52. doi:10.2340/16501977-1165
Kang, H.-J., Stewart, R., Park, M.-S., Bae, K.-Y., Kim, S.-W., Kim, J.-M., Shin, I.-S., Cho, K.-
H., Yoon, J.-S., 2013. White matter hyperintensities and functional outcomes at 2 weeks
and 1 year after stroke. Cerebrovasc. Dis. 35, 138–45. doi:10.1159/000346604
Karibe, H., Shimizu, H., Tominaga, T., Koshu, K., Yoshimoto, T., 2000. Diffusion-weighted
magnetic resonance imaging in the early evaluation of corticospinal tract injury to predict
functional motor outcome in patients with deep intracerebral hemorrhage. J. Neurosurg. 92,
58–63. doi:10.3171/jns.2000.92.1.0058
Kasner, S.E., 2006. Clinical interpretation and use of stroke scales. Lancet Neurol. 5, 603–612.
Neurological Biomarkers of Post-Stroke Motor Recovery 148

doi:10.1016/S1474-4422(06)70495-1
Kelly, P.J., Furie, K.L., Shafqat, S., Rallis, N., Chang, Y., Stein, J., 2003. Functional recovery
following rehabilitation after hemorrhagic and ischemic stroke. Arch. Phys. Med. Rehabil.
84, 968–972. doi:10.1016/S0003-9993(03)00040-6
Kim, B., Winstein, C., 2017. Can Neurological Biomarkers of Brain Impairment Be Used to
Predict Poststroke Motor Recovery? A Systematic Review. Neurorehabil. Neural Repair 31,
3–24. doi:10.1177/1545968316662708
Kim, J.H., Haldar, J., Liang, Z.-P., Song, S.-K., 2009. Diffusion tensor imaging of mouse brain
stem and cervical spinal cord. J. Neurosci. Methods 176, 186–91.
doi:10.1016/j.jneumeth.2008.09.005
Kim, K.H., Kim, Y.-H., Kim, M.S., Park, C., Lee, A., Chang, W.H., 2015. Prediction of Motor
Recovery Using Diffusion Tensor Tractography in Supratentorial Stroke Patients With
Severe Motor Involvement. Ann. Rehabil. Med. 39, 570. doi:10.5535/arm.2015.39.4.570
Kirschen, R.H., O’higgins, E.A., Lee, R.T., 2000. The Royal London Space Planning: an
integration of space analysis and treatment planning: Part I: Assessing the space required to
meet treatment objectives. Am. J. Orthod. Dentofacial Orthop. 118, 448–55.
doi:10.1067/mod.2000.109031
Kitago, T., Krakauer, J.W., 2013. Motor learning principles for neurorehabilitation, in:
Handbook of Clinical Neurology. pp. 93–103. doi:10.1016/B978-0-444-52901-5.00008-3
Kleim, J.A., Chan, S., Pringle, E., Schallert, K., Procaccio, V., Jimenez, R., Cramer, S.C., 2006.
BDNF val66met polymorphism is associated with modified experience-dependent plasticity
in human motor cortex. Nat. Neurosci. 9, 735–737. doi:10.1038/nn1699
Kleim, J. a, Barbay, S., Cooper, N.R., Hogg, T.M., Reidel, C.N., Remple, M.S., Nudo, R.J.,
Neurological Biomarkers of Post-Stroke Motor Recovery 149

2002. Motor learning-dependent synaptogenesis is localized to functionally reorganized
motor cortex. Neurobiol. Learn. Mem. 77, 63–77. doi:10.1006/nlme.2000.4004
Koh, C.-L., Pan, S.-L., Jeng, J.-S., Chen, B.-B., Wang, Y.-H., Hsueh, I.-P., Hsieh, C.-L., 2015.
Predicting Recovery of Voluntary Upper Extremity Movement in Subacute Stroke Patients
with Severe Upper Extremity Paresis. PLoS One 10, e0126857.
doi:10.1371/journal.pone.0126857
Koski, L., Mernar, T.J., Dobkin, B.H., 2004. Immediate and long-term changes in corticomotor
output in response to rehabilitation: correlation with functional improvements in chronic
stroke. Neurorehabil. Neural Repair 18, 230–249. doi:10.1177/1545968304269210
Koyama, T., Tsuji, M., Miyake, H., Ohmura, T., Domen, K., 2012. Motor outcome for patients
with acute intracerebral hemorrhage predicted using diffusion tensor imaging: an
application of ordinal logistic modeling. J. Stroke Cerebrovasc. Dis. 21, 704–11.
doi:10.1016/j.jstrokecerebrovasdis.2011.03.004
Koyama, T., Tsuji, M., Nishimura, H., Miyake, H., Ohmura, T., Domen, K., 2013. Diffusion
tensor imaging for intracerebral hemorrhage outcome prediction: comparison using data
from the corona radiata/internal capsule and the cerebral peduncle. J. Stroke Cerebrovasc.
Dis. 22, 72–9. doi:10.1016/j.jstrokecerebrovasdis.2011.06.014
Krakauer, J.W., 2004. Functional imaging of motor recovery after stroke: remaining challenges.
Curr. Neurol. Neurosci. Rep. 4, 42–6.
Krakauer, J.W., 2005. Arm function after stroke: from physiology to recovery. Semin. Neurol.
25, 384–95. doi:10.1055/s-2005-923533
Krakauer, J.W., 2006. Motor learning: its relevance to stroke recovery and neurorehabilitation.
Curr. Opin. Neurol. 19, 84–90.
Neurological Biomarkers of Post-Stroke Motor Recovery 150

