Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Learning reaching skills in non-disabled and post-stroke individuals
(USC Thesis Other)
Learning reaching skills in non-disabled and post-stroke individuals
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
LEARNING REACHING SKILLS
IN NON-DISABLED AND POST-STROKE INDIVIDUALS
by
Hyeshin Park
A dissertation submitted in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
(Biokinesiology)
in the University of Southern California
August 2015
Doctoral Committee:
Associate Professor Nicolas Schweighofer, Chair
Professor James Gordon
Professor Carolee Winstein
Professor Francisco Valero-Cuevas
Adjunct Professor Robert Gregor
© Hyeshin Park
2015
ii
DEDICATION
To my God and family
iii
ACKNOWLEDGMENTS
I would like to express my appreciation to a number of individuals. First and
foremost, I express my deepest gratitude to Dr. Nicolas Schweighofer for his great
guidance during my doctoral studies. I also acknowledge the helpful suggestions and
valuable support given by my doctoral committee members, Dr. James Gordon, Dr.
Carolee Winstein, Dr. Francisco Valero-Cuevas, and Dr. Robert Gregor.
Computational Neuro-Rehabilitation Laboratory (CNRL) is a fabulous laboratory
that encourages the intellectual growth. I thank all former and current CNRL members as
well as the Motor Behavior and Neurorehabilitation Laboratory (MBNL) members. I
extend warm thanks to my friends, Sooyeon, Boyea, Nattakarn, Okjoo, and Kayoung, as
well as my cello teacher, Mr. Chicco, and the Bella orchestra in Burbank. I greatly
appreciate all of the participants who contributed to my study. I have enjoyed and valued
your friendship. I thank Dr. Bradley for her readiness to help me complete my experiment.
My family has shown tremendous love, encouragement, and support over the years.
I especially remember grandfather, who died before I finished this work. My parents-in-
law always encouraged me and supported me. Finally, I would like to share this honor
with my husband and my Fitts.
iv
TABLE OF CONTENTS
DEDICATION .............................................................................................................. ii
ACKNOWLEDGMENTS .......................................................................................... iii
TABLE OF CONTENTS ............................................................................................ iv
LIST OF FIGURES .................................................................................................. viii
LIST OF TABLES ..................................................................................................... xii
ABSTRACT ............................................................................................................... xiii
CHAPTER 1: Introduction ...................................................................................... 1
1.1. Statement of the problem.................................................................................... 1
1.2. Current understanding of upper extremity stroke rehabilitation ........................ 3
1.3. Aims and hypotheses .......................................................................................... 5
1.4. Overview of dissertation structure...................................................................... 8
CHAPTER 2: Design and development of Arm Reach Training system ........... 12
2.1. Introduction ...................................................................................................... 12
2.2. Methods ............................................................................................................ 13
v
2.2.1. The Arm Reach Training system .............................................................. 13
2.2.2. Calibration ................................................................................................ 14
2.2.3. Target scheduling ...................................................................................... 17
2.2.4. Feedback ................................................................................................... 19
CHAPTER 3: How non-disabled individuals learn to move fast ........................ 22
3.1. Introduction ...................................................................................................... 22
3.2. Methods ............................................................................................................ 24
3.2.1. Participants ............................................................................................... 24
3.2.2. Study design .............................................................................................. 25
3.2.3. Data analysis ............................................................................................. 26
3.3. Results .............................................................................................................. 30
3.3.1. Learning effect on hand trajectories, kinematic outcomes, and
performance ............................................................................................. 31
3.3.2. Change in movement time ........................................................................ 32
3.3.3. Improved feedforward control .................................................................. 34
3.3.4. Decreased variability ................................................................................ 36
3.3.5. Increased initial elbow height and 3D kinematic analysis ........................ 38
3.4. Discussion ........................................................................................................ 40
Appendix A. ............................................................................................................ 44
CHAPTER 4: Short-duration and intensive training improves long-term……
reaching performance in individuals with chronic stroke ........ 52
4.1. Introduction ...................................................................................................... 52
4.2. Methods ............................................................................................................ 55
vi
4.2.1. Participants ............................................................................................... 55
4.2.2. Study design .............................................................................................. 57
4.2.3. Clinical assessments ................................................................................. 58
4.2.4. Data analysis ............................................................................................. 58
4.2.5. Mixed regression models .......................................................................... 59
4.3. Results .............................................................................................................. 60
4.3.1. Demographic information and overall training effect .............................. 60
4.3.2. Movement time at baseline ....................................................................... 62
4.3.3. Decrease in movement time with training ................................................ 63
4.3.4. Change in movement smoothness ............................................................ 65
4.3.5. Score change in box and block test ........................................................... 67
4.3.6. Relationship between initial performance and change in
performance ............................................................................................. 67
4.4. Discussion ........................................................................................................ 68
Appendix B. ............................................................................................................. 72
CHAPTER 5: Prediction of long-term gains due to intensive reach training……
in individuals with chronic stroke ................................................. 73
5.1. Introductions ..................................................................................................... 73
5.2. Methods ............................................................................................................ 76
5.2.1. Participants ............................................................................................... 76
5.2.2. Clinical assessments ................................................................................. 76
5.2.3. Study design .............................................................................................. 76
5.2.4. Data analysis ............................................................................................. 77
5.2.5. Nonlinear statistical model with mixed effect .......................................... 77
vii
5.3. Results .............................................................................................................. 79
5.3.1. Demographic information ......................................................................... 79
5.3.2. Nonlinear statistical model with mixed effect in movement time ............ 79
5.3.3. Relationship between individual model parameters and demographic
information parameters ............................................................................ 87
5.3.4. Prediction of long-term (1-month) performance ...................................... 88
5.3.5. Learning rate in the control and stroke groups ......................................... 92
5.4. Discussion ........................................................................................................ 93
Appendix C .............................................................................................................. 97
CHAPTER 6: Conclusion and future work ........................................................ 106
REFERENCES ......................................................................................................... 109
viii
LIST OF FIGURES
Figure 1.3.1. Goals of the four studies. ............................................................................... 6
Figure 2.2.1. The Arm Reach Training (ART) system ..................................................... 14
Figure 2.2.2. Calibration procedure for collocation. The center of magnetic position
sensor system represented cube is (0,0,0). After the translation, ⑧ is (0,0,0) and
called the home-position. .................................................................................................. 16
Figure 2.2.3. Diagram showing the location of the 35 test targets in the 2-dimensional
workspace. The five training targets are the black circles at 25 cm. ................................ 19
Figure 2.2.4. Procedure of a trial. ..................................................................................... 19
Figure 2.2.5. Five possible visuo-auditory feedback cues at the end of a training trial
based on comparison of
on line
MT
to the mean and standard deviation of the
movement time ( on line MT and std) computed in 20 previous trials (except for the first
block: see method). ........................................................................................................... 20
Figure 2.2.6. Elbow auditory alarm in real time during training session. ......................... 21
Figure 3.2.1. Diagram showing the timing of the five visits over a 5-week period. ........ 26
Figure 3.2.2. Angles ( ,, at shoulder and at elbow) defined by the posture of
the arm in 3-dimensional workspace. ............................................................................... 29
Figure 3.3.1. Examples of hand paths and tangential hand velocities before and after
training from a subject in Pre1-test and in 1-day and 1-month retention tests. First
row: Hand path. Second row: Tangential velocities and number of peaks (indicated
with filled symbols). Note that the subject shows a large decrease in movement time,
ix
increased peak velocity, and that the velocity profiles become symmetrical with a
single peak. ....................................................................................................................... 32
Figure 3.3.2. A: Mean movement time across the test sessions. B: Overall movement
time before, during, and after training shows a long-lasting (1 month) reduction of
MT following training. C: Regression coefficient of target distance (Test D) in each
test in the mixed regression model shows a long-lasting reduction of the effect of
distance on MT. D: Regression coefficient of target angle (Test cos(150-q)) in each
test in the mixed regression model shows a long-lasting reduction in the effect of
angle on MT. E: Mean movement time during training. The dashed vertical lines
separate the four visit days (indicated with filled symbols: circle for Pre-test, square
for Post-test, triangle for 1-day retention test, and diamond for 1-month retention
test). Error bars show standard errors of the mean of MT. * p < 0.007, ** p < 0.0001. .. 34
Figure 3.3.3. A: Mean symmetry ratio across the test sessions. B: Overall symmetry
ratio before, during, and after training shows a long-lasting (1 month) reduction of
symmetry ratio following training. C: Regression coefficient of target distance (Test x
D) in each test in the mixed regression model shows a long-lasting reduction of the
effect of distance on symmetry ratio. D: Regression coefficient of target angle (Test x
cos(150-q)) in each test in the mixed regression model shows a long-lasting reduction
in the effect of angle on symmetry ratio. E: Mean symmetry ratio during training. The
dashed vertical lines separate the four visit days (indicated with filled symbols: circle
for Pre-test, square for Post-test, triangle for 1-day retention test, and diamond for 1-
month retention test). Error bars show standard errors of the mean of symmetry ratio.
** p < 0.0001. ................................................................................................................... 36
Figure 3.3.4. First row: Mean maximum acceleration during 2-day training. Second
row: mean standard deviation of maximum acceleration. Third row: Coefficient of
variance (CV) on maximum acceleration. The dashed vertical lines separate the two
training days. ..................................................................................................................... 37
Figure 3.3.5. A: Change of initial elbow height during training. B: Change of final
elbow height during training. The dashed vertical lines separate the four visit days
(indicated with filled symbols: circle for Pre-test, square for Post-test, triangle for 1-
day retention test, and diamond for 1-month retention test). Error bars show standard
errors of the mean of MT. ................................................................................................. 38
Figure 3.3.6. Range of motion of shoulder and elbow angles in 3-dimensional
workspace. A: Range of motion of shoulder vertical angle during training. B: Range
of motion of shoulder horizontal angle during training. C: Range of motion of torsion
angle during training. D: Range of motion of elbow angle during training. The dashed
vertical lines separate the four visit days (indicated with filled symbols: circle for Pre-
test, square for Post-test, triangle for 1-day retention test, and diamond for 1-month
retention test). ................................................................................................................... 40
x
Figure 4.2.1. A: Diagram showing the timing of the five visits over a 6-week period
for the stroke group. .......................................................................................................... 57
Figure 4.3.1. Examples of hand paths and tangential hand velocities before and after
training for two subjects post-stroke (Subject 10 and Subject 5 from Table 1) in Pre1-
test and in 1-day and 1-month retention tests. First row: Hand path. Second row:
Tangential velocities and number of peaks (indicated with filled symbols: circle for
Pre1-test, diamond for 1-day retention test, and square for 1-month retention test).
Note that subject 10 on the left, with relatively high severity score (FM = 30/66),
shows a large decrease in movement time and number of peaks compared with
subject 5 on the right with a lesser severity score (FM = 51/66). ..................................... 61
Figure 4.3.2. A: Mean movement time in the stroke group across the test sessions. B:
Mean movement time in the control group. C: Overall movement time before, during,
and after training in the stroke group shows a long-lasting (1 month) reduction of MT
following training. D: Regression coefficient of target distance (Test D) in each test
in the mixed regression model shows a long-lasting reduction of the effect of distance
on MT in the stroke group. E: Regression coefficient of target angle (Test cos(150-
q)) in each test in the mixed regression model shows a long-lasting reduction in the
effect of angle on MT in the stroke group. * p < 0.007, ** p < 0.0001. ........................... 64
Figure 4.3.3. A: Mean number of peak in the stroke group across test sessions (results
for the control group are not shown because the mean number of peaks is close to 1
for all movements). B: Overall number of peaks before, during, and after training
shows a long-lasting (1 month) reduction in the number of peaks following training.
C: Regression coefficient of target distance (Test D) in each test in the mixed
regression model shows a long-lasting reduction of the effect of distance on number
of peak in the stroke group. D: Regression coefficient of target angle (Test cos(150-
q)) in each test in the mixed regression model shows a long-lasting reduction in the
effect of angle on number of peak in the stroke group. ** p < 0.0001. ............................ 66
Figure 4.3.4. Relationship between initial performance and change in performance
between Pre1-test and 1-day retention test in stroke group. A: Linear relationship is
significant between initial MT and ΔMT. B: Linear relationship is significant between
initial number of peaks and Δpeaks. ................................................................................. 68
Figure 5.3.1. Individual MT at Target 5 in the young control group (n = 9). Red,
green, cyan, blue, and purple represent individual trial-by-trial MT. Black solid lines
represent the model fitting based on each participant’s movement time; widely spaced
dash lines represent the single exponential fitting; narrowly spaced dash lines
represent the amount of fatigue using the equation (1). The dashed vertical lines
separate the four visit days (indicated with filled symbols: circle for Pre-test, square
for Post-test, triangle for 1-day retention test, and diamond for 1-month retention
test). ................................................................................................................................... 82
xi
Figure 5.3.2. Individual MT at Target 5 in the age-matched control group (n = 10).
Red, green, cyan, blue, and purple represent individual trial-by-trial MT. Black solid
lines represent the model fitting based on each participant’s movement time; widely
spaced dash lines represent the single exponential fitting; narrowly spaced dash lines
represent the amount of fatigue using the equation (1). The dashed vertical lines
separate the four visit days (indicated with filled symbols: circle for Pre-test, square
for Post-test, triangle for 1-day retention test, and diamond for 1-month retention
test). ................................................................................................................................... 84
Figure 5.3.3. Individual MT at Target 5 in the stroke group (n = 16). Red, green,
cyan, blue, and purple represent individual trial-by-trial MT. Black solid lines
represent the model fitting based on each participant’s movement time; widely spaced
dash lines represent the single exponential fitting; narrowly spaced dash lines
represent the amount of fatigue using the equation (1). The dashed vertical lines
separate the last four visit days among five visits (indicated with filled symbols: circle
for Pre-test, square for Post-test, triangle for 1-day retention test, and diamond for 1-
month retention test). ........................................................................................................ 86
Figure 5.3.4. Prediction of 1-month performance at Target 5 in the control group
(n=19). First row: (A) as a result of the equation (1) with fatigue term. Second row:
(B) as a results of equation (2) without fatigue term. Left panel: Relationship between
the ratio of “amplitude of the learning curve over initial performance” and the ratio of
“MT between the Pre1-test and the 1-month retention test over Pre1”. Right panel:
Relationship between the ratio of “Performance at the end of training over initial
performance” and the ratio of “MT between the Pre1-test and the 1-month retention
test over Pre1”. Red solid line represents least-squares fit to individual data. ................. 89
Figure 5.3.5. Prediction of 1-month performance at Target 5 in the stroke group
(n=16). First row: (A) as a result of the equation (1) with fatigue term. Second row:
(B) as a results of the equation (2) without fatigue term. Left panel: Relationship
between the ratio of “amplitude of the learning curve over initial performance” and
the ratio of “MT between the Pre1-test and the 1-month retention test over Pre1”.
Right panel: Relationship between the ratio of “Performance at the end of training
over initial performance” and the ratio of “MT between the Pre1-test and the 1-month
retention test over Pre1”. Red solid line represents least-squares fit to individual data. .. 90
Figure 5.3.6. Comparison of two examples (i.e., S1 and S15) with and without the
fatigue term effect at Target 5 in the stroke group. Left panel: Fitting results using the
equation (1). Right panel: Fitting results using the equation (2). Black solid lines
represent the model fitting based on each participant’s MT; widely spaced dash lines
represent the single exponential fitting; narrowly spaced dash lines represent the
amount of fatigue. ............................................................................................................. 92
xii
LIST OF TABLES
Table 3.2.1. Demographic information for the 19 right-handed participants in the
study. MMSE = Mini-mental state examination scores; SE = Standard error. ................. 25
Table 4.2.1. Demographic information for the 16 participants in the stroke group.
MMSE = Mini-mental state examination scores; FM = UE score of Fugl-Meyer
motor test; BBT = Box and Block Test; SE = Standard error. ......................................... 56
Table 5.3.1. Correlation between individual model parameters and Pre1 (MT before
training) in the control and stroke groups and the UE FM scores in the stroke group. *
p < 0.05, ** p < 0.01, *** p < 0.001. ................................................................................ 88
xiii
ABSTRACT
Many individuals with stroke who have impaired movement patterns benefit from
programs that provide intensive individualized rehabilitation under the supervision of a
therapist. Some individuals post-stroke, however, cannot receive the ideal number of
rehabilitation sessions due to therapy cost. Since most daily activities involve the upper
extremity, the recovery of reaching abilities is critical to maintain independent living and
quality of life.
This dissertation work focuses on the design and development of an arm reach
training (ART) system consisting of relatively inexpensive, reliable hardware and
software to be used in testing and training sessions. The ART system without any
assistance from a therapist provides intensive reach training with adaptive visuo-auditory
feedback to enhance the motor performance of an affected arm of individuals post-stroke.
Second, to better understand change of reaching performance through short-duration
intensive training with the ART system, this dissertation investigates how non-disabled
individuals learn to move quickly. We identify specific characteristics of motor skill
learning and the learning patterns by analyzing both spatial and temporal kinematics.
Third, this dissertation evaluates the long-term and generalization effects of short-
duration and intensive reach training in individuals with chronic stroke and mild to
xiv
moderate impairments and in age-matched non-disabled right-handed individuals. We
investigate change of clinical scores after training as well as the learning patterns, by
again analyzing both spatial and temporal kinematics in stroke.
Finally, this dissertation develops a nonlinear statistical model with mixed effect to
predict long-term performance change due to training in individuals with chronic stroke.
We test the model on three types of participants (i.e., stroke group, age-matched control
group, and young control group). We find that the model parameters can be indicators to
predict individualized long-term performance in the three groups and that the control
group learns more quickly than the stroke group.
We suggest that the results of this dissertation can inform future designs of stroke
upper extremity rehabilitation systems for clinical and in-home use. The results may also
inspire efforts to improve the rehabilitation of other neuromuscular disorders. Other
applications, such as sports rehabilitation and training may benefit from the results of this
dissertation.
1
CHAPTER 1: Introduction
1.1. Statement of the problem
Stroke leads to long-term motor disability and the disability result in behavioral
abnormalities. There has been about 795,000 people each year who suffer a stroke in the
United States (Dhamoon et al., 2009; Mozaffarian et al., 2015). In the United States, the
number of people who survive stroke is steadily increasing due to successful medical
technology (Lackland et al., 2014). Stroke disrupts the behavioral ability to carry out
basic activities of daily living, for example, reaching, grasping, walking, and jumping
(Carr & Shepherd, 2003). The motor ability deficits of the upper extremity in 65%
individuals with stroke become chronic after six months (Dobkin, 2005; Lum et al., 2009).
The affected upper extremity (UE) after stroke typically exhibit several kinematic
and kinetic characteristics of slowness of movements (Cirstea, Ptito, & Levin, 2003;
Coderre et al., 2010; Rohrer et al., 2002), abnormal movement posture (Cirstea,
Mitnitski, Feldman, & Levin, 2003; Trombly, 1992), lack of range of motion (Cirstea,
Mitnitski, et al., 2003; Dewald, Pope, Given, Buchanan, & Rymer, 1995; Sukal, Ellis, &
2
Dewald, 2007), abnormal timing between UE joints within a movement (Cirstea,
Mitnitski, et al., 2003; Dewald et al., 1995), and loss of inter-joint coordination (Beer,
Dewald, Dawson, & Rymer, 2004; Beer, Dewald, & Rymer, 2000; Cirstea & Levin,
2007; Cirstea & Levin, 2000; Cirstea, Mitnitski, et al., 2003; Dewald, Sheshadri,
Dawson, & Beer, 2001; Levin, 1996; Levin, Michaelsen, Cirstea, & Roby-Brami, 2002).
Therefore, the recovery of functional movements of stroke-affected arm is critical for
individuals with stroke to maintain their quality of life (i.e., better mobility and
independent living) (Carr & Shepherd, 2003).
Individuals with stroke can benefit from rehabilitation therapy with a therapist to
regain UE motor function. There is evidence that therapist-assisted training combined
with increasing the total hours of rehabilitation dedicated to the affected limb is
associated with positive effect on arm functional recovery (Oujamaa, Relave, Frooger,
Mottet, & Pelessier, 2009). Since often cost is an issue for individuals with stroke (i.e.,
treatment for stroke in the U.S. exceeded $65 billion in 2008) (Demaerschalk, Hwang, &
Leung, 2010; Heidenreich et al., 2011), some individuals with stroke cannot participate in,
or finish the ideal number of rehabilitation sessions due to the cost of therapy
(Demaerschalk et al., 2010). Therefore, there has been much interest in developing less
costly rehabilitation systems that can supplement conventional UE therapist-assisted
therapy.
3
1.2. Current understanding of upper extremity stroke rehabilitation
Arm reaching ability is considered the most important factor in individuals with
stroke, since most activities of daily living require reaching movements (Coster et al.,
2004; Page, Sisto, Levine, & McGrath, 2004). A simple reaching movement of
individuals post-stroke exhibits the characteristics of slowness of movements (Cirstea,
Ptito, et al., 2003; Kamper, McKenna-Cole, Kahn, & Reinkensmeyer, 2002; McCrea &
Eng, 2005), lack of range of motion (Sukal et al., 2007), impaired movement smoothness
(e.g., more than one peak on the velocity profile) (Cirstea, Ptito, et al., 2003; Kamper et
al., 2002; McCrea & Eng, 2005), indirectness of movements (e.g. increased trajectory
curvature) (Kamper et al., 2002; McCrea & Eng, 2005), disrupted inter-joint coordination
(Beer et al., 2004; Dewald et al., 2001; Levin, 1996), abnormal timing between UE joints
(Beer et al., 2004; Dewald et al., 2001; Levin, 1996), and/or compensatory movement
(e.g., excessive trunk movement during reaching) (Levin et al., 2002; Michaelsen &
Levin, 2004). Moreover, the kinematic outcomes of individuals post-stroke significantly
correlate with their motor impairment level (Kamper et al., 2002).
The positive effects of reach training movements of the affected arm with
rehabilitation systems have been demonstrated by several methods. For example, MIT-
MANUS, a 2 degree-of-freedom (DOF) planar robot and video program with visual and
auditory feedback, provides goal-directed reach training having passive and active
movements of the shoulder and elbow joints (e.g., flexion-extension and abduction-
adduction movements) (Aisen, Krebs, Hogan, McDowell, & Volpe, 1997; Krebs, Hogan,
Aisen, & Volpe, 1998; Kwakkel, Kollen, & Krebs, 2007). ARM-Guide is designed to
4
assist or resist arm movement following linearly different directions for patients post-
stroke who are unable to actively move the affected arm (D. J. Reinkensmeyer, Emken, &
Cramer, 2004; D. J. Reinkensmeyer et al., 2000). MIME, a 6-DOF robot, provides
unassisted shoulder-elbow movements to only the affected arm or to both arms through
the mirror image movement (Lum et al., 2006). Bi-Manu-Track engages passive and
active distal arm movements (e.g., elbow pronation-supination and wrist flexion-
extension) (Hesse et al., 2005). ARMin, a 6-DOF semi-exoskeleton robot, provides task-
oriented repetitive and intensive arm training including active and passive movement
modalities (Nef & Riener, 2005). The rehabilitation systems are likely to have little
potential for use in clinical and/or home-based rehabilitation settings, due to the cost and
operating complexity of the instrumentation required.
Active participation of the affected arm post-stroke has been associated more with
better performance and cortical reorganization of surrounding the damaged brain area
than passive movements (Carel et al., 2000; Conner, Culberson, Packowski, Chiba, &
Tuszynski, 2003; Jeffery & Good, 1995; Karni, 1995; Lotze, Braun, Birbaumer, Anders,
& Cohen, 2003; Muellbacher, Ziemann, Boroojerdi, Cohen, & Hallett, 2001; Nudo,
Wise, SiFuentes, & Milliken, 1996). Indeed, intensive single or multiple session training
can result in faster, more precise, and smoother movements (Aisen et al., 1997;
Blennerhassett & Dite, 2004; Cirstea, Mitnitski, et al., 2003; DeJong, Schaefer, & Lang,
2012; Dipietro et al., 2012; Harris-Love, Morton, Perez, & Cohen, 2011; Hesse et al.,
2005; Lo et al., 2010; Lum et al., 2006; Michaelsen & Levin, 2004; Rohrer et al., 2002;
Volpe et al., 1999). The faster movement duration is associated with better
shoulder/elbow movement timing during reach practice (D. J. Reinkensmeyer, Cole,
5
Kahn, & Kamper, 2002; Roby ‐Brami et al., 2003) and improved smoothness in
movements is characteristics of stroke recovery (Rohrer et al., 2002). Therefore, intensive
training of the affected arm post-stroke can improve motor control and increased
participation of the affected arm, and may promote cortical reorganization of the area
surrounding the damaged brain (Nudo et al., 1996; Taub, Uswatte, & Elbert, 2002).
1.3. Aims and hypotheses
There are four main goals of this dissertation: 1) to design and develop the arm
reach training (ART) system for rehabilitation of UE functions post-stroke; 2) to
investigate how non-disabled individuals learn to move quickly; 3) to study how short-
duration and intensive training can improve long-term reaching performance in
individuals with chronic stroke; and 4) to explore whether improvement in performance
during intensive arm reach training can predict long-term gains in individuals post-stroke.
Figure 1.3.1 illustrates the four goals and the next paragraph gives the details.
F
A
re
T
se
tr
tr
on
au
igure 1.3.1.
Aim 1: To
ehabilitatio
The ART sys
essions: to
raining (testi
raining (train
n a particip
uditory alarm
Goals of the
design an
n of UE fun
stem is desig
compare th
ing session)
ning session)
ant’s previo
ms to minim
e four studie
nd develop
nctions post-
gned to 1) pr
he change o
and monitor
); 3) provide
ous performa
mize compens
6
s.
p the arm
-stroke
rovide unass
of kinematic
r the change
e real-time v
ance in a tra
satory elbow
m reach tra
sisted intens
c/kinetic pe
of kinemati
visuo-auditor
aining sessio
w movement
aining (AR
sive reach tra
erformance b
ic/kinetic per
ry feedback
on; and 4) p
.
RT) system
aining; 2) be
before and
rformance d
adaptively b
provide real
m for
e two
after
during
based
l-time
7
Aim 2: To investigate how non-disabled individuals learn to move fast
In non-disabled right-handed individuals, short-duration and intensive arm reach training
induces the following:
H2a: increase in speed
H2b: long-lasting improvement in performance
H2c: generalization to untrained movements
H2d: increased reliance on feedforward control
H2e: decrease in variability
H2f: improvement in movement efficiency
Aim 3: To study the effect of short-duration and intensive arm reach training in
individuals with chronic stroke
In individuals with chronic stroke, short-duration and intensive arm reach training
induces the following:
H3a: long-lasting improvement in movement time and movement smoothness
H3b: If H3a is true, the decreased in movement time and improved smoothness are
affected by the effect of distance and large inertia
H3c: durable generalization effect
H3d: benefits in clinical assessment
8
Aim 4: To develop and assess a novel data-driven statistical model and explore the
model components influencing the prediction of long-term gains in individuals with
chronic stroke
A novel nonlinear statistical model with mixed effect, developed as a result of short-
duration and intensive reach training in non-disabled individuals and individuals with
chronic stroke, finds the following:
H4a) amplitude of learning curve in performance due to training predicts long-term
performance
H4b) performance at the end of training does not predict long-term performance
H4c) although total numbers of trials remain constant for all participants, learning is
faster in the control group than in the stroke group
1.4. Overview of dissertation structure
Chapter 2, Design of Arm Reach Training (ART) system, describes the design and
development of Arm Reach Training system which is the computational foundation for
the remaining chapters. Motivated by the aforementioned limitations in the existing
assistive systems such as high cost and systematic operating complexity, this chapter
presents the design and development of the Arm Reach Training (ART) system that
engages individuals with stroke actively, intensively, and adaptively to improve reaching
ability in real time. The ART system consists of inexpensive, reliable hardware and
software combined with testing and training sessions.
9
Chapter 3, How non-disabled individuals learn to move fast, is based on a journal
paper co-authored by James Gordon and Nicolas Schweighofer, which investigated the
characteristics of motor skill learning. This chapter tests whether two sessions of
intensive reach training lead to motor skill learning in non-disabled individuals. If true,
the chapter then investigates the characteristics of the motor skill learning. Nine (24.4 ±
1.1 years, referred to as young group) and ten (56.5 ± 2.9 years, referred to as elder
group) non-disabled right-handed participants are instructed to reach to circular targets
with the index finger of the dominant hand as quickly as possible. Each undergoes two
sessions of intensive reach training, with 600 training movements per session. The results
show significant and long-lasting (1-month) improvements in movement time (36% and
39% on average in the young and elder groups, respectively) and symmetry ratio (19% on
average in both groups). The improvement in symmetry ratio indicates that velocity
profiles are initially asymmetric (right-skewed) but become symmetric due to improved
feedforward control. Moreover, the coefficient of variation of the peak acceleration
decreases with training and the training induced generalization to non-trained targets,
which persists in 1-day and 1-month retention tests. The results also show that training in
the young group only benefits efficiency of movement by increasing the initial elbow
height (i.e., increased initial elbow height reduces the range of motion in shoulder
movement and increases the range of motion in elbow movements during reaching). This
chapter interprets the results to mean that preparing the posture with increased initial
elbow height reduces the inertia of arm or energy cost to move fast.