Krakauer, J.W., Carmichael, S.T., Corbett, D., Wittenberg, G.F., 2012. Getting
neurorehabilitation right: what can be learned from animal models? Neurorehabil. Neural
Repair 26, 923–31. doi:10.1177/1545968312440745
Kusano, Y., Seguchi, T., Horiuchi, T., Kakizawa, Y., Kobayashi, T., Tanaka, Y., Seguchi, K.,
Hongo, K., 2009. Prediction of functional outcome in acute cerebral hemorrhage using
diffusion tensor imaging at 3T: a prospective study. AJNR. Am. J. Neuroradiol. 30, 1561–5.
doi:10.3174/ajnr.A1639
Kuzu, Y., Inoue, T., Kanbara, Y., Nishimoto, H., Fujiwara, S., Ogasawara, K., Ogawa, A., 2012.
Prediction of motor function outcome after intracerebral hemorrhage using fractional
anisotropy calculated from diffusion tensor imaging. Cerebrovasc. Dis. 33, 566–73.
doi:10.1159/000338904
Kwakkel, G., Kollen, B.J., 2013. Predicting activities after stroke: What is clinically relevant?
Int. J. Stroke. doi:10.1111/j.1747-4949.2012.00967.x
Kwakkel, G., Wagenaar, R.C., Kollen, B.J., Lankhorst, G.J., 1996. Predicting disability in
stroke--a critical review of the literature. Age Ageing 25, 479–489.
Kwon, H.G., Hong, J.H., Jang, S.H., 2011. Anatomic location and somatotopic arrangement of
the corticospinal tract at the cerebral peduncle in the human brain. AJNR. Am. J.
Neuroradiol. 32, 2116–9. doi:10.3174/ajnr.A2660
Kwon, Y.H., Jeoung, Y.J., Lee, J., Son, S.M., Kim, S., Kim, C., Jang, S.H., 2012. Predictability
of motor outcome according to the time of diffusion tensor imaging in patients with cerebral
infarct. Neuroradiology 54, 691–7. doi:10.1007/s00234-011-0972-x
Lai, C., Wang, C., Tsai, P., Chan, R., Lin, S., Lin, F., Hsieh, C., 2015. Corticospinal Integrity
and Motor Impairment Predict Outcomes After Excitatory Repetitive Transcranial Magnetic
Neurological Biomarkers of Post-Stroke Motor Recovery 151

Stimulation : A Preliminary Study. Arch. Phys. Med. Rehabil. 96, 69–75.
doi:10.1016/j.apmr.2014.08.014
Lai, C., Zhang, S.Z., Liu, H.M., Zhou, Y.B., Zhang, Y.Y., Zhang, Q.W., Han, G.C., 2007. White
matter tractography by diffusion tensor imaging plays an important role in prognosis
estimation of acute lacunar infarctions. Br. J. Radiol. 80, 782–9. doi:10.1259/bjr/99366083
Lang, C.E., Edwards, D.F., Birkenmeier, R.L., Dromerick, A.W., 2008. Estimating Minimal
Clinically Important Differences of Upper-Extremity Measures Early After Stroke. Arch.
Phys. Med. Rehabil. 89, 1693–1700. doi:10.1016/j.apmr.2008.02.022
Langhorne, P., Coupar, F., Pollock, A., 2009. Motor recovery after stroke: a systematic review.
Lancet Neurol. 8, 741–754. doi:10.1016/S1474-4422(09)70150-4
Levin, M.F., Kleim, J.A., Wolf, S.L., 2009. What do motor “recovery” and “compensation”
mean in patients following stroke? Neurorehabil. Neural Repair 23, 313–319.
doi:10.1177/1545968308328727
Levin, M.F., Michaelsen, S.M., Cirstea, C.M., Roby-Brami, A., 2002. Use of the trunk for
reaching targets placed within and beyond the reach in adult hemiparesis. Exp. Brain Res.
143, 171–180. doi:10.1007/s00221-001-0976-6
Liew, S.-L., Jahanshad, N., Aziz-Zadeh, L., Birbaumer, N., Borich, M., Boyd, L., Byblow, W.,
Craddock, C., Dimyan, M., Ermer, E., Goud, A., Lang, C.E., Li, J., Liu, J., Nichols, T.,
Ramos, A., Roberts, P., Sanossian, N., Soekadar, S., Stinear, C., Ward, N., Wang, J.,
Westlye, L.T., Kuceyeski, A., Winstein, C.J., Wittenberg, G.F., Yu, C., Cramer, S.C.,
Thompson, P.M., 2017. Abstract 14: Effects of Lesion Laterality on Post-Stroke Motor
Performance: An ENIGMA Stroke Recovery Analysis. Stroke 48, A14.
Liggins, J.T.P., Yoo, A.J., Mishra, N.K., Wheeler, H.M., Straka, M., Leslie-Mazwi, T.M.,
Neurological Biomarkers of Post-Stroke Motor Recovery 152

Chaudhry, Z.A., Kemp, S., Mlynash, M., Bammer, R., Albers, G.W., Lansberg, M.G., 2013.
A score based on age and DWI volume predicts poor outcome following endovascular
treatment for acute ischemic stroke. Int. J. Stroke 44. doi:10.1111/ijs.12207
Lin, J.-H., Hsu, M.-J., Sheu, C.-F., Wu, T.-S., Lin, R.-T., Chen, C.-H., Hsieh, C.-L., 2009.
Psychometric comparisons of 4 measures for assessing upper-extremity function in people
with stroke. Phys. Ther. 89, 840–850. doi:10.2522/ptj.20080285
Lin, K.C., Hsieh, Y.W., Wu, C.Y., Chen, C.L., Jang, Y., Liu, J.S., 2009. Minimal detectable
change and clinically important difference of the Wolf Motor Function Test in stroke
patients. Neurorehabil Neural Repair 23, 429–434. doi:10.1177/1545968308331144
Lindenberg, R., Renga, V., Zhu, L.L., Betzler, F., Alsop, D., Schlaug, G., 2010. Structural
integrity of corticospinal motor fibers predicts motor impairment in chronic stroke.
Neurology 74, 280–287. doi:10.1212/WNL.0b013e3181ccc6d9
Lindenberg, R., Zhu, L.L., Rüber, T., Schlaug, G., 2012. Predicting functional motor potential in
chronic stroke patients using diffusion tensor imaging. Hum. Brain Mapp. 33, 1040–1051.
doi:10.1002/hbm.21266
Lindenberg, R., Zhu, L.L., Schlaug, G., 2009. Corticospinal Tract Integrity Predicts Recovery
Potential in Chronic Stroke Patients. Ann. Neurol. 66, S30–S30.
Liu, X., Tian, W., Qiu, X., Li, J., Thomson, S., Li, L., Wang, H.Z., 2012. Correlation analysis of
quantitative diffusion parameters in ipsilateral cerebral peduncle during Wallerian
degeneration with motor function outcome after cerebral ischemic stroke. J. Neuroimaging
22, 255–60. doi:10.1111/j.1552-6569.2011.00617.x
Livingston-Thomas, J., Nelson, P., Karthikeyan, S., Antonescu, S., Jeffers, M.S., Marzolini, S.,
Corbett, D., 2016. Exercise and Environmental Enrichment as Enablers of Task-Specific
Neurological Biomarkers of Post-Stroke Motor Recovery 153

Neuroplasticity and Stroke Recovery. Neurotherapeutics 13, 395–402. doi:10.1007/s13311-
016-0423-9
Loubinoux, I., Carel, C., Pariente, J., Dechaumont, S., Albucher, J.-F., Marque, P., Manelfe, C.,
Chollet, F., 2003. Correlation between cerebral reorganization and motor recovery after
subcortical infarcts. Neuroimage 20, 2166–80. doi:10.1016/j.neuroimage.2003.08.017
Ma, C., Liu, A., Li, Z., Zhou, X., Zhou, S., 2014. Longitudinal study of diffusion tensor imaging
properties of affected cortical spinal tracts in acute and chronic hemorrhagic stroke. J. Clin.
Neurosci. 21, 1388–1392. doi:10.1016/j.jocn.2013.11.032
Ma, L., Gao, P., Hu, Q., Lin, Y., Jing, L., Xue, J., Chen, Z., Wang, Y., Liu, M., Cai, Y., 2011.
Effect of baseline magnetic resonance imaging (MRI) apparent diffusion coefficient lesion
volume on functional outcome in ischemic stroke. Neurol. Res. 33, 494–502.
doi:10.1179/016164111X13007856084124
Maeshima, S., Osawa, A., Nishio, D., Hirano, Y., Kigawa, H., Takeda, H., 2013. Diffusion
tensor MR imaging of the pyramidal tract can predict the need for orthosis in hemiplegic
patients with hemorrhagic stroke. Neurol. Sci. 34, 1765–70. doi:10.1007/s10072-013-1330-
z
Marumoto, K., Koyama, T., Hosomi, M., Takebayashi, T., Hanada, K., Ikeda, S., Kodama, N.,
Domen, K., 2013. Diffusion tensor imaging predicts the outcome of constraint-induced
movement therapy in chronic infarction patients with hemiplegia: A pilot study. Restor.
Neurol. Neurosci. 31, 387–96. doi:10.3233/RNN-120285
Massey, F.J., 1951. The Kolmogorov-Smirnov Test for Goodness of Fit. J. Am. Stat. Assoc. 46,
68–78. doi:10.1080/01621459.1951.10500769
May, A., 2011. Experience-dependent structural plasticity in the adult human brain. Trends
Neurological Biomarkers of Post-Stroke Motor Recovery 154

Cogn. Sci. doi:10.1016/j.tics.2011.08.002
Michaelsen, S.M., Jacobs, S., Roby-Brami, A., Levin, M.F., 2004. Compensation for distal
impairments of grasping in adults with hemiparesis. Exp Brain Res 157, 162–173.
doi:10.1007/s00221-004-1829-x
Muñoz Maniega, S., Bastin, M.E., Armitage, P.A., Farrall, A.J., Carpenter, T.K., Hand, P.J.,
Cvoro, V., Rivers, C.S., Wardlaw, J.M., 2004. Temporal evolution of water diffusion
parameters is different in grey and white matter in human ischaemic stroke. J. Neurol.
Neurosurg. Psychiatry 75, 1714–8. doi:10.1136/jnnp.2003.033852
Nascimbeni, A., Gaffuri, A., Imazio, P., 2006. Motor evoked potentials: prognostic value in
motor recovery after stroke. Funct. Neurol. 21, 199–203.
Nouri, S., Cramer, S.C., 2011. Anatomy and physiology predict response to motor cortex
stimulation after stroke. Neurology 77, 1076–83. doi:10.1212/WNL.0b013e31822e1482
Nudo, R.J., 2013. Recovery after brain injury: mechanisms and principles. Front. Hum.
Neurosci. 7, 887. doi:10.3389/fnhum.2013.00887
Nuutinen, J., Liu, Y., Laakso, M.P., Karonen, J.O., Roivainen, R., Vanninen, R.L., Partanen, K.,
Østergaard, L., Sivenius, J., Aronen, H.J., 2006. Assessing the outcome of stroke: a
comparison between MRI and clinical stroke scales. Acta Neurol. Scand. 113, 100–7.
doi:10.1111/j.1600-0404.2005.00550.x
O’Shea, J., Boudrias, M.-H.H., Stagg, C.J., Bachtiar, V., Kischka, U., Blicher, J.U., Johansen-
Berg, H., 2014. Predicting behavioural response to TDCS in chronic motor stroke.
Neuroimage 85, 924–933. doi:10.1016/j.neuroimage.2013.05.096
Park, C.-H., Kou, N., Boudrias, M.-H., Playford, E.D., Ward, N.S., 2013. Assessing a
standardised approach to measuring corticospinal integrity after stroke with DTI.
Neurological Biomarkers of Post-Stroke Motor Recovery 155