10
Chapter 4, Short-duration and intensive training improves long-term reaching
performance in individuals with chronic stroke, is based on a journal paper, co-authored
by Sujin Kim, Carolee Winstein, James Gordon, and Nicolas Schweighofer, which
compared the reach performance before and after short-duration and intensive reach
training in the stroke group to the baseline performance in the age-matched control group.
While long-term training regimens are often not applicable in actual clinical settings, this
chapter tests whether two sessions of intensive reach training lead to long-term and
generalization effects in 16 individuals with chronic stroke and mild to moderate
impairments. Each undergoes two sessions of intensive reach training, with 600 training
movements per session. The results show significant and long-lasting (1-month)
improvements in movement time (20.4% on average) and movement smoothness (22.7%
on average) in the stroke group. The greatest improvements occur in post-stroke
individuals with the greatest initial motor impairments. In addition, training induces
generalization to non-trained targets, which persists in 1-day and 1-month retention tests.
There is a significant improvement in the Box and Block test from baseline to 1-month
retention test (23.1% on average). Thus, short-duration and intensive reach training can
lead to generalized and durable benefits in individuals with chronic stroke and mild to
moderate impairments.
Chapter 5, Predicting the long-term gains due to intensive reach training in
individuals post-stroke, is based on a journal paper, co-authored by Nicolas Schweighofer,
which developed a data-driven nonlinear statistical model with mixed effect to predict
long-term performance in individuals with chronic stroke. The model, consisting of
11
exponential decay, fatigue, and intercept was fit based on individual MTs in 19 non-
disabled individuals and 16 individuals with chronic stroke. As a result, all individual
fitted curves from the model significantly correlate with individual MTs during the first
day training (p < 0. 00001). The model parameter demonstrates that the control group
learns more quickly than the stroke group. After describing the model components
influencing the prediction of long-term gains in individuals with chronic stroke, this
chapter suggests that amplitude of learning curve in performance significantly predicts
long-term performance (
2
R = 0.31, p = 0.026), whereas the performance at the end of
training does not predict long-term performance (
2
R = 0.157, p = 0.129). Thus, the
prediction from the model can provide an efficiently individualized training schedule for
stroke UE rehabilitation.
Chapter 6, Conclusion and future work, summarizes the major findings and gives
recommendations for additional studies.
12
CHAPTER 2: Design and development of Arm Reach Training system
2.1. Introduction
In order to regain and facilitate their functional mobility during reaching, grasping,
and walking, therapy programs with intensive and individualized rehabilitation under the
supervision of a therapist are beneficial to individuals with stroke. However, many
individuals with stroke cannot afford to receive the ideal number of rehabilitation
sessions due to the expensive care cost from a therapist. To meet the great needs of
receiving cost-effective rehabilitation training, various rehabilitation systems have
investigated to improve UE motor functions in individuals with stroke (e.g., MIT-
MANUS, ARM Guide, MIME, InMotion, Shoulder-Elbow Robot, BiManu-Track, and
etc.). However the existing assistive systems are impractical for use in clinical or home
rehabilitation training regimens due to their cost and operating complexity of the systems.
Motivated by the high costs of post-stroke therapeutic assistance, this chapter
describes the design and development of a user-friendly system that helps individuals
actively, intensively, and adaptively to improve their reaching abilities in real time. The
13
Arm Reach Training (ART) system consists of inexpensive, reliable hardware and
software combined with testing and training sessions. The ART system can be used in
clinical and in-home settings.
2.2. Methods
2.2.1. The Arm Reach Training system
The Arm Reach Training (ART) system has a 2-dimensional wooden table, an
overhead projector, a magnetic position sensor system with two sensors, and a computer
(see Figure 2.2.1). Custom software implemented with Linux OS captures the position
signals of the two magnetic sensors, displays a visual target on the wooden table using
the projector, and stores the position signals for data analysis. Before proceeding to the
experiment with the ART system, a participant removes all metal objects to avoid
possibility of magnetic interference with the magnetic sensor system, and then sits down
in a chair equipped with a seat belt to restrain trunk movements. In addition, the
participant adjusts the position of the chair in order to reach to the farthest target naturally
and comfortably in the ART workspace. Experiments begin with a calibration procedure
that measures the positions of the elbow and index finger in the 3-dimensional
workspace.
F
C
th
p
S
2
co
igure 2.2.1.
Magneti
Corporation)
he stroke g
erformance,
ensor data a
.2.2. Calibr
For co-
oordinates a
The Arm Re
c sensors (
are attached
group and t
and to the l
are recorded
ation
-location in
are collocat
each Trainin
(six degree-
d to the parti
the domina
lateral epico
at 120 Hz w
n a 3-dime
ted with th
14
ng (ART) sys
-of-freedom
icipant’s ind
ant hand in
ondyle of the
with accuracy
ntional virt
he relative l
stem
miniBird 5
dex fingertip
n the contro
e humerus to
y better than
tual reality
location, or
500, Ascens
of the more
ol group to
o monitor el
n 1 mm.
environme
rientation, a
sion Techno
e affected ha
o monitor r
lbow movem
ent, the gr
and scale o
ology
and in
reach
ments.
aphic
f the
15
magnetic sensor coordinates. Then the locations of the two magnetic sensors are
translated by using a homogeneous matrix with rotation, translation, and scaling. Each
sensor needed the following procedures for the calibration. Figure 2.2.2 illustrates the
procedure for calibrating the two magnetic sensors’ co-location, shows the one of nine
standard points displayed in the ART table in sequence (e.g., ① to ⑨). Then each
magnetic sensor is put on the illuminated standard points until hearing beep sound (i.e.,
the beep sounds when an experimenter pushes the enter button in the keyboard). The
homogeneous 4 by 4 matrix is calculated by saving the coordinates of the real points and
the standard points collocated with respect to the participant’s view. The new sensor
positions are registered based on the homogeneous matrix acquired from the calibration
mode.
F
se
th
T
(R
m
igure 2.2.2.
ensor system
he home-pos
arg
arg
tet
tet
x
y
In equati
These variabl
R11, R12, R21
matrix by the
Calibration
m represente
sition.
11 12
21 22
RR
RR
ion (1), 6 un
les were com
, R22, d1, and
e selection o
n procedure
d cube is (0
1
2
sensor
sensor
x d
yd
nknown varia
mputed using
d d2). Determ
of 3 points o
16
e for colloca
0,0,0). After
1
2
ables (i.e., R
g 13 equation
mine the tran
on a triangle
ation. The
the translat
(
R11, R12, R21,
ns by taking
nslation and
13 times ite
center of m
tion, ⑧ is (
(1)
, R22, d1, and
g the average
rotation of t
erations (Fig
magnetic pos
(0,0,0) and c
d d2) were ex
e of each var
the homogen
gure 2.2.2) t
sition
called
xisted.
riable
neous
to get
17
the combinations (1,2,3), (1,3,4), (1,2,5), (2,3,5), (3,4,5), (1,4,5), (6,7,8), (7,8,9), (9,6,7),
(5,6,7), (5,7,8), (5,8,9), and (5,6,9). Register the new sensor positions based on the
homogeneous matrix acquired from the calibration mode, as:
11 12
21 22
1
2
sensor
sensor
x RR x d
yR R y d
(2)
For the z axis, the z-values measured during calibration (i.e., above the ART table)
are zeros. Before experiment starts, all participants’ shoulder position based on the home-
position, the coordinates of the origin (0, 0, 0), were measured.
2.2.3. Target scheduling
As mentioned, participants are instructed to reach a target with the index finger of
their most affected arm in the stroke group and dominant arm in the control group as
quickly as possible. The diameter of each target is 3 cm. For each test trial, the projector
displays one of 35 targets in a pseudorandom order – distributed in 20-degree increments
between 30° and 150° (0 degree is on the right x-axis, approximately parallel to the
trunk) along 7 circular arcs centered on the home position, with a distance of 5cm
between arcs, from 10 cm to 30 cm away from the home position as shown in Figure
2.2.3.
tr
w
or
in
ra
tr
on
p
b
Trainin
rials per bloc
was comprise
rder (see Fig
ncluded in th
anging from
rials remaine
n the perform
Testing:
seudo-rando
elow). The t
g: Each train
ck. Blocks w
ed of 20 tria
gure 2.2.3 an
he training
m 50° to 130
ed constant f
mance and r
: Each test c
om order, an
target numbe
ning session
were separat
als to each o
nd below). In
set -- locate
0° (see Figu
for each part
esting period
consisted of
nd then 35 tri
ers for these
18
ns consists o
ted by rest p
of the 5 train
n other word
ed 25 cm fro
ure 2.2.3 and
ticipant, but
ds (participa
60 trials: 5
ials, to one t
35 trials we
f 600 trials,
periods of at
ning targets,
ds, a projecto
om the hom
d below). N
the duration
ants could re
trials to eac
to each of th
ere selected p
given in six
t least 5-min
presented in
or displays o
me resting po
Note that the
n of each ses
equest longer
ch of the 5 t
he 35 targets
pseudo-rand
x blocks with
nutes. Each b
n pseudo-ran
one of five ta
osition on a
e total numb
sion varied b
r rest periods
training targe
s (Figure 1B
domly.
h 100
block
ndom
argets
an arc
ber of
based
s).
ets in
B, and
F
w
h
R
ta
al
in
ap
lo
F
2
ch
p
igure 2.2.3.
workspace. T
Figure 2
is/her index
Ready sound
arget as quic
llowed is 5
ndex finger
ppears at on
ocations in a
igure 2.2.4.
.2.4. Feedba
Feedbac
hallenging.
ossible feedb
Diagram sh
The five train
2.2.4 illustra
x finger on t
s. After hear
ckly as possib
seconds. Af
to the home
ne of the othe
a pseudo-rand
Procedure o
ack
k as extrin
At the end
back signals
howing the
ning targets a
ates the pro
the home po
ring the Go
ble and tries
fter 500 ms,
e position. A
er four locat
dom order.
of a trial.
nsic inform
of each tra
s based on th
19
location of
are the black
ocedure of a
osition, a ta
sound, the p
s to stay on t
the target d
After 1 seco
tions during
mation helps
aining trial,
he participan
the 35 test
k circles at 2
a reach trial
arget appear
participant m
the target. Th
disappears an
nd, the next
training, or
s to keep
the ART s
nt's on-line m
targets in th
5 cm.
l. After a p
rs on the tab
moves the in
he maximum
nd the partic
t target from
at one of th
rehabilitatio
system prov
movement tim
he 2-dimens
participant p
ble and an a
ndex finger t
m movement
cipant return
m the trainin
e 34 non-tra
on and tra
ided one of
me, referred
sional
places
audio
to the
t time
ns the
ng set
aining
aining
f five
d to as
M
in
se
(t
fo
co
co
lo
fe
b
b
g
1
p
ta
F
b
on line
MT
, com
nterval from
ensor on the
time when th
or 500 ms).
omparing M
omputed fro
ower values
eedback, an
lock, in trial
ecome avail
ood” sound)
0, no feedba
articipant ho
arget location
igure 2.2.5.
ased on com
mpared to pr
m the movem
e index fing
he fingertip e
. At each tr
on line
MT
to the
om the 20 tr
s of on l MT
d higher va
ls 41 to 100,
able. In trial
) was provid
ack was give
ow fast to m
ns, and to m
Five possib
mparison of
revious mov
ment start (tim
ertip first cr
entered the t
rial, the vis
e mean ( MT
rials to the s
(0.5 line std
alues of MT
on lin e MT an
ls 11 to 40, a
ded when MT
en. The obje
move in ord
maintain the p
ble visuo-au
on line
MT
to
20
vement times
me at which
rossed the 3
target, with t
suo-auditory
on line MT ) and
same target
) d compare
(0.5 on line T
nd std were c
a single type
on line
T
was sm
ectives of th
der to receiv
participant’s
uditory feedb
the mean a
s.
on line
MT
w
h the tangen
30 cm/s thre
the condition
y feedback c
the standard
in the previ
ed to the m
) std “nega
computed wi
e of feedback
maller than in
he visuo-aud
ve positive f
motivation.
back cues a
and standard
was compute
ntial velocity
eshold) to th
n that the fin
cues was se
d deviation
ious block (
mean gener
ative” feedb
ith 20 data p
k (i.e., “gree
n previous tr
ditory feedba
feedback acr
.
at the end o
d deviation o
ed using the
y of the mag
he movemen
nger remains
elected base
(std) of MT
(Figure 1C),
rating “posi
back. In the
points, as the
en” cue and “
rials. In trial
ack are to te
ross the diff
of a training
of the move
time
gnetic
nt end
s on it
ed on
on lin e T
with
itive”
e first
e data
“very
s 1 to
ell the
ferent
g trial
ement
ti
m
un
ep
ac
ep
w
F
ime ( on l MT
method).
To minim
npleasant “b
picondyle m
cromion (Fi
picondyle di
who participa
igure 2.2.6.
line and std)
mize the par
beep” sound
moves above
igure 2.2.6)
id not excee
ated in this st
Elbow audit
computed i
rticipant’s c
ds when the
a point two
). There is
ed the two th
tudy.
tory alarm in
21
in 20 previo
ompensatory
magnetic se
o-thirds of th
evidence th
hirds of the h
n real time d
ous trials (ex
y elbow mo
ensor attache
he height bet
hat the mag
height in all
during trainin
xcept for the
ovements du
ed to the pa
tween the A
gnetic senso
l 19 non-dis
ng session.
e first block
uring reachin
articipant’s la
ART table an
or on the la
abled indivi
k: see
ng, an
ateral
nd the
ateral
iduals
22
CHAPTER 3: How non-disabled individuals learn to move fast
3.1. Introduction
Unlike performance that temporarily improves a skill through practice, learning is
associated with a long-lasting performance change and generalization in different
conditions (Cahill, McGaugh, & Weinberger, 2001; C. J. Winstein et al., 1996). Recent
studies, however, have emphasized that motor adaptation is a type of motor learning in
which motor commands are altered to compensate for external disturbances from the
environment. The motor adaptation literatures have examined the arm reaching
movements of non-disabled individuals in experimental settings such as prism adaptation
(Martin, Keating, Goodkin, Bastian, & Thach, 1996; Wolpert, Diedrichsen, & Flanagan,
2011), visuomotor adaptation (Shadmehr & Mussa-Ivaldi, 1994; Thoroughman &
Shadmehr, 2000), and dynamic force-field adaptation (Smith, Ghazizadeh, & Shadmehr,
2006). In these settings with artificial perturbation, participants learned to execute the
movement leading error reduction by recalibration of their movements. After training
23
with the perturbation, however, the improved performance did not translate into skills in
the setting with no perturbation.
Intensive task-specific practice has been identified as an important factor for
improving sports performance beyond present capabilities in the functional task (Keetch,
Schmidt, Lee, & Young, 2005) and rehabilitation (Birkenmeier, Prager, & Lang, 2010;
Blennerhassett & Dite, 2004; Ellis, Sukal-Moulton, & Dewald, 2009; Lo et al., 2010; D.
J. Reinkensmeyer et al., 2004; Rohrer et al., 2002; Thielman, Dean, & Gentile, 2004;
Wolf et al., 2006). This study focuses on motor skill learning leading durable improved
performance and generalization in response to practice without any artificial perturbation.
A speed-accuracy trade-off function (SAF) (Krakauer & Mazzoni, 2011; Wulf &
Lewthwaite, 2010) is a feature of motor skill learning as is a reduction in movement
variability. Shmuelof et al. trained fifty non-disabled subjects to move a cursor from one
circle to another circle inside a circular channel by only using wrist movements. After
three consecutive training days, the skilled wrist movement induced faster speed and
better accuracy of the cursor trajectories and reduced trajectory variability (Shmuelof,
Krakauer, & Mazzoni, 2012). Hung et al. have proposed that the change of joint
coordination pattern leads to motor skill learning in a complex task involving multiple
joints (Hung, Kaminski, Fineman, Monroe, & Gentile, 2008). In a study of six non-
disabled individuals who practiced a multi-joint throwing task with 100 trials a day, three
times per week for 13 sessions, the authors found improved accuracy and variability of
both end-point path and joint coordination (Hung et al., 2008).
This chapter proposes that short-duration and intensive reach training promotes
long-lasting and generalization effects in reaching performance and induces motor skill
24
learning with several characteristics. This chapter discusses the following characteristics:
1) increase in speed; 2) long-lasting improvement in performance; 3) generalization to
untrained movements; 4) increased reliance on feedforward control; 5) decrease in
variability; and 6) improvement in movement efficiency. The next section explains the
methodology.
3.2.Methods
3.2.1.Participants
Nine non-disabled participants (24.4 ± 1.1 years; 4 males, 5 females), subsequently
referred to as the young group, and ten non-disabled participants (56.5 ± 2.9 years; 5
males, 5 females), referred to as the elder group (i.e., as compared with the young group),
were recruited. Exclusion criteria included any central neurological dysfunction. Only
right-handed people were included according to the Edinburgh Handedness Inventory
(Oldfield, 1971). Table 3.2.1 summarizes the demographic information.
This study was approved by the University of Southern California Institutional Review
Board, and all participants naïve to the purpose of the experiments read and signed an
informed consent prior to study enrollment.
T
M
3
re
w
ac
re
w
sc
Table 3.2.1. D
MMSE = Min
.2.2. Study d
This proo
epeated-mea
week period.
ctual session
etention test
week follow
chedule.
Demographi
ni-mental sta
design
of-of-concep
asures design
First, the p
ns consisted
session. Th
wing training
ic informatio
ate examinat
t study des
n. All partici
participants
of three con
he participan
g for the o
25
on for the 1
tion scores;
scribed in
ipants visite
familiarized
nsecutive da
nts returned
one-month r
9 right-hand
SE = Standa
this chapte
ed the labora
d themselves
ays, i.e., two
to the labor
retention te
ded participa
ard error.
er was a w
atory for fou
s with the A
o training da
ratory one ti
st. Figure
ants in the s
within-partici
ur visits over
ART system
ays and a one
ime in the f
3.2.1 shows
study.
ipant,
r a 5-
. The
e-day
fourth
s the
F
3
u
d
v
w
in
P
pr
igure 3.2.1.
.2.3. Data a
Speed
The sen
sing a zero p
ata analysis
elocity of th
which the ta
nside of the
earson), vali
Symmetr
To meas
rofile was es
Diagram sho
nalysis
sor position
phase-lag se
was compu
he index fin
ngential vel
target.
on
MT
idating the o
ry ratio
sure the rel
stimated as:
owing the tim
n data were
econd-order
uted by the
gertip first e
locity fell b
n line
and MT
on-line feedb
iance of fee
26
ming of the
low-pass fi
Butterworth
interval be
exceeded 5%
below 5% o
T were highl
back method
edforward c
five visits ov
iltered with
h filter. Mov
etween the t
% of maxim
of maximum
ly correlated
d (see Chapte
control, the
ver a 5-week
a cutoff fre
vement time
time at whic
mum velocity
m velocity w
d (r = 0.987
er 2.2.4).
symmetry
k period.
equency of
(MT) for o
ch the tange
y and the tim
with the fing
and p < 0.0
ratio of vel
7 Hz
ffline
ential
me at
gertip
0001,
locity
27
Time at maximum velocity
symmetry ratio
Total movement time
(2)
If a symmetry ratio was less than 0.5, the shape of the velocity profile was right-skewed.
If a ratio was greater than 0.5, the shape of the velocity profile was left-skewed. If a ratio
is exactly 0.5, the velocity profile was symmetrical.
Variability
To measure the movement variability, the coefficients of variation (CV) of
maximum acceleration were calculated as:
std
CV
mean
(3)
where std was standard deviation. A moving average with 20 bins was used to analyze
the data for each training target.
Three-dimensional kinematic analyses
Reaching task with ART system was performed in 3-dimensional workspace. It was
assumed that shoulder position did not change because of the participant’s trunk was
restrained by the chair’s seat belt. Figure 3.3.2 illustrates the coordinate system used to
define the angles of the arm posture: the coordinates of the origin (0, 0, 0) were at each
participant’s fixed shoulder position. Based on all right-handed participants, the arrows in
the X, Y, Z axis in Figure 3.3.2 represent lateral-, forward-, and up- direction,
respectively.
an
W
h
d
th
p
fl
The ang
nd represent
When the for
orizontal an
irection, the
he humeral
lane, the
lexion or ext
gle , referr
ted the arm
rearm was on
ngle, was ca
was zero
axis of the
was zero. T
tension. Whe
ed to as a s
’s elevation
n Z-axis, the
alculated in X
o. The angle
upper arm.
The angle
en the elbow
28
houlder vert
(Soechting,
e was zero
X-Y plane.
, referred
When the u
, referred to
w was fully e
tical angle,
, Buneo, He
o. The angle
When the f
d to as a torsi
upper arm a
o as an elbo
extended, the
was calcula
errmann, &
e , referred
forearm was
ion, represen
and forearm
ow angle, re
e was zero
ated in Y-Z p
Flanders, 1
d to as a sho
s on X-axis
nted a rotatio
lay in a ve
epresented e
o.
plane
995).
oulder
right
on on
ertical
elbow
29
Figure 3.2.2. Angles ( ,, at shoulder and at elbow) defined by the posture of the
arm in 3-dimensional workspace.
22 2 2 2
cos
tan
cos
2
cos '
e
upper
e
e
hh h upper forearm
upper forearm
final final
z
l
y
x
xy z l l
ll
eh up
2
(' )
() ( )
( , , ) (0,0,0)
;
()
unit unit
xx x x
yy y y
zz zz
xy z
xx x
yy y
zz z
xy
where eh eh Sh eh Sh
he h e
se se
he h e
norm se norm se
he he
where s s s s
es e
se e s e
es e
norm se e e
22
22 2
22 2
()
(' )
00
0(0 )
() ()
11
()
z
final
xy z
unit unit
final
xy z
e
eh eh
eh
norm eh
eh eh eh
up up Sh up Sh
se se
norm se norm se
up up
up
norm up
up up up
where indicated inner product.
30
Statistical Analysis
To identify the temporal and spatial characteristics performance and change in
performance over time, a linear mixed model analysis (with SPSS 18) with MT as the
dependent variable and subjects’ random intercept was conducted. To test for short- and
long-term training effects, Test (pre-, post-, 1-day and 1-month follow-up session) was
included as a fixed repeated factor. Target distance (D) and cosine of the target angle
(cos(150-q), where q was the target angle ranging from 30 ° to 150 °) were included as co-
variates. To test for possible training-induced generalization effect of target distance and
target angles, interaction terms Test D and Test cos(150-q) were included. Model
comparisons (using the Bayesian information criterion, BIC, as selection factor) showed
that repeated measures were best modeled with an auto-regressive moving average
(ARMA(1,1)). Bonferroni correction was used for multiple pairwise comparisons. The
level of statistical significance was set at p < 0.05. All average results were reported with
mean ± standard errors (SE) of the corresponding mean.
3.3. Results
All participants completed the 600 trials of the two-day training session and
successfully reached all targets within 5 seconds. The young and elder groups
significantly differed in age (t-test, p < 0.0001). Since some participants took longer time
than others because of different reaching performance level and/or longer breaks,
performing the 600 movements in each Training session lasted on average 77.2 ± 2.9
31
minutes (range: 61.9 to 110.6 minutes) in the young group, and 90.3 ± 3.8 minutes
(range: 64.7 to 117.7 minutes) in the elder group including breaks.
The next section reports the results of the young group. Appendix A reports the
results of the elder group.
3.3.1. Learning effect on hand trajectories, kinematic outcomes, and performance
Figure 3.3.1, which gives examples of movement data for Pre1-test (before training),
1-day and 1-month retention test from a participant in the young group, show how MT
decreased following training as result of increased peak velocity. Velocity profiles that
were initially asymmetric (right-skewed) became symmetric with training. Figure
3.3.1(A) shows that when subject 8 reached a target on the left from the home position,
MT for each Testing session corresponded to 460 ms (Pre1), 322 ms (1-day), and 338 ms
(1-month). In addition, the maximum velocity corresponded to 98.8 cm/s (Pre1), 139.6
cm/s (1-day), and 138.4 cm/s (1-month). Furthermore, the mean symmetry ratio
increased from 0.339 (Pre1), to 0.407 (1-day) and 0.435 (1-month). Figure 3.3.1(B)
shows that when subject 8 reached a target on the right from the home position, MT for
each Testing session corresponded to 443 ms (Pre1), 270 ms (1-day), and 286 ms (1-
month). In addition, the maximum velocity corresponded to 145.5 cm/s (Pre1), 205.7
cm/s (1-day), and 199.9 cm/s (1-month). Furthermore, the mean symmetry ratio increased
from 0.354 (Pre1), 0.448 (1-day), and 0.423 (1-month).
F
tr
H
sy
v
3
ta
m
p
si
igure 3.3.1.
raining from
Hand path. S
ymbols). No
elocity, and
.3.2. Chang
Figure 3
argets in the
ms) indicate
articipants s
Post ho
ignificantly g
Examples
m a subject in
econd row:
ote that the su
that the velo
ge in movem
3.3.2(A) sho
ART system
es longer a
howed large
oc pairwise
greater MT i
of hand pat
n Pre1-test a
Tangential v
ubject show
ocity profiles
ment time
ows an avera
m workspac
and shorter
er MTs with
compariso
in Pre1 (i.e.,
32
ths and tang
and in 1-day
velocities an
ws a large dec
s become sy
age moveme
e. The black
MT, respe
larger distan
on tests wi
, 379 ± 15 m
gential hand
y and 1-mon
nd number o
crease in mo
ymmetrical w
ent time (M
k and white
ectively. Th
nce and angl
ith Bonferr
ms) than in P
d velocities
nth retention
f peaks (ind
ovement tim
with a single
MT) map, rep
colors (rang
he figure a
les prior to tr
roni correct
Post1 (i.e., 26
before and
n tests. First
dicated with
me, increased
peak.
presenting a
ge 213.9 to 4
also reveals
raining.
tions show
62 ± 15 ms),
after
row:
filled
d peak
all 35
472.1
that
ed a
, Pre2
33
(i.e., 281 ± 15 ms), Post2 (i.e., 247 ± 15 ms), 1-day (i.e., 245 ± 15 ms), and 1-month (i.e.,
248 ± 15 ms) (all p < 0.0001), greater MT in Post1 than in Pre2 (p < 0.009), and greater
MT in Pre2 than in Post2, 1-day, and 1-month (p < 0.003). Both the distance and angular
coefficients in the mixed regression model (more precisely for Test D and Test
cos(150 –q), respectively) decreased with training, with most of the decrease occurring in
the first training session. The decrease in both the distance and angular coefficients was
maintained in 1-day and 1-month retention tests (Figure 3.3.2(C) and 3.3.2(D)). The
distance coefficients of each test session decreased from 4.1 ± 0.3 in Pre1 to 1.3 ± 0.3 at
1-month (Figure 3.3.2(C)). The angular coefficients decreased from 74.4 ± 3.6 in Pre1 to
32.0 ± 3.6 (1-month) (Figure 3.3.2(D)). All fixed factor terms in the model (Test, Test
D, and Test cos(150-q)) were significant (p < 0.0001). A comparison of the average
MTs between the first block of the first training day and the last block of the second
training day, showed that the average MTs at Target 1, 3, and 5 significantly decreased (p
< 0.001) (Figure 3.3.2(E)).
F
b
fo
th
M
m
M
d
d
er
3
th
igure 3.3.2.
efore, durin
ollowing trai
he mixed reg
MT. D: Regr
mixed regres
Mean movem
ays (indicate
ay retention
rrors of the m
.3.3. Improv
Figure 3
he ART syst
A: Mean mo
ng, and afte
ining. C: Re
gression mo
ression coef
sion model
ment time du
ed with filled
n test, and d
mean of MT
ved feedforw
3.3.3(A) show
tem worksp
ovement tim
er training s
egression co
odel shows a
fficient of ta
shows a lon
uring trainin
d symbols: c
diamond for
T. * p < 0.007
ward contro
ws an averag
ace, as a fu
34
me across the
shows a lon
efficient of
a long-lastin
arget angle
ng-lasting red
ng. The dash
circle for Pre
r 1-month re
7, ** p < 0.0
ol
ge symmetry
unction of th
e test session
ng-lasting (
target distan
ng reduction
(Test cos
duction in th
hed vertical
e-test, square
etention test
0001.
y ratio map
he target loc
ns. B: Overal
(1 month) r
nce (Test D
n of the effe
s(150-q)) in
he effect of
lines separ
e for Post-te
t). Error bar
representing
cations. The
ll movement
reduction of
D) in each te
ct of distanc
each test i
angle on M
ate the four
est, triangle f
rs show stan
g all 35 targ
black and w
t time
f MT
est in
ce on
n the
MT. E:
r visit
for 1-
ndard
ets in
white
35
colors (range 0.290 to 0.455) indicates larger and smaller ratio, respectively. The figure
also reveals that participants showed a smaller ratio with larger distance and angles
(Figure 3.3.3(A)) prior to training.