NeuroImage. Clin. 2, 521–33. doi:10.1016/j.nicl.2013.04.002
Piron, L., Piccione, F., Tonin, P., Dam, M., 2005. Clinical correlation between motor evoked
potentials and gait recovery in poststroke patients. Arch. Phys. Med. Rehabil. 86, 1874–8.
doi:10.1016/j.apmr.2005.03.007
Puig, J., Blasco, G., Daunis-I-Estadella, J., Thomalla, G., Castellanos, M., Figueras, J., Remollo,
S., van Eendenburg, C., Sánchez-González, J., Serena, J., Pedraza, S., 2013. Decreased
corticospinal tract fractional anisotropy predicts long-term motor outcome after stroke.
Stroke. 44, 2016–8. doi:10.1161/STROKEAHA.111.000382
Puig, J., Blasco, G., Schlaug, G., Stinear, C.M., Daunis-i-Estadella, P., Biarnes, C., Figueras, J.,
Serena, J., Hernández-Pérez, M., Alberich-Bayarri, A., Castellanos, M., Liebeskind, D.S.,
Demchuk, A.M., Menon, B.K., Thomalla, G., Nael, K., Wintermark, M., Pedraza, S., 2017.
Diffusion tensor imaging as a prognostic biomarker for motor recovery and rehabilitation
after stroke. Neuroradiology 59, 343–351. doi:10.1007/s00234-017-1816-0
Puig, J., Pedraza, S., Blasco, G., Daunis-i-Estadella, J., Prados, F., Remollo, S., Prats-Galino, A.,
Soria, G., Boada, I., Castellanos, M., Serena, J., 2011. Acute Damage to the Posterior Limb
of the Internal Capsule on Diffusion Tensor Tractography as an Early Imaging Predictor of
Motor Outcome after Stroke. Am. J. Neuroradiol. 32, 857–863. doi:10.3174/ajnr.A2400
Puig, J., Pedraza, S., Blasco, G., Daunis-I-Estadella, J., Prats, A., Prados, F., Boada, I.,
Castellanos, M., Sánchez-González, J., Remollo, S., Laguillo, G., Quiles, A.M., Gómez, E.,
Serena, J., 2010. Wallerian degeneration in the corticospinal tract evaluated by diffusion
tensor imaging correlates with motor deficit 30 days after middle cerebral artery ischemic
stroke. AJNR. Am. J. Neuroradiol. 31, 1324–30. doi:10.3174/ajnr.A2038
Qiu, M., Darling, W.G., Morecraft, R.J., Ni, C.C., Rajendra, J., Butler, A.J., 2011. White Matter
Neurological Biomarkers of Post-Stroke Motor Recovery 156

Integrity Is a Stronger Predictor of Motor Function Than BOLD Response in Patients With
Stroke. Neurorehabil. Neural Repair 25, 275–284. doi:10.1177/1545968310389183
Qiu, T.-M., Zhang, Y., Wu, J.S., 2008. Preliminary application of pyramidal tractography in
evaluating prognosis of patients with hypertensive intracerebral hemorrhage. Acta
Neurochir. Suppl. 105, 165–70. doi:10.1007/978-3-211-09469-3_33
Quallo, M.M., Price, C.J., Ueno, K., Asamizuya, T., Cheng, K., Lemon, R.N., Iriki, A., 2009.
Gray and white matter changes associated with tool-use learning in macaque monkeys.
Proc. Natl. Acad. Sci. U. S. A. 106, 18379–84. doi:10.1073/pnas.0909751106
Quinlan, E.B., Dodakian, L., Otr, L., See, J., Mckenzie, A., Le, V., Wojnowicz, M., Shahbaba,
B., Cramer, S.C., 2014. Neural Function , Injury , and Stroke Subtype Predict Treatment
Gains after Stroke. Ann. Neurol. doi:10.1002/ana.24309
Ramos-Cejudo, J., Gutierrez-Fernandez, M., Otero-Ortega, L., Rodriguez-Frutos, B., Fuentes, B.,
Vallejo-Cremades, M.T., Hernanz, T.N., Cerdan, S., Diez-Tejedor, E., 2015. Brain-Derived
Neurotrophic Factor Administration Mediated Oligodendrocyte Differentiation and Myelin
Formation in Subcortical Ischemic Stroke. Stroke 46, 221–228.
doi:10.1161/STROKEAHA.114.006692
Rapisarda, G., Bastings, E., de Noordhout, A.M., Pennisi, G., Delwaide, P.J.J., 1996. Can motor
recovery in stroke patients be predicted by early transcranial magnetic stimulation? Stroke.
27, 2191–6. doi:10.1161/01.STR.27.12.2191
Rehme, A.K., Eickhoff, S.B., Wang, L.E., Fink, G.R., Grefkes, C., 2011. Dynamic causal
modeling of cortical activity from the acute to the chronic stage after stroke. Neuroimage
55, 1147–58. doi:10.1016/j.neuroimage.2011.01.014
Rehme, A.K., Volz, L.J., Feis, D., Eickhoff, S.B., Fink, G.R., Grefkes, C., 2015. Individual
Neurological Biomarkers of Post-Stroke Motor Recovery 157