The mean symmetry ratio of each test session was 0.355 ± 0.013 (Pre1), 0.411 ±
0.013 (Post1), 0.407 ± 0.013 (Pre2), 0.427 ± 0.013 (Post2), 0.427 ± 0.013 (1-day), and
0.421 ± 0.013 (1-month), as shown in Figure 3.3.3(A) and 3.3.3(B). Post hoc pairwise
comparison tests with Bonferroni corrections showed a significantly higher symmetry
ratio in Pre1 than in Post1, Pre2, Post2, and 1-day, and 1-month (all p < 0.0001).
Therefore, the increased symmetry ratio after 2-day reach training was sustained at 1-day
and 1-month (Figure 3.3.3(B)).
Both the distance and angular coefficients in the mixed regression model for
symmetry ratio increased with training, with most of the increase occurring in the first
training session (Figure 3.3.3(C) and 3.3.3(D)). The increase in both the distance and
angular coefficients were maintained in 1-day and 1-month retention tests (Figure 3.3.3(C)
and 3.3.3(D)). The distance coefficients of each test session increased from -0.00138 ±
0.00037 in Pre1 to 0.00069 ± 0.00037 at 1-month (Figure 3.3.3(C)). The angular
coefficients increased from -0.038 ± 0.005 in Pre1 to -0.009 ± 0.005 at 1-month (Figure
3.3.3(D)). The covariate fixed factor terms in the model (Test D and Test cos(150-q))
were significant (p < 0.0001). Comparing average symmetry ratio between the first block
of first training day and the last block of second training day, the average symmetry ratio
at Target 1, 3, and 5 were significantly decreased (p < 0.001) (Figure 3.3.3(E)).
F
b
ra
te
on
te
on
se
P
b
3
d
igure 3.3.3.
efore, during
atio followin
est in the mix
n symmetry
est in the mi
n symmetry
eparate the f
ost-test, tria
ars show sta
.3.4. Decrea
Compari
ay and the la
A: Mean sy
g, and after
ng training.
xed regressio
y ratio. D: Re
ixed regress
y ratio. E: M
four visit da
angle for 1-d
andard errors
ased variabi
ing the mean
ast block of
ymmetry rati
training sho
C: Regressi
on model sh
egression co
ion model s
Mean symme
ays (indicate
day retention
s of the mean
ility
n maximum
second train
36
o across the
ows a long-l
ion coefficie
hows a long-
oefficient of
shows a long
etry ratio du
ed with filled
n test, and di
n of symmet
acceleration
ning day, the
e test session
lasting (1 mo
ent of target
lasting reduc
f target angle
g-lasting red
uring trainin
d symbols: c
iamond for 1
try ratio. ** p
n between th
e mean maxim
ns. B: Overal
onth) reduct
t distance (T
ction of the
e (Test x cos
duction in th
ng. The dash
circle for Pr
1-month rete
p < 0.0001.
he first block
mum acceler
ll symmetry
tion of symm
Test x D) in
effect of dis
s(150-q)) in
he effect of
hed vertical
re-test, squar
ention test).
k of first tra
ration at Tar
y ratio
metry
each
stance
n each
angle
lines
re for
Error
aining
rget 1,
3
st
tr
m
(F
d
0
F
m
(C
, and 5 were
tandard devi
raining day a
maximum ac
Figure 3.3.4
ay and the l
.001) (Figur
.
igure 3.3.4.
mean standar
CV) on max
e significantl
iation of the
and the last b
cceleration a
, second row
last block of
re 3.3.4, third
First row: M
rd deviation
imum accele
ly increased
e mean max
block of sec
at Target 1,
w). Compar
f second trai
d row).
Mean maxim
of maximu
eration. The
37
(p < 0.001)
ximum acce
cond training
3, and 5 w
ing the CV
ining day, th
mum accelera
um accelerat
dashed verti
) (Figure 3.3
eleration betw
g day, the sta
were signific
between th
he CV were
ation during
tion. Third r
ical lines sep
.4, first row
tween the fi
andard devia
cantly decrea
e first block
significantl
g 2-day train
row: Coeffic
parate the tw
w). Comparin
rst block of
ation of the m
ased (p < 0
k of first tra
ly decreased
ning. Second
cient of var
wo training d
ng the
f first
mean
0.001)
aining
d (p <
d row:
riance
days.
3
C
an
3
th
w
F
h
w
te
o
h
.3.5. Increa
Figure 3.
Comparing av
nd the last b
, and 5 signi
he final elbo
were no signi
igure 3.3.5.
eight during
with filled sy
est, and diam
f MT.
Figure 3
orizontal, an
sed initial e
3.5(A) show
verage initia
lock of the s
ificantly incr
ow heights in
ificant chang
A: Change o
g training. T
ymbols: circ
mond for 1-m
3.3.6 shows
nd torsion,
elbow height
ws that the
al elbow hei
second traini
reased (p <
ncreased wi
ges at Target
of initial elb
The dashed v
le for Pre-te
month retenti
the ranges o
and elbow
38
t and 3D kin
e initial el
ights betwee
ing day, the
0.001) (Figu
ith training a
t 5 (p > 0.05)
bow height d
vertical line
est, square f
ion test). Err
of motion in
angles duri
nematic ana
lbow height
en the first b
average init
ure 3.3.5(A)
at Target 1 a
).
during trainin
s separate th
for Post-test,
ror bars show
n shoulder a
ing training
alysis
ts increased
block of the
tial elbow he
). Figure 3.3
and 3 (p < 0
ng. B: Chan
he four visi
, triangle fo
w standard e
angles, i.e., s
g. Comparin
d with trai
first training
eights at Targ
3.5(B) show
0.001), and
nge of final e
it days (indi
r 1-day rete
errors of the m
shoulder ver
g average i
ining.
g day
get 1,
s that
there
elbow
icated
ention
mean
rtical,
initial
39
elbow heights between the first block of the first training day and the last block of the
second training day, the ranges of motion in shoulder vertical angles were significantly
decreased with training at Target 3 and 5 (p < 0.001), but there were no significant
change at Target 1 (p > 0.05) (Figure 3.3.6(A)). Figure 3.3.6(B) shows that the ranges of
motion in shoulder horizontal angles significantly decreased with training at Target 1, 3,
and 5 (p < 0.001). Figure 3.3.6(C) shows that the ranges of motion in torsion angles
significantly decreased with training at Target 1 and 3 (p < 0.001), and there was no
significant change at Target 5 (p > 0.05). Figure 3.3.6(D) shows that the ranges of motion
in elbow angles were significantly increased with training at Target 1 (p < 0.001), but
there were no significant change at Target 5 (p > 0.05).
F
A
sh
tr
se
P
3
un
igure 3.3.6.
A: Range of
houlder hori
raining. D: R
eparate the f
ost-test, trian
.4. Discussio
The exp
nassisted int
Range of m
motion of s
izontal angle
Range of mo
four visit da
ngle for 1-da
on
periments d
tensive reach
motion of sho
shoulder ver
e during tra
otion of elb
ays (indicate
ay retention
discussed in
h training in
40
oulder and el
rtical angle
aining. C: R
ow angle du
ed with filled
test, and dia
n this chap
n young and
lbow angles
during train
Range of mo
uring trainin
d symbols: c
amond for 1-
pter demons
d elder non-d
in 3-dimens
ning. B: Ran
otion of tors
ng. The dash
circle for Pr
-month reten
strate that t
disabled righ
sional works
nge of motio
sion angle d
hed vertical
re-test, squar
ntion test).
two session
ht-handed gr
space.
on of
during
lines
re for
ns of
roups
41
can induce motor skill learning in fast reaching movements. In addition, the experiments
identifies the key characteristics that induce non-disabled individuals to achieve motor
skill learning: 1) increase in speed; 2) long-lasting improvement in performance; 3)
generalization to untrained movements; 4) increased reliance on feedforward control; 5)
decrease in variability; and 6) improvement in movement efficiency.
The intensive reach training shows significant and durable motor performance
improvement in both groups. In the young group, compared to Pre-test, movement time
decreases on average 34.5 ± 2.9% at the 1-day retention test and 33.7 ± 2.8% at the 1-
month retention test. Similarly, the symmetry ratio of the velocity profile increases on
average 21.7 ± 5.9% at the 1-day retention test and 20.8 ± 5.9% at the 1-month retention
test. In the elder group, compared to Pre-test, movement time decreases on average 39.3
± 3.3% at the 1-day retention test and 38.0 ± 4.1% at the 1-month retention test. Similarly,
the symmetry ratio of the velocity profile increases on average 21.7 ± 5.9% at the 1-day
retention test and 20.8 ± 5.9% at the 1-month retention test. The improvements in both
movement time and symmetry ratio are not only long lasting but the training induces
broad generalization of improvement to movements to non-trained targets. The training to
the five targets induces reduction of movement time and symmetry ratio to the 35 testing
targets, with performance improving for all targets (Figure 3.3.2 and Figure 3.3.3).
Besides reducing overall movement time, training induces a “flattening” of movement
time over all target locations, as shown by the decreased distance and angular regression
coefficients during training (Figure 3.4.2). We interpret the flattening effects as follows.
Movement time varies as a function of the distance (Fitts's law; speed and accuracy trade-
off (Fitts, 1954; Fitts & Peterson, 1964; van Beers, Haggard, & Wolpert, 2004) and angle
42
(inertia (Beer et al., 2004; Gordon, Ghilardi, Cooper, & Ghez, 1994)). For a function of
the angle, the flattening effect provides evidence that movements in the direction of the
long axis of the arm inertial ellipse are longer than movements in the direction of the
short axis (Gordon et al., 1994). Therefore, the flattening effect supports the ideathat two
possible non-exclusives mechanisms explain these findings. First, participants only use
an imperfect internal model of the dynamics, and thus generate similar initial acceleration
at the independent hand directions (Gordon et al., 1994). Second, previous optimal
control models suggest that participants minimize a constant effort for all reaches extent
and directions (Guigon, Baraduc, & Desmurget, 2007). However, the flattening effect as
a result of the UE intensive training can be interpreted in two different ways. First, the
participants may learn to update an internal model of the dynamics by generating
different acceleration at the independent hand directions. Second, the participants do not
use a “constant effort” strategy to compensate large inertia at the direction of the long
axis of the arm inertial ellipse.
The symmetry ratios increased up to 0.5 in both groups (see Figure 3.3.3(E) and
Appendix 3(E) in chapter 3). We interpret that the increased symmetry ratio represents
more reliable feedforward control for the rapid reaching movements. We can also
interpret the decreased CV as decreased movement variability.
All participants in the young group show that increased initial elbow height induces
less range of motion (ROM) in shoulder angles and more ROM in elbow angles during
reaching. We suggest that the new strategy to move quickly by changing initial posture
helps to minimize the inertia of arm (i.e., move elbow joint more than both shoulder and
elbow joints) as a result of efficiency of movements in joint coordination. Compared to
43
the inertia of whole arm movements, inertia with more weighted forearm may be
relatively small. In conclusion, the results of the experiment described in this chapter
support our hypotheses that an ART system used with intensive training can lead to faster
motor skill learning. Therefore, the future study is to investigate the reason of the initial
posture change mathematically and to verify the efficiency of movements in joint
coordination.
This study thus demonstrates that short-duration intensive training induces motor
skill learning with specific characteristics to move fast. In addition, to our knowledge,
previous studies that mostly have reported characteristics of motor skill learning either in
settings with artificial perturbation or in uncommon setting in real-life. Thus, our results
support the view that intensive training leading the characteristics might be effective to
lead motor skill learning and this work might contribute to achieve greater and faster
benefic in sports and rehabilitation.
Acknowledgements
This work was in part supported by NSF BCS 1031899 and NIH R01 HD065438.
A
T
el
F
tr
H
sy
v
p
an
cm
Appendix A.
This appendix
lder group.
Figure A1.
raining from
Hand path. S
ymbols). No
elocity, and
Figure A
osition, the M
nd 243 ms (
m/s (Pre1),
.
x A discusse
Examples o
m a subject in
econd row:
ote that the su
that the velo
A1 shows th
MT for each
1-month) (se
212.5 cm/s
es the experi
of hand path
n Pre1-test a
Tangential v
ubject show
ocity profiles
hat when sub
h Testing ses
ee Figure A1
(1-day), and
44
ment describ
hs and tang
and in 1-day
velocities an
ws a large dec
s become sy
bject 3 reac
sion corresp
1(A)). The m
d 208.5 cm/
bed in Chapt
gential hand
y and 1-mon
nd number o
crease in mo
ymmetrical w
ched a targe
ponded to 45
maximum ve
/s (1-month)
ter 2 as perf
d velocities
nth retention
f peaks (ind
ovement tim
with a single
et on the lef
52 ms (Pre1)
elocity corre
). The mean
formed by th
before and
n tests. First
dicated with
me, increased
peak.
ft from the h
), 225 ms (1-
sponded to
n symmetry
he
after
row:
filled
d peak
home
-day),
124.1
ratio
45
increased from 0.270 (Pre1), to 0.422 (1-day) and 0.395 (1-month). Figure A1(B) shows
that when subject 3 reached a target on the right from the home position, the MT for each
Testing session corresponded to 278 ms (Pre1), 173 ms (1-day), and 283 ms (1-month).
The maximum velocity corresponded to 172.7 cm/s (Pre1), 300.8 cm/s (1-day), and 272.5
cm/s (1-month), and the mean symmetry ratio increased from 0.407 (Pre1), 0.445 (1-day),
and 0.426 (1-month).