prediction of chronic motor outcome in the acute post-stroke stage: Behavioral parameters
versus functional imaging. Hum. Brain Mapp. 36, 4553–65. doi:10.1002/hbm.22936
Rickards, T., Sterling, C., Taub, E., Perkins-Hu, C., Gauthier, L., Graham, M., Griffin, A., Davis,
D., Mark, V.W., Uswatte, G., 2014. Diffusion tensor imaging study of the response to
constraint-induced movement therapy of children with hemiparetic cerebral palsy and adults
with chronic stroke. Arch. Phys. Med. Rehabil. 95, 506–514.e1.
doi:10.1016/j.apmr.2013.08.245
Rickards, T., Taub, E., Sterling, C., Graham, M.J., Barghi, A., Uswatte, G., Mark, V.W., 2012.
Brain parenchymal fraction predicts motor improvement following intensive task-oriented
motor rehabilitation for chronic stroke. Restor. Neurol. Neurosci. 30, 355–61.
doi:10.3233/RNN-2012-110211
Rosso, C., Colliot, O., Pires, C., Delmaire, C., Valabrègue, R., Crozier, S., Dormont, D., Baillet,
S., Samson, Y., Lehéricy, S., 2010. Early ADC changes in motor structures predict outcome
of acute stroke better than lesion volume. J. Neuroradiol. 38, 105–112.
doi:10.1016/j.neurad.2010.05.001
Sachdev, P., 2004. Homocysteine, cerebrovascular disease and brain atrophy. J. Neurol. Sci. 226,
25–29. doi:10.1016/j.jns.2004.09.006
Sandor, S., Leahy, R., 1997. Surface-based labeling of cortical anatomy using a deformable atlas.
IEEE Trans. Med. Imaging 16, 41–54. doi:10.1109/42.552054
Sato, Y., Chin, Y., Kato, T., Tanaka, Y., Tozuka, Y., Mase, M., Ageyama, N., Ono, F., Terao,
K., Yoshikawa, Y., Hisatsune, T., 2009. White matter activated glial cells produce BDNF in
a stroke model of monkeys. Neurosci. Res. 65, 71–78. doi:10.1016/j.neures.2009.05.010
Schaechter, J.D., Moore, C.I., Connell, B.D., Rosen, B.R., Dijkhuizen, R.M., 2006. Structural
Neurological Biomarkers of Post-Stroke Motor Recovery 158

and functional plasticity in the somatosensory cortex of chronic stroke patients. Brain 129,
2722–2733. doi:10.1093/brain/awl214
Schiemanck, S.K., Kwakkel, G., Post, M.W.M., Kappelle, L.J., Prevo, A.J.H., 2006a. Predicting
Long-Term Independency in Activities of Daily Living After Middle Cerebral Artery
Stroke: Does Information From MRI Have Added Predictive Value Compared With
Clinical Information? Stroke 37, 1050–1054. doi:10.1161/01.STR.0000206462.09410.6f
Schiemanck, S.K., Kwakkel, G., Post, M.W.M., Prevo, A.J.H., 2006b. Predictive Value of
Ischemic Lesion Volume Assessed With Magnetic Resonance Imaging for Neurological
Deficits and Functional Outcome Poststroke: A Critical Review of the Literature.
Neurorehabil. Neural Repair 20, 492–502. doi:10.1177/1545968306289298
Schlaug, G., Marchina, S., Norton, A., 2009. Evidence for plasticity in white-matter tracts of
patients with chronic broca’s aphasia undergoing intense intonation-based speech therapy,
in: Annals of the New York Academy of Sciences. pp. 385–394. doi:10.1111/j.1749-
6632.2009.04587.x
Scholz, J., Klein, M.C., Behrens, T.E.J., Johansen-Berg, H., 2009. Training induces changes in
white-matter architecture. Nat. Neurosci. 12, 1370–1. doi:10.1038/nn.2412
Schwamm, L.H., Koroshetz, W.J., Sorensen, A.G., Wang, B., Copen, W.A., Budzik, R., Rordorf,
G., Buonanno, F.S., Schaefer, P.W., Gonzalez, R.G., 1998. Time Course of Lesion
Development in Patients With Acute Stroke : Serial Diffusion- and Hemodynamic-
Weighted Magnetic Resonance Imaging. Stroke 29, 2268–2276.
doi:10.1161/01.STR.29.11.2268
Shattuck, D.W., Sandor-Leahy, S.R., Schaper, K.A., Rottenberg, D.A., Leahy, R.M., 2001.
Magnetic Resonance Image Tissue Classification Using a Partial Volume Model.
Neurological Biomarkers of Post-Stroke Motor Recovery 159

Neuroimage 13, 856–876. doi:10.1006/nimg.2000.0730
Soares, J.M., Marques, P., Alves, V., Sousa, N., 2013. A hitchhiker’s guide to diffusion tensor
imaging. Front. Neurosci. 7, 31. doi:10.3389/fnins.2013.00031
Song, F., Zhang, F., Yin, D.-Z., Hu, Y.-S., Fan, M.-X., Ni, H.-H., Nan, X.-L., Cui, X., Zhou, C.-
X., Huang, C.-S., Zhao, Q., Ma, L.-H., Xu, Y.-M., Xia, Q.-J., 2012. Diffusion tensor
imaging for predicting hand motor outcome in chronic stroke patients. J. Int. Med. Res. 40,
126–33.
Song, S.-K., Sun, S.-W., Ju, W.-K., Lin, S.-J., Cross, A.H., Neufeld, A.H., 2003. Diffusion
tensor imaging detects and differentiates axon and myelin degeneration in mouse optic
nerve after retinal ischemia. Neuroimage 20, 1714–1722.
doi:10.1016/j.neuroimage.2003.07.005
Soper, D.S., 2015. Post-hoc Statistical Power Calculator for Multiple Regression [Software]
[WWW Document]. URL http://www.danielsoper.com/statcalc
Spampinato, M.V., Chan, C., Jensen, J.H., Helpern, J.A., Bonilha, L., Kautz, S.A., Nietert, P.J.,
Feng, W., 2017. Diffusional Kurtosis Imaging and Motor Outcome in Acute Ischemic
Stroke. Am. J. Neuroradiol. 38, 1328–1334. doi:10.3174/ajnr.A5180
Steele, C.J., Bailey, J. a, Zatorre, R.J., Penhune, V.B., 2013. Early musical training and white-
matter plasticity in the corpus callosum: evidence for a sensitive period. J. Neurosci. 33,
1282–90. doi:10.1523/JNEUROSCI.3578-12.2013
Sterr, A., Dean, P.J. a., Szameitat, a. J., Conforto, a. B., Shen, S., 2014. Corticospinal Tract
Integrity and Lesion Volume Play Different Roles in Chronic Hemiparesis and Its
Improvement Through Motor Practice. Neurorehabil. Neural Repair 28, 335–343.
doi:10.1177/1545968313510972
Neurological Biomarkers of Post-Stroke Motor Recovery 160

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., Barber, P.A., Petoe, M., Anwar, S., Byblow, W.D., 2012. The PREP algorithm
predicts potential for upper limb recovery after stroke. Brain 135, 2527–2535.
doi:10.1093/brain/aws146
Stinear, C.M., Barber, P.A., Smale, P.R., Coxon, J.P., Fleming, M.K., Byblow, W.D., 2007.
Functional potential in chronic stroke patients depends on corticospinal tract integrity. Brain
130, 170–80. doi:10.1093/brain/awl333
Stinear, C.M., Byblow, W.D., 2014. Predicting and accelerating motor recovery after stroke.
Curr. Opin. Neurol. 27, 624–630. doi:10.1097/WCO.0000000000000153
Stinear, C.M., Byblow, W.D., Ackerley, S.J., Barber, P.A., Smith, M.-C., 2017. Predicting
Recovery Potential for Individual Stroke Patients Increases Rehabilitation Efficiency.
Stroke 48, 1011–1019. doi:10.1161/STROKEAHA.116.015790
Stinear, C.M., Byblow, W.D., Ward, S.H., 2014. An update on predicting motor recovery after
stroke. Ann. Phys. Rehabil. Med. 57, 489–498. doi:10.1016/j.rehab.2014.08.006
Stinear, Stinear, C.M., 2010. Prediction of recovery of motor function after stroke. Lancet
Neurol. 9, 1228–1232. doi:10.1016/S1474-4422(10)70247-7
Symposium recommendations for methodology in stroke outcome research. Task Force on
Stroke Impairment, Task Force on Stroke Disability, and Task Force on Stroke Handicap.,
1990. . Stroke. 21, II68-73.
Tao, W.-D., Wang, J., Schlaug, G., Liu, M., Selim, M.H., 2014. A Comparative Study of
Fractional Anisotropy Measures and ICH Score in Predicting Functional Outcomes After
Intracerebral Hemorrhage. Neurocrit. Care 21, 417–425. doi:10.1007/s12028-014-9999-2
Neurological Biomarkers of Post-Stroke Motor Recovery 161

Taubert, M., Draganski, B., Anwander, A., Müller, K., Horstmann, A., Villringer, A., Ragert, P.,
2010. Dynamic properties of human brain structure: learning-related changes in cortical
areas and associated fiber connections. J. Neurosci. 30, 11670–7.
doi:10.1523/JNEUROSCI.2567-10.2010
Taubert, M., Lohmann, G., Margulies, D.S., Villringer, A., Ragert, P., 2011. Long-term effects
of motor training on resting-state networks and underlying brain structure. Neuroimage 57,
1492–8. doi:10.1016/j.neuroimage.2011.05.078
Thiebaut de Schotten, M., Ffytche, D.H., Bizzi, A., Dell’Acqua, F., Allin, M., Walshe, M.,
Murray, R., Williams, S.C., Murphy, D.G.M., Catani, M., 2011. Atlasing location,
asymmetry and inter-subject variability of white matter tracts in the human brain with MR
diffusion tractography. Neuroimage 54, 49–59. doi:10.1016/j.neuroimage.2010.07.055
Tibshirani, R., 1996. Regression Shrinkage and Selection via the Lasso. J. R. Stat. Soc. Ser. B
58, 267–288.
Timmerhuis, T.P.J., Hageman, G., Oosterloo, S.J., Rozeboom, A.R., 1996. The prognostic value
of cortical magnetic stimulation in acute middle cerebral artery infarction compared to other
parameters. Clin. Neurol. Neurosurg. 98, 231–236. doi:10.1016/0303-8467(96)00034-0
Tuch, D.S., 2004. Q-ball imaging. Magn. Reson. Med. 52, 1358–72. doi:10.1002/mrm.20279
Tuch, D.S., Reese, T.G., Wiegell, M.R., Makris, N., Belliveau, J.W., 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
Van Kordelaar, J., Van Wegen, E., Kwakkel, G., 2014. Impact of Time on Quality of Motor
Control of the Paretic Upper Limb After Stroke. Arch. Phys. Med. Rehabil. 95, 338–44.
doi:10.1016/j.apmr.2013.10.006
Neurological Biomarkers of Post-Stroke Motor Recovery 162