F
b
fo
th
M
m
M
w
te
o
w
in
M
Figure A2. A
efore, durin
ollowing trai
he mixed reg
MT. D: Regr
mixed regres
Mean MT dur
with filled sy
est, and diam
f MT. * p <
Figure A
workspace. T
ndicated long
MTs with larg
A: Mean mov
ng, and afte
ining. C: Re
gression mo
ression coef
sion model
ring training
ymbols: circ
mond for 1-m
0.05, ** p <
A2 shows an
The black a
ger and shor
ger distance
vement time
er training s
egression co
odel shows a
fficient of ta
shows a lon
g. The dashe
le for Pre-te
month retenti
< 0.0001.
average MT
and white co
rter MT, resp
and angles (
46
e across the
shows a lon
efficient of
a long-lastin
arget angle
ng-lasting red
d vertical lin
est, square f
ion test). Err
T map, repre
olors (range
pectively. Be
(Figure A2(A
test sessions
ng-lasting (
target distan
ng reduction
(Test cos
duction in th
nes separate
for Post-test,
ror bars show
esenting all 3
e 210.4 to
efore trainin
A)).
s. B: Overal
(1 month) r
nce (Test D
n of the effe
s(150-q)) in
he effect of
the four vis
, triangle fo
w standard e
35 targets in
534.7 ms) i
ng, participan
ll movement
reduction of
D) in each te
ct of distanc
each test i
angle on M
it days (indi
r 1-day rete
errors of the m
n the ART sy
in Figure A
nts showed l
t time
f MT
est in
ce on
n the
MT. E:
icated
ention
mean
ystem
A2(A)
larger
47
Overall MT decreased with training (Figure A2(A) and A2(B)). Post hoc pairwise
comparison tests with Bonferroni corrections show a significantly greater MT in Pre1
(i.e. 423.5 ± 14 ms) than in Post1 (i.e. 281 ± 14 ms), Pre2 (i.e. 289 ± 13 ms), Post2 (i.e.
257 ± 13 ms), 1-day (i.e. 250 ± 14 ms), and 1-month (i.e. 255 ± 14 ms) (p < 0.0001),
greater MT in Post1 than in 1-day (p = 0.040), and greater MT in Pre2 than in Post2, 1-
day, and 1-month (p < 0.022). Both the distance and angular coefficients in the mixed
regression model (more precisely for Test D and Test cos(150 –q), respectively)
decreased with training, with most of the decrease occurring in the first training session.
The decrease was maintained in 1-day and 1-month retention tests (Figure A2(C) and
A2(D)). The distance coefficients of each test session decreased from 4.4 ± 0.3 in Pre1 to
2.1 ± 0.3 at 1-month (Figure A2(C)). The angular coefficients decreased from 61.7 ± 3.4
in Pre1 to 40.5 ± 3.4 (1-month) (Figure A2(D)). All fixed factor terms in the model (Test,
Test D, and Test cos(150-q)) were significant (p < 0.0001). Comparing the average
MT between the first block of the first training day and the last block of second training
day, the average MT at Target 1, 3, and 5 significantly decreased (p < 0.001) (Figure
A2(E)).
F
b
ra
te
on
te
on
se
P
b
th
co
tr
Figure A3. A
efore, during
atio followin
est in the mix
n symmetry
est in the mi
n symmetry
eparate the f
ost-test, tria
ars show sta
Figure A
he ART sys
olors (range
raining, parti
A: Mean sym
g, and after
ng training.
xed regressio
y ratio. D: Re
ixed regress
y ratio. E: M
four visit da
angle for 1-d
andard errors
A3(A) shows
tem worksp
e 0.274 to 0
icipants show
mmetry ratio
training sho
C: Regressi
on model sh
egression co
ion model s
Mean symme
ays (indicate
day retention
s of the mean
s an average
pace as a fun
.442) indica
wed a smalle
48
o across the
ows a long-l
ion coefficie
hows a long-
oefficient of
shows a long
etry ratio du
ed with filled
n test, and di
n of symmet
e symmetry
nction of th
ated a larger
er ratio with
test session
lasting (1 mo
ent of target
lasting reduc
f target angle
g-lasting red
uring trainin
d symbols: c
iamond for 1
try ratio. ** p
ratio map, r
he target loc
r and smalle
larger distan
s. B: Overal
onth) reduct
t distance (T
ction of the
e (Test cos
duction in th
ng. The dash
circle for Pr
1-month rete
p < 0.0001.
representing
cations. The
er ratio, resp
nce and ang
ll symmetry
tion of symm
Test D) in
effect of dis
s(150-q)) in
he effect of
hed vertical
re-test, squar
ention test).
g all 35 targe
black and w
pectively. B
les.
ratio
metry
each
stance
each
angle
lines
re for
Error
ets in
white
Before
49
Overall symmetry ratio increased with training (Figure A3(A) and A3(B)). The
mean symmetry ratio of each test session was 0.349 ± 0.010 (Pre1), 0.410 ± 0.010 (Post1),
0.401 ± 0.009 (Pre2), 0.416 ± 0.009 (Post2), 0.416 ± 0.010 (1-day), and 0.409 ± 0.010 (1-
month) (Figure A3(B)). Post hoc pairwise comparison tests with Bonferroni corrections
show a significantly higher symmetry ratio in Pre1 than in Post1, Pre2, Post2, and 1-day,
and 1-month (all p < 0.0001). Therefore, the increased symmetry ratio after 2-day reach
training was sustained at 1-day and 1-month (Figure A3(B)).
Both the distance and angular coefficients in the mixed regression model for
symmetry ratio increased with training, with most of the increase occurring in the first
training session (Figure A3(C) and A3(D)). The increase was maintained in 1-day and 1-
month retention tests (Figure A3(C) and A3(D)). The distance coefficients of each test
session increased from -0.00001 ± 0.00037 in Pre1 to 0.00022 ± 0.00037 at 1-month
(Figure A3(C)). The angular coefficients increased from -0.021 ± 0.005 in Pre1 to -0.005
± 0.005 at 1-month (Figure A3(D)). The fixed factor terms in the model (Test and Test
cos(150-q)) were significant (p < 0.0001). Comparing average symmetry ratio between
the first block of the first training day and the last block of second training day, the
average symmetry ratio at Target 1, 3, and 5 significantly decreased (p < 0.001) (see
Figure A3(E)).
F
m
(C
d
3
st
tr
m
(F
Figure A4. F
mean standar
CV) on max
Comparin
ay and the la
, and 5 were
tandard devi
raining day a
maximum ac
Figure A4, s
First row: M
rd deviation
imum accele
ng the mean
ast block of
e significant
iation of the
and the last b
cceleration a
second row).
Mean maximu
of maximu
eration. The
maximum
second train
tly increased
e mean max
block of sec
at Target 1,
. Comparing
50
um accelera
um accelerat
dashed verti
acceleration
ning day, the
d (p < 0.001
ximum acce
cond training
3, and 5 w
g the CV bet
ation during
tion. Third r
ical lines sep
n between th
e mean maxim
1) (Figure A
eleration betw
g day, the sta
were signific
tween the fir
2-day traini
row: Coeffic
parate the tw
he first block
mum acceler
A4, first row)
tween the fi
andard devia
cantly decrea
rst block of
ing. Second
cient of var
wo training d
k of first tra
ration at Tar
). Comparin
rst block of
ation of the m
ased (p < 0
first training
row:
riance
days.
aining
rget 1,
ng the
f first
mean
0.001)
g day
51
and the last block of second training day, the CV were significantly decreased (p < 0.001)
(Figure A4, third row).
52
CHAPTER 4: Short-duration and intensive training improves long-term reaching
performance in individuals with chronic stroke
4.1. Introduction
About 65% of stroke survivors
experience long-term limitations in upper extremity
(UE) functions (Lum et al., 2009). In particular, long-term limitations are prominent in
arm reaching performance, which correlates strongly with patients’ general impairment
levels (Kamper et al., 2002; van Dokkum et al., 2013). After stroke, reaching movements
show increased movement time, multiple velocity peaks, and high variability (Cirstea,
Mitnitski, et al., 2003; Kamper et al., 2002; McCrea & Eng, 2005). Movements in
directions that require inter-joint coordination are most impaired (Beer et al., 2004;
Levin, 1996). In addition, there is a greater increase of movement time as a function of
distance with the affected arm than with the less affected arm (McCrea & Eng, 2005).
Rehabilitation of UE functions in individuals post-stroke is clinically important
because a substantial number of daily activities involve use of the arms and hands (Coster
et al., 2004). Task-specific intensive training over multiple sessions can improve UE
function in individuals post-stroke (Aisen et al., 1997; Birkenmeier et al., 2010;
53
Blennerhassett & Dite, 2004; Dipietro et al., 2012; Ellis et al., 2009; Lo et al., 2010; Lum
et al., 2006; D. J. Reinkensmeyer et al., 2004; J. Reinkensmeyer, 2001; Rohrer et al.,
2002; Schaefer, Patterson, & Lang, 2013; Thielman et al., 2004; C. J. Winstein et al.,
2004; Wolf et al., 2006). In particular, reach training post-stroke has been extensively
studied, either with or without rehabilitation robotic systems (Aisen et al., 1997;
Birkenmeier et al., 2010; Dipietro et al., 2012; Ellis et al., 2009; Lo et al., 2010; Lum et
al., 2006; D. J. Reinkensmeyer et al., 2004; J. Reinkensmeyer, 2001; Rohrer et al., 2002;
Schaefer et al., 2013; Thielman et al., 2004; C. J. Winstein et al., 2004; Wolf et al., 2006).
Multiple sessions of reach training improve movement speed and smoothness (Dipietro et
al., 2012; Ellis et al., 2009; J. Reinkensmeyer, 2001; Rohrer et al., 2002; Thielman et al.,
2004), as well as clinical task scores (Birkenmeier et al., 2010; Blennerhassett & Dite,
2004; Dipietro et al., 2012; Lo et al., 2010; C. J. Winstein et al., 2004; Wolf et al., 2006).
For example, clinical trials of UE training with large number of sessions (e.g., 18 and 36
sessions in (Dipietro et al., 2012) and (Lo et al., 2010), respectively) led to long-term
gains and generalization to untrained tasks, the two main requirements of effective
rehabilitation.
In practical clinical settings, however, both overall training time and number of
sessions are typically smaller than in clinical trials (Keith & Cowell, 1987; Lang et al.,
2009; Lincoln, Willis, Philips, Juby, & Berman, 1996; Mackey, Ada, Heard, & Adams,
1996). Thus, it is of high clinical importance to develop short-duration UE training
methods that lead to long-lasting gains and that generalize beyond the trained tasks. The
following studies have reported effects of short-duration training on UE function post-
stroke. First, seventy reach trials in one training session were found to improve
54
movement time and elbow-shoulder coordination (Cirstea, Mitnitski, et al., 2003).
Second, sixty reach-to-grasp trials in a training session with trunk restraint induced more
elbow extension, better inter-joint coordination, and fewer trunk movements (Michaelsen
& Levin, 2004). Third, two hundred reach trials in a training session were found to
improve response time; the improvements were maintained 24 hours later, but not 1
month later (Harris-Love et al., 2011). Finally, training on a specific (feeding) task for 50
trials a day for 5 consecutive days was found to improve performance and also
generalized two other (i.e., sorting and dressing) untrained tasks (Schaefer et al., 2013).
These studies thus provide limited evidence (Class II, Level B (Ringleb et al., 2010)) that
short-duration reach training promote short-term gains in UE performance post-stroke.
The effects of short-duration training on long-term gains, however, remain unclear.
This chapter is to evaluate the long-term and generalization effects of short-duration
and intensive reach training in individuals with chronic stroke and mild to moderate
impairments. Previous research suggest that training of arm movements post-stroke
should contain three important characteristics: First, because the amount of practice is the
most important parameter in motor learning, it has been recommended that individuals
post-stroke perform as many repetitions per training session as they can tolerate (Party,
2008). Second, movements need to be constrained to minimize redundant degrees of
freedom to perform the task. In particular, there is good evidence that trunk restraint
during reaching is important to enhance recovery of arm function (Michaelsen & Levin,
2004; Roby‐Brami et al., 2003). Third, practicing challenging tasks, but not simple
repetitive tasks, is likely to elicit motor learning and associated neural reorganization
(Nudo et al., 1996; Plautz, Milliken, & Nudo, 2000; Sanger, 2004). Performance-based
55
positive and negative feedback helps to keep tasks challenging (Duff et al., 2010; Molier,
Van Asseldonk, Hermens, & Jannink, 2010). We thus tested the hypothesis that two
sessions of unassisted reach training of 600 movements each, with trunk restraint, and
with display of performance-based feedback at each trial, will lead to long-lasting (1
month) improvements in movement time and smoothness both for trained and for
untrained movements.
4.2. Methods
4.2.1. Participants
Sixteen participants with ischemic or hemorrhagic stroke with mild to moderate
impairments (63.2 ± 2.7 years; 14 males, 2 females), subsequently referred to as the
stroke group, were recruited. Potential participants were included if they (1) were at least
6 months post-stroke; (2) had residual capability to move their UE (Upper Extremity
Fugl-Meyer motor score > 19/66); (3) had the ability to follow and remember instruction
(Mini-Mental State Examination (MMSE) score > 25/30) (Folstein, Folstein, & McHugh,
1975); and (4) were able to perform an unassisted reach to the farthest target displayed at
40 cm from the anterior edge of the Arm Reach Training (ART) system table.
Participants were excluded if they had (1) any neurologic diagnoses other than stroke; (2)
peripheral movement restrictions, such as neuropathy; (3) orthopedic disorders affecting
the paretic UE; (4) severe pain or sensory impairment; or (5) visual neglect (more than
4% of lines left uncrossed on Albert’s test) (Fullerton, McSherry, & Stout, 1986). All
en
d
su
p
E
U
an
T
=
=
nrolled part
emographic
Ten non
ubsequently
erformance.
Edinburgh H
University of
nd signed an
Table 4.2.1. D
= Mini-menta
= Box and Bl
ticipants had
information
n-disabled ag
referred to a
Only righ
Handedness
f Southern C
n informed c
Demographi
al state exam
lock Test; SE
d a score o
n for the strok
ge-matched
as the contro
ht-hand dom
Inventory (
California’s I
onsent prior
c informatio
mination sco
E = Standard
56
of 0 on Alb
ke group.
participants
ol group, wer
minant peop
Oldfield, 19
Institutional
r to study enr
on for the 16
res; FM = U
d error.
bert’s test. T
(56.6 ± 2.9
re recruited
ple were in
971). This s
Review Bo
rollment.
6 participants
UE score of
Table 4.2.1
9 years; 5 m
for a compa
ncluded as
study was a
oard, and all
s in the strok
Fugl-Meyer
summarize
males, 5 fem
arison of reac
assessed by
approved by
participants
ke group. M
r motor test;
es the
males),
ching
y the
y the
s read
MMSE
BBT
4
P
(s
w
co
pr
w
la
p
4
p
F
th
.2.2. Study d
This pro
articipants i
see Figure 4
with the AR
onsecutive
receded by a
was given. In
ab one more
articipants in
.2.1). Partici
erform a fam
igure 4.2.1.
he stroke gro
design
oof-of-conce
in the stroke
4.2.1). The f
RT system.
days. The f
a pre-test an
n the fourth w
e time for a
n the stroke
ipants in the
miliarization
A: Diagram
oup.
ept study w
e group visi
first visit co
In the follo
first two da
d followed b
week follow
1-month ret
group: Pre1
control grou
session with
m showing th
57
was a within
ited the labo
omprised cli
owing week
ays were tr
by a post-tes
wing the two
tention test.