van Kuijk, A.A., Pasman, J.W., Hendricks, H.T., Zwarts, M.J., Geurts, A.C.H., 2009. Predicting
hand motor recovery in severe stroke: the role of motor evoked potentials in relation to
early clinical assessment. Neurorehabil. Neural Repair 23, 45–51.
doi:10.1177/1545968308317578
Várkuti, B., Guan, C., Pan, Y., Phua, K.S., Ang, K.K., Kuah, C.W.K., Chua, K., Ang, B.T.,
Birbaumer, N., Sitaram, R., 2013. Resting state changes in functional connectivity correlate
with movement recovery for BCI and robot-assisted upper-extremity training after stroke.
Neurorehabil. Neural Repair 27, 53–62. doi:10.1177/1545968312445910
Walhovd, K.B., Fjell, A.M., Reinvang, I., Lundervold, A., Dale, A.M., Eilertsen, D.E., Quinn,
B.T., Salat, D., Makris, N., Fischl, B., 2005. Effects of age on volumes of cortex, white
matter and subcortical structures. Neurobiol. Aging 26, 1261–1270.
doi:10.1016/j.neurobiolaging.2005.05.020
Wan, C.Y., Zheng, X., Marchina, S., Norton, A., Schlaug, G., 2014. Intensive therapy induces
contralateral white matter changes in chronic stroke patients with Broca’s aphasia. Brain
Lang. 136C, 1–7. doi:10.1016/j.bandl.2014.03.011
Wang, D.-M., Li, J., Liu, J.-R., Hu, H.-Y., 2012. Diffusion tensor imaging predicts long-term
motor functional outcome in patients with acute supratentorial intracranial hemorrhage.
Cerebrovasc. Dis. 34, 199–205. doi:10.1159/000341857
Wedeen, V.J., Wang, R.P., Schmahmann, J.D., Benner, T., Tseng, W.Y.I., Dai, G., Pandya,
D.N., Hagmann, P., D’Arceuil, H., de Crespigny, a. J., 2008. Diffusion spectrum magnetic
resonance imaging (DSI) tractography of crossing fibers. Neuroimage 41, 1267–1277.
doi:10.1016/j.neuroimage.2008.03.036
Weimar, C., König, I.R., Kraywinkel, K., Ziegler, A., Diener, H.C., German Stroke Study
Neurological Biomarkers of Post-Stroke Motor Recovery 163

Collaboration, 2004. Age and National Institutes of Health Stroke Scale Score within 6
hours after onset are accurate predictors of outcome after cerebral ischemia: development
and external validation of prognostic models. Stroke. 35, 158–62.
doi:10.1161/01.STR.0000106761.94985.8B
Winstein, C., Lewthwaite, R., Blanton, S.R., Wolf, L.B., Wishart, L., 2014. Infusing motor
learning research into neurorehabilitation practice: a historical perspective with case
exemplar from the accelerated skill acquisition program. J. Neurol. Phys. Ther. 38, 190–
200. doi:10.1097/NPT.0000000000000046
Winstein, C.J., Merians, A.S., Sullivan, K.J., 1999. Motor learning after unilateral brain damage.
Neuropsychologia 37, 975–87. doi:10.1016/S0028-3932(98)00145-6
Winstein, C.J., Wolf, S.L., Dromerick, A.W., Lane, C.J., Nelsen, M.A., Lewthwaite, R., Cen,
S.Y., Azen, S.P., 2016. Effect of a Task-Oriented Rehabilitation Program on Upper
Extremity Recovery Following Motor Stroke. JAMA 315, 571.
doi:10.1001/jama.2016.0276
Winters, C., van Wegen, E.E.H., Daffertshofer, A., Kwakkel, G., 2014. Generalizability of the
Proportional Recovery Model for the Upper Extremity After an Ischemic Stroke.
Neurorehabil. Neural Repair. doi:10.1177/1545968314562115
Wöhrle, J.C., Behrens, S., Mielke, O., Hennerici, M.G., 2004. Early motor evoked potentials in
acute stroke: Adjunctive measure to MRI for assessment of prognosis in acute stroke within
6 hours. Cerebrovasc. Dis. 18, 130–134. doi:10.1159/000079265
Wolf, S.L., Catlin, P. a., Ellis, M., Archer, a. L., Morgan, B., Piacentino, A., 2001. Assessing
Wolf Motor Function Test as Outcome Measure for Research in Patients After Stroke.
Stroke 32, 1635–1639. doi:10.1161/01.STR.32.7.1635
Neurological Biomarkers of Post-Stroke Motor Recovery 164

Yoshioka, H., Horikoshi, T., Aoki, S., Hori, M., Ishigame, K., Uchida, M., Sugita, M., Araki, T.,
Kinouchi, H., 2008. Diffusion tensor tractography predicts motor functional outcome in
patients with spontaneous intracerebral hemorrhage. Neurosurgery 62, 97–103.
doi:10.1227/01.NEU.0000311066.03121.B8
Yu, C., Zhu, C., Zhang, Y., Chen, H., Qin, W., Wang, M., Li, K., 2009. A longitudinal diffusion
tensor imaging study on Wallerian degeneration of corticospinal tract after motor pathway
stroke. Neuroimage 47, 451–8. doi:10.1016/j.neuroimage.2009.04.066
Zarahn, E., Alon, L., Ryan, S.L., Lazar, R.M., Vry, M.-S., Weiller, C., Marshall, R.S., Krakauer,
J.W., 2011. Prediction of Motor Recovery Using Initial Impairment and fMRI 48 h
Poststroke. Cereb. Cortex 21, 2712–2721. doi:10.1093/cercor/bhr047
Zatorre, R.J., Fields, R.D., Johansen-Berg, H., 2012. Plasticity in gray and white: neuroimaging
changes in brain structure during learning. Nat. Neurosci. 15, 528–36. doi:10.1038/nn.3045
Zhang, S., Zhu, W., Zhang, Y., Yao, Y., Shi, J., Wang, C.-Y., Zhu, W., 2017. Diffusional
kurtosis imaging in evaluating the secondary change of corticospinal tract after unilateral
cerebral infarction. Am J Transl Res 9, 1426–1434.
Zhu, L.L., Lindenberg, R., Alexander, M.P., Schlaug, G., 2010. Lesion load of the corticospinal
tract predicts motor impairment in chronic stroke. Stroke. 41, 910–5.
doi:10.1161/STROKEAHA.109.577023

Abstract (if available)

Abstract Prediction of functional recovery after stroke is crucial to improving the efficacy of rehabilitation for stroke survivors. Previous studies reported that a Diffusion tensor imaging (DTI)-derived corticospinal tract (CST) biomarker is predictive of motor recovery early after stroke. However, little is known about whether or not a DTI-derived CST biomarker is an essential predictor of motor improvement in chronic stroke survivors. This dissertation aimed to examine whether a DTI-derived CST biomarker can be used to predict motor improvement in chronic stroke survivors with mild-to-moderate motor impairment. Further, we examined whether subcortical white matter structural changes over a 3-month period are measurable using research quality diffusion images. Finally, we sought to determine if there was a relationship between CST brain structural changes and improvement in motor behavior in this cohort of chronic stroke survivors. Our results showed that the CST biomarker can be used to predict motor improvement over a 3-month period. Further, a cross-validated multimodal model that included CST fractional anisotropy (FA) asymmetry, upper extremity Fugl-Meyer impairment score, and age more accurately predicted improvement in motor performance than a simple linear regression model. However, there was no detectable change in subcortical white matter structure of either the ipsilesional or contralesional CST. In spite of the lack of measurable subcortical white matter structural changes, there was a small but significant improvement in motor behavior as measured by the distal component of the Wolf Motor Function test (WMFT-Distal). These findings provide strong evidence that a DTI-derived CST biomarker is a significant predictor of improvement in motor behavior in chronic stroke survivors with mild-to-moderate motor impairment. Future research is necessary to develop gold standard methods to quantify both structural and functional brain changes for use in predicting functional recovery in stroke survivors.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Learning reaching skills in non-disabled and post-stroke individuals

PDF

The brain and behavior of motor learning: the what, how and where

PDF

The effects of fast walking, biofeedback, and cognitive impairment on post-stroke gait

PDF

Hemisphere-specific deficits in the control of bimanual movements after stroke

PDF

Computational model of stroke therapy and long term recovery

PDF

Experimental and computational explorations of different forms of plasticity in motor learning and stroke recovery

PDF

A multi-site neuroimaging approach to studying hippocampal damage in chronic stroke

PDF

Deficits and rehabilitation of upper-extremity multi-joint movements in individuals with chronic stroke

PDF

Relationship between interhemispheric inhibition and bimanual coordination in musicians

PDF

Planning of unconstrained reach actions after unilateral sensorimotor stroke

PDF

Using ecological momentary assessment to study the impact of social-cognitive factors on paretic hand use after stroke

PDF

Impact of enhanced efficacy expectation on motor learning in individuals with Parkinson’s disease

PDF

Development and implementation of a modular muscle-computer interface for personalized motor rehabilitation after stroke

PDF

Virtual surgeries as a tool for studying motor learning

PDF

Relationships between lifetime chronic stress exposure, vascular risk, cognition, and brain structure in HIV

PDF

The structure of kinematic variability during hopping: a window into motor control factors in overuse injury

PDF

Brain-based prediction of chronic pain progression: a longitudinal study of urologic chronic pelvic pain syndrome using baseline resting state connectivity from the periaqueductal gray

PDF

Locomotor skill learning in virtual reality in healthy adults and people with Parkinson disease

PDF

Corticomotor excitability of gluteus maximus: influence on hip extensor strength and hip mechanics

PDF

Trunk control during dynamic balance: effects of cognitive dual-task interference and a history of recurrent low back pain

Asset Metadata

Creator Kim, Bokkyu (author)

Core Title Relationship between brain structure and motor behavior in chronic stroke survivors

School School of Dentistry

Degree Doctor of Philosophy

Degree Program Biokinesiology

Publication Date 11/14/2017

Defense Date 10/09/2017

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag corticospinal tract,diffusion tensor imaging,motor recovery,neuroimaging,OAI-PMH Harvest,prognosis,stroke rehabilitation

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Winstein, Carolee J. (committee chair), Fisher, Beth E. (committee member), Gordon, James (committee member), Leahy, Richard M. (committee member), Schweighofer, Nicolas (committee member)

Creator Email bokkyuki@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-456974

Unique identifier UC11266922

Identifier etd-KimBokkyu-5916.pdf (filename),usctheses-c40-456974 (legacy record id)

Legacy Identifier etd-KimBokkyu-5916.pdf

Dmrecord 456974

Document Type Dissertation

Rights Kim, Bokkyu

Type texts

Source 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 a...

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