1, Post1, Pre
up visited th
h the ART sy
he timing of
n-participant
oratory five
inical tests a
k, participan
raining days
st. On the thi
training ses
Thus, a tot
e2, Post2, 1-
he laboratory
ystem and th
f the five vis
t, repeated-m
times over
and a famili
nts visited
s, with the
ird day, a 1-
sions, partic
tal of six tes
day and 1-m
y one time, d
hen a single
sits over a 6
measures de
a 6-week p
iarization se
the lab in
training se
day retentio
cipants visite
sts were giv
month (see F
during which
arm reach te
6-week perio
esign.
period
ession
three
ession
n test
ed the
ven to
Figure
h they
est.
od for
58
4.2.3. Clinical assessments
The upper-extremity score of the Fugl-Meyer motor (FM) test (Fugl-Meyer, Jääskö,
Leyman, Olsson, & Steglind, 1974) was performed at baseline. The Box and Block test
(BBT) (Mathiowetz, Volland, Kashman, & Weber, 1985) was performed at baseline test
(i.e., before training), and in the 1-day and 1-month retention test sessions to assess any
transfer effect of the ART training on arm and hand function. All participants
successfully completed all arm reach training and test sessions, and the clinical
assessments, with the exception of one participant who could not complete the BBT at
the baseline test due to a schedule conflict.
4.2.4. Data analysis
The sensor position data were low-pass filtered with a cutoff frequency of 5 Hz
using a zero phase-lag second-order Butterworth filter. Movement time (MT) for offline
data analysis was computed by the interval between the time at which the tangential
velocity of the index fingertip first exceeded 5% of maximum velocity (Coderre et al.,
2010) and the time at which the velocity fell below 5% of maximum velocity with the
fingertip inside of the target.
on line
MT
and MT were highly correlated (r=0.988 and p <
0.0001, Pearson), validating the on-line feedback method. Movement smoothness was
determined by counting the number of positive peaks in the velocity profile. The positive
peaks were counted by finding inflection points that are points on the velocity profile at
59
which the sign of the acceleration profile (i.e., time derivatives of velocity) changes from
positive to negative.
4.2.5. Mixed regression models
To identify the temporal and spatial characteristics of performance and changes in
performance over time, we performed linear mixed model analysis (using SPSS 18) with
either MT or number of peaks as the dependent variable and subjects’ random intercept.
To test for short- and long-term training effects, Test (Pre-, Post-, 1-day and 1-month
follow-up session) was included as a fixed repeated factor. Target distance (D), the
distance between the center of home-position and the center of presented target ranging
from 10 to 30 cm, with 5 cm increments, and cosine of the target angle (cos(150-q),
where q is the target angle ranging from 30 to 150 degrees) were included as co-variates.
To test for possible training-induced generalization effect of target distance and target
angles, interaction terms Test D and Test cos(150-q) were included. Model
comparisons (using the Bayesian information criterion, BIC, as selection factor) showed
that repeated measures were best modeled with an auto-regressive moving average
(ARMA(1,1)).
Because of our small data pool, we combined the data from participants with left and
right hemiparesis. To remove left-right effects, the data were “flipped” along the mid-
line, so that all participants in the stroke group “behaved” as right hemiparetic
participants. To validate this approach, we fitted individual models of MT in Pre1-test
60
with target distance (D) and cosine of the target angle (cos(150-q)) to determine whether
spatial movement characteristics were qualitatively similar between subjects and
symmetric across the midline for left and right hemiparesis. Bonferroni correction was
used for multiple pairwise comparisons. The level of statistical significance was set at p <
0.05. All results were reported with mean ± standard errors (SE) of the corresponding
mean.
4.3. Results
4.3.1. Demographic information and overall training effect
Average stroke duration was 72 ± 11 months and the average FM score was 48.7 ±
2.5 (Table 1). The control and stroke groups did not differ in age (t-test, p = 0.117). All
participants completed 600 trials of the two-day training session and successfully reached
all targets within 5 seconds. Performing the 600 movements in each Training block lasted
on average 106.0 ± 2.9 minutes (range: 63.9 to 143.5 minutes) in the stroke group and
90.3 ± 3.8 minutes (range: 64.7 to 117.7 minutes) in the control group including break
times.
Figure 4.3.1 shows examples of movement data for Pre1-test (before training), 1-day
and 1-month retention test from two participants post-stroke. These examples indicate
that MT decreased following training concomitant with a decrease in the number of peaks.
This is especially striking for the more impaired participant (Subject 10, FM = 30) for the
target shown in Figure 4.3.1(A), i.e., mean MT decreased from 1518 ms (Pre1), to 721
m
fr
F
tr
an
v
d
1
m
se
th
ms (1-day), a
rom 5 (Pre1)
igure 4.3.1.
raining for tw
nd in 1-day
elocities an
iamond for
0 on the left
movement tim
everity score
Overall,
he initial FM
and to 686 m
), to 1 (1-day
Examples
wo subjects
and 1-mont
d number o
1-day retenti
t, with relativ
me and num
e (FM = 51/6
for the stro
M score and
ms (1-month)
y), and incre
of hand pat
post-stroke
th retention
of peaks (in
ion test, and
vely high sev
mber of peak
66).
oke group, th
the initial M
61
). Furthermo
ased to 2 (1-
ths and tang
(Subject 10
tests. First r
ndicated wi
d square for 1
verity score
s compared
here was a
MT (
2
R = 0.5
ore, the mean
-month).
gential hand
and Subject
row: Hand p
ith filled sy
1-month rete
(FM = 30/6
with subjec
significant n
53, p < 0.00
n number of
d velocities
t 5 from Tab
path. Second
ymbols: circ
ention test).
66), shows a
ct 5 on the ri
negative cor
01). The init
f peaks decre
before and
ble 1) in Pre
d row: Tange
cle for Pre1
Note that su
large decrea
ight with a l
rrelation bet
ial FM score
eased
after
1-test
ential
1-test,
ubject
ase in
lesser
tween
e and
62
the initial number of peaks were negatively correlated (
2
R = 0.65, p < 0.0002). As a
result, there was a significant positive correlation between the initial MT and the initial
number of peaks (
2
R = 0.85, p < 0.001).
4.3.2. Movement time at baseline
Before training, participants in the stroke group moved on average slower than the
control group (MT in the stroke group: 740 ± 47 ms; control group: 400 ± 19 ms; t-test, p
< 0.0001). Note that both groups showed larger MTs in movements with larger target
distances and at greater angles from midline (Figure 4.3.2(A)-Pre1 and 4.3.2(B)).
We first fitted individual models of MT in Pre1-test with target distance (D) and
cosine of the target angle (cos(150-q)). In the control group, six out of the ten participants
showed a significant effect of target distance, and nine out of ten participants showed a
significant effect of target angle. In the stroke group, all participants showed a significant
effect of target distance, and thirteen participants showed a significant effect of target
angle. For example, six of the eight participants with left hemiparesis in the stroke group
exhibited a significant negative effect of target angle, indicating increased MT for targets
to the right, whereas all participants with right dominant hand in the control group and
seven of the eight participants with right hemiparesis in the stroke group exhibited a
significant positive effect of target angle, indicating increased MT for targets to the left.
63
4.3.3. Decrease in movement time with training
Overall MT decreased with training (Figure 4.3.2(A) and 4.3.2(C)). Post hoc
pairwise comparison tests with Bonferroni corrections showed a significantly greater MT
in Pre1 (740 ± 47 ms) than in Post1 (614 ± 47 ms), Pre2 (617 ± 47 ms), Post2 (535 ± 47
ms), 1-day (548 ± 47 ms), and 1-month (556 ± 47 ms) (all p < 0.0001) and greater MT in
Pre2 than in Post2 (p < 0.0001) and 1-day (p = 0.007). Both the distance and angular
coefficients in the mixed regression model decreased with training, with decrease
maintained in 1-day and 1-month retention tests (see Figure 4.3.2(D) and 4.3.2€). The
distance coefficients of each test session decreased from 14.1 ± 0.7 in Pre1 to 6.9 ± 0.7 at
1-month (Figure 4.3.2(D)). The angular coefficients decreased from 98.3 ± 9.0 in Pre1 to
39.0 ± 9.0 at 1-month (Figure 4.3.2(E)). For comparison, the coefficients of distance and
angle were 4.11 ± 0.3 and 46.1 ± 4.3, respectively, in the control group. All fixed factor
terms in the model (Test, Test D, and Test cos(150-q)) were significant (p < 0.014 or
less).
F
M
af
fo
th
M
ea
an
igure 4.3.2.
Mean movem
fter training
ollowing trai
he mixed reg
MT in the str
ach test in t
ngle on MT
A: Mean m
ment time in t
g in the stro
ining. D: Re
gression mo
roke group.
the mixed re
in the stroke
movement ti
the control g
oke group
egression co
odel shows a
E: Regressi
egression mo
e group. * p
64
ime in the s
group. C: Ov
shows a lon
efficient of
a long-lastin
on coefficie
odel shows
< 0.007, **
stroke group
verall movem
ng-lasting (
target distan
ng reduction
ent of target
a long-lastin
p < 0.0001.
p across the
ment time be
(1 month) r
nce (Test D
n of the effe
angle (Test
ng reduction
test session
efore, during
reduction of
D) in each te
ct of distanc
t cos(150-q
n in the effe
ns. B:
g, and
f MT
est in
ce on
q)) in
ect of
65
4.3.4. Change in movement smoothness
The control group generated a single peak for most movements (even without training),
with no patterns across distance and angle (mean number of peaks, 1.06 ± 0.02, range
from 1.0 to 1.3), whereas the average number of peaks before training in the stroke group
was significantly greater than in the control group (t-test, p < 0.0001). The mean number
of peaks (i.e., 2.22 ± 0.19) during the Pre1 in the stroke group ranged from 1.5 to 2.8.
The overall number of peaks decreased with training in the stroke group (Figure
4.3.3(B)). The number of peaks in Pre1 (i.e., 2.22 ± 0.19) was greater than in Post1 (i.e.,
1.68 ± 0.18), Pre2 (i.e., 1.80 ± 0.18), Post2 (i.e., 1.44 ± 0.18), and 1-day (i.e., 1.66 ± 0.18),
and 1-month (i.e., 1.67 ± 0.19) (all p < 0.0001). Number of peaks in was greater in Pre2
than in Post2 (p < 0.0001) (Figure 4.3.3(B)).
Similarly to movement time, both the distance and angular coefficients in the mixed
regression model for movement peaks decreased with training (Test D and Test
cos(150-q), both p < 0.0001), with decrease maintained in 1-day and 1-month retention
tests (Figure 4.3.3(C) and 4.3.3(D)). The distance coefficients of each test session
decreased from 0.027 ± 0.005 in Pre1 to -0.001 ± 0.005 at 1-month (Figure 4.3.3(C)). The
angular coefficients decreased from 0.435 ± 0.06 in Pre1 to 0.071 ± 0.06 at 1-month
(Figure 4.3.3(D)). For comparison, the coefficients of distance and angle were -0.0002 ±
0.002 and 0.011 ± 0.024, respectively, in the control group.
F
th
m
la
co
a
R
re
in
igure 4.3.3.
he control gr
movements).
asting (1 mo
oefficient of
long-lasting
Regression c
egression mo
n the stroke g
A: Mean nu
roup are not
B: Overall n
onth) reduct
f target dista
g reduction o
coefficient o
odel shows a
group. ** p <
umber of pea
t shown bec
number of p
tion in the n
ance (Test
of the effect
of target ang
a long-lastin
< 0.0001.
66
ak in the stro
ause the me
peaks before,
number of p
D) in each t
of distance o
gle (Test ng reduction
oke group ac
ean number
, during, and
peaks follow
test in the m
on number o
cos(150-q))
in the effect
cross test ses
of peaks is
d after traini
wing training
mixed regress
of peak in the
) in each te
t of angle on
ssions (resul
close to 1 f
ing shows a
g. C: Regre
sion model s
e stroke grou
est in the m
n number of
lts for
for all
long-
ession
shows
up. D:
mixed
f peak
67
4.3.5. Score change in box and block test
From the baseline to the 1-day and the 1-month retention tests, the stroke group
showed improvements in the BBT. At both the 1-day and 1-month retention tests, the
number of blocks moved with the affected limb increased 23% from baseline (paired t-
test, p = 0.004 and p = 0.004, respectively. Baseline: 21.3 ± 1.9; 1-day 26.3 ± 2.5; and 1-
month 26.3 ± 2.4 blocks). In addition, the change in BBT between baseline and 1 month
significantly correlated with the change in MT between Pre-1 and 1-month (
2
R = 0.56, p
= 0.001; changes between baseline and 1 day;
2
R = 0.025, p = 0.061). Note that for the
BBT, the minimum detectable change is 18% (Chen, Chen, Hsueh, Huang, & Hsieh,
2009). Ten out of fifteen participants exceeded this threshold between baseline and 1-day,
and between baseline and 1 month.
4.3.6. Relationship between initial performance and change in performance
The degree to which participants improved in MT and number of peaks directly
related to performance at the beginning of training. The mean initial MT showed a
significant linear association with mean change in MT (i.e., ΔMT) between Pre1 and 1-
day (
2
R = 0.77, p < 0.001) and between Pre1 and 1-month (
2
R = 0.68, p < 0.001) (Figure
4.3.4(A)). The initial number of peaks showed a significant linear association with mean
change in number of peaks (i.e., Δpeaks) between Pre1 and 1-day (
2
R = 0.80, p < 0.001)
and between Pre1 and 1-month (
2
R = 0.84, p < 0.001) (Figure 4.3.4(B)).
F
b
si
in
4
re
an
ch
d
in
at
im
th
(2
igure 4.3.4.
etween Pre1
ignificant be
nitial numbe
.4. Discussio
The mai
each training
nd moveme
hronic durat
ay retention
n the velocit
t the 1-mo
mpairments,
he Pre1-test
23% on ave
. Relationsh
1-test and 1
etween initia
r of peaks an
on
in result of
g with visuo
ent smoothn
tion. Compa
test and 20.
ty profile de
nth retentio
i.e., particip
showed the
erage) on th
hip between
1-day retenti
al MT and Δ
nd Δpeaks.
this proof-o
o-auditory f
ess in indiv
ared to the P
4% at the 1-
creased on a
on test. Th
pants with th
e greatest ga
he BBT at o
68
n initial per
ion test in
ΔMT. B: Li
of-concept st
feedback can
viduals with
Pre1-test, MT
-month reten
average 22.8
e improvem
he slowest m
ains in both
one month,
rformance a
stroke grou
inear relatio
tudy is that
n significan
h mild to m
T decreased
ntion test. Si
8% at the 1-
ments were
movements o
h quantities
and the sig
and change
up. A: Linea
onship is sig
short-durati
ntly improve
moderate stro
d on average
imilarly, the
day retentio
proportiona
r greatest nu
in absolute
gnificant cor
in perform
ar relationsh
gnificant bet
ion and inte
e movement
oke severity
e 22.8% at t
number of p
on test and 2
al to the i
umber of pea
terms. The
rrelation bet
mance
hip is
tween
ensive
time
y and
the 1-
peaks
22.7%
initial
aks at
gain
tween
69
changes in MT and changes in BBT at one month, suggest that the effect of task-specific
arm reach training transfers to untrained motor tasks and is lasting.
In both groups, movement times were longest for far targets to the left and shortest for
close targets to the right. The increase of movement time with increasing target distance
was formally described by Fitts (1954) now known as Fitts’s Law (Fitts, 1954). Pre-
training results in the stroke group (Figure 3D, Pre1-test) are in line with a previous study
showing that Fitts’s slope and intercept were greater for the affected arm compared to the
unaffected arm post-stroke (McCrea & Eng, 2005). In addition, the post-training results
show that training brings the Fitts’s slope closer (Figure 4.3.2(D)) to that of non-disabled
participants (Figure 4.3.2(D), right), possibly because more force can be generated post-
training. Increased movement times in the direction of the greatest inertia in planar
reaching movements (i.e., for movements near the direction of left-most training target)
have been previously reported in non-disabled participants (Gordon et al., 1994), and
have been explained by a “constant effort” strategy (Guigon et al., 2007). Studies have
found that participants post-stroke have difficulties controlling movements to leftward
targets because of a reduced ability to account for interaction torques (Beer et al., 2000).
Thus, although movement times to the left were longer than movement times to the right
in both groups (as a reminder, MT and smoothness data from left hemiparetic subject
were flipped from the left to right), participants in the stroke group showed greater
modulation of MT as a function of target angle. Training, however, reduced this angular
influence on movement time (Figure 4.3.2(E)) and number of peaks (Figure 4.3.3(D)).
Movements became smoother after training, notably in movements to leftward targets,
70
possibly because of a decrease in weakness and improvement in compensating the
interaction torques for these movements.
When instructed to move quickly, individuals post-stroke generated reaching
movements that are both faster and smoother than those produced when instructed to
move at a preferred speed (DeJong et al., 2012). The findings suggest that training based
on movement time induces a decrease in movement time and number of peaks that is
simply context-dependent, with no actual improvements. The training instructions
emphasize moving as quickly as possible, and not at a preferred speed from the start of
training. In addition, the scores of the BBT increased significantly overall after training
and the improvements were sustained 1-month later.
The main limitation of our feasibility study is the within-sample design with no
stroke control group. Nonetheless, training seems to induce robust changes in patients
with mild to moderate impairment, as shown by the strong correlation (
2
R > 0.77)
between the initial MT and the gain in MT for instance (see Figure 4.3.4(A)). In addition,
there is little possibility that the improvements in performance between initial
performance and the 1-day and 1-month retention tests after training was due to
spontaneous recovery, because the participants were in chronic stage (i.e., the minimum
time post-stroke was 12 months, with an average of 72±11 months). We note that there
was no correlation between duration since stroke and change in the MT (
2
R =0.03,
p=0.50 at 1-month) or number of peaks (
2
R =0.007, p=0.76 at 1-month). There is
however the possibility that the tests themselves lead to the improvements that we
observed. This limitation should be addressed in a future Phase 1 RCT. A second
limitation is that, based on the observation of left-right symmetry in Pre1-test MTs, we
71
did not differentiate between right and left hemiparesis. A future larger study should
include either left or right affected individuals, or better both, to account for differences
in control and learning between individuals with left and right brain damage (Mani et al.,
2013; Mani, Przybyla, Good, Haaland, & Sainburg, 2014). A third limitation is that we
did not collect data about the lesion (i.e., ischemic or hemorrhagic location, etc), nor did
we collect data about depression and fatigue that are associated with quality of life (van
de Port, Kwakkel, van Wijk, & Lindeman, 2006) and functional outcome (de Groot,
Phillips, & Eskes, 2003), respectively.
In summary, our results suggest that two sessions of unassisted intensive reach
training can induce long-term performance changes in patients with mild to moderate
impairments, who can perform reaching movements without gravity support. Although it
is probably infeasible to increase the number of movements per day much beyond 600
movements, the number of training sessions can be increased. Still unknown is the most
effective amount of training, since more is not necessarily better (Dromerick et al., 2009).
Acknowledgements
This work was supported in part by NIH grant R01 HD065438 and MC-IIF 299687.
A
T
Appendix B.
Table B1. AR
RT experime ental protoco
72
ol for 5 visits s in the strok ke group.
73
CHAPTER 5: Prediction of long-term gains due to intensive reach training in
individuals with chronic stroke
5.1. Introductions
The number of individuals who had a stroke leading long-term motor disability is
increasing in the United States (Mozaffarian et al., 2015). The reason is because
successful medical technology development has led to a decrease in mortality (Lackland
et al., 2014). Moreover, about 65% of stroke survivors
experience long-term limitations
in upper extremity (UE) functions (Dobkin, 2005; Lum et al., 2009). Since numerous
activities involve the UEs (Coster et al., 2004), even mild impairments post-stroke impact
quality of life (Carr & Shepherd, 2003) and participation (Duncan et al., 1999). Task-
specific training over multiple sessions can improve UE function post-stroke (Wolf et al.,
2006). In particular, motor training improves reaching movements’ speed, smoothness,
and range, decreases duration, and improves clinical scores (Cirstea, Mitnitski, et al.,
2003; Kamper et al., 2002; McCrea & Eng, 2005; van Dokkum et al., 2013). In addition,
tens of thousands of movements are needed for clinically significant improvement in
74
functions (Nudo et al., 1996; Pavlides, Miyashita, & Asanuma, 1993; Wolf et al., 2006).
Therefore, the need of ideal number of rehabilitation sessions with therapists is
increasing.
As the given recent emphasis on reducing healthcare costs, therapy with semi-
autonomous rehabilitation system providing intensive training with kinematic analysis
has been investigated. However, to maximize the cost-effectiveness and regain the loss of
motor function faster and greater, predictive methods by individualizing the dose and
schedule of UE motor therapy are needed. Then the predictive model based on the
premise that long term recovery occurs in response to motor learning determine a
personalized training schedule with an adaptive dose rather than with a fixed dose based
on present performance.
The initial performance changes are mostly due to “spontaneous recovery”, which is
maximally expressed in the first 4 weeks, but continues until about 6 months (Duncan,
Goldstein, Matchar, Divine, & Feussner, 1992; Kwakkel, Kollen, & Krakauer).
Therefore, several studies investigated the statistical methods to predict the long-term UE
performance using kinematic, kinetic, and/or clinical variables during the spontaneous
recovery period (Krebs et al., 2014; Lo et al., 2010; Mirbagheri, Tsao, & Rymer, 2008).
For example, kinematic measures with MIT-Manus (Lo et al., 2010) and clinical scores in
208 patients improved prediction of recovery at five time points (between 7 days and 3
month after stroke onset) (Krebs et al., 2014). In addition, kinematic, kinetic, and clinical
measures were tracked to investigate voluntary elbow movement in 20 participants at five
time points (between 1 and 12 month after stroke onset) (Mirbagheri et al., 2008).
Although Rohrer et al. investigated the recovery process due to multiple therapy sessions
75
in two groups: 15 inpatients (less than 1 month post-stroke) and 26 outpatients (greater
than 12 months post-stroke) (Rohrer et al., 2002), the statistical model predicting the
long-term behavioral performance change due to short-duration and intensive training in
individual with chronic stroke has not yet been fully investigated.
The performance during training is influenced by short-term memory, attention,
motivation, and etc (Cahill et al., 2001). The change in performance during motor
training occurs at a variety of time scales, as short as 10s of seconds (Kording,
Tenenbaum, & Shadmehr, 2007; Smith et al., 2006). If the task difficulty is based on only
the performance during training, it is uncertain whether the shown performance during
training is maintained over time. Therefore, the well-known learning-performance
distinction (Cahill et al., 2001; C. Winstein, Wing, & Whitall, 2003) illustrates this
variability in time scales and is seen for instance when patients can perform a motor task
in the clinic, but not after going back home. Moreover fatigue affected by performance
intense/duration is a complex area physically and psychologically. Therefore, the
predictive model of recovery is in this premise, distinct between learning and
performance with fatigue effect.
Chapter 5 introduces a nonlinear statistical model with mixed effect to predict
individualized long-term performance. This chapter examines the model’s ability to
determine if the amplitude of the learning curve and the performance at the end of
training can predict long-term performance, and whether the control group can learn
more quickly than the stroke group. The next sections describe the methodology.
76
5.2. Methods
5.2.1. Participants
Sixteen stroke patients with ischemic or hemorrhagic stroke with mild to moderate
impairments (63.2 ± 2.7 years; 14 males, 2 females), referred to as “stroke group”, ten
age-matched non-disabled individuals (56.6 ± 2.9 years; 5 males, 5 females), referred to
as “age-matched control group”, and nine young non-disabled individuals (24.4 ± 1.1
years; 4 males, 5 females), referred to as “young control group”, participated in the proof-
of-concept study.
Table 3.2.1 and Table 4.2.1 summarize the demographic information for the three
groups. This study was approved by the University of Southern California’s Institutional
Review Board, and all participants read and signed an informed consent prior to study
enrollment.
5.2.2. Clinical assessments
See Chapter 4.2.3.
5.2.3. Study design
See Figure 3.2.1 for the control group and Figure 4.2.1 for the stroke group.
77
5.2.4. Data analysis
MATLAB (The MathWorks, Natick, MA) was used to process the recorded data.
The sensor position data were low-pass filtered with a cutoff frequency of 5 Hz using a
zero phase-lag second-order Butterworth filter. The MT for offline data analysis was
computed by the interval between the time at which the tangential velocity of the index
fingertip first exceeded 5% of maximum velocity (Coderre et al., 2010) and the time at
which the velocity fell below 5% of maximum velocity with the fingertip inside of the
target. The model parameters for each participant with his/her training data were
estimated using the MATLAB function nlmefitsa.
5.2.5. Nonlinear statistical model with mixed effect
The data of all participants in the three groups were considered. Since MT is a prime
candidate and a good indicator of motor skill learning and change of performance during
training, a nonlinear statistical model with mixed effect based on MT was formulated as:
,, , , ,
j
i
t
tau
ijk i i j ik i jk
MT A e C t D ε
(1)
where
ijk
MT
is the movement time, for subject number i (i=1~35), trial number
j
t
(j=1~120) and target number k (k=1~5),
i
A is the mixed effect amplitude parameter,
78
i
tau is the mixed effect rate parameter,
i
C is the “fatigue” parameter including meaning
of both physical and psychological fatigue as the trial number increases,
ik
D is the
intercept for each training target, and a
ijk
ε
is a noise. Equation (1) has three terms:
j
i
t
tau
i
Ae
referred to as exponential decay,
ij
Ct referred to as “fatigue”, and
, ik
D
referred to as intercept. In order to validate whether we predict long-term gain or not
using the model parameters,
, iik
A D is defined as the “initial performance”,
120
i
tau
ii
AA e
is defined as the “Amplitude of learning curve” and
120
,
120
i
tau
ii ik
Ae C D
is defined as the “Performance at the end of training”. In
Appendix C gives the details.
To compare the data-driven model with or without the fatigue term, the equation (2)
including no fatigue term was formulated as:
,, , , ,
j
i
t
tau
ijk i ik ijk
MT A e D ε
(2)
We also tested the linear relationship between the model parameters (i.e., from the
equation (1) and (2)) and demographic information parameters (i.e., age, stroke duration,
initial UE Fugl-Meyer scores, Wechsler’s memory scores, initial MT (before training),
and side affected). The level of statistical significance was set at p < 0.05. All results
were reported with the mean ± standard errors (SE) of the corresponding mean.
79
5.3. Results
5.3.1. Demographic information
Average stroke duration was 72 ± 11 months and the average FM score was 48.7 ±
2.5 (Table 4.2.1). The age-matched control and stroke groups did not differ in age (t-test,
p = 0.117). The young and age-matched control groups significantly differed in age (t-test,
p < 0.0001). All participants completed 600 trials of the two-day training session and
successfully reached all targets within 5 seconds. Performing the 600 movements in each
Training block lasted on average 106.0 ± 2.9 minutes (range: 63.9 to 143.5 minutes) in
the stroke group, 90.3 ± 3.8 minutes (range: 64.7 to 117.7 minutes) in the age-matched
control group, and 77.2 ± 2.9 minutes (range: 61.9 to 110.6 minutes) in the young group
including break times.
5.3.2. Nonlinear statistical model with mixed effect in movement time
Figure 5.3.1 shows the individual MT during training in the young control group.
The individual fitted curves obtained from the nonlinear statistical model were
significantly correlated with the individual MT data at Target 1 (
2
R = 0.64, p < 0.00001)
and at Target 5 (
2
R = 0.57, p < 0.00001). Figure 5.3.2 shows the individual MT during
training in the age-matched control group. The individual fitted curves obtained from the
nonlinear statistical model were significantly correlated with the individual MT data at
Target 1 (
2
R = 0.31, p < 0.00001) and at Target 5 (
2
R = 0.42, p < 0.00001). Figure 5.3.3
80
shows the individual MT during training in the stroke group. The individual fitted curves
obtained from the nonlinear statistical model were significantly correlated with the
individual MT data at Target 1 (
2
R = 0.47, p < 0.00001) and at Target 5 (
2
R = 0.49, p <
0.00001). See Appendix C for the details.
81
82
Figure 5.3.1. Individual MT at Target 5 in the young control group (n = 9). Red, green,
cyan, blue, and purple represent individual trial-by-trial MT. Black solid lines represent
the model fitting based on each participant’s movement time; widely spaced dash lines
represent the single exponential fitting; narrowly spaced dash lines represent the amount
of fatigue using the equation (1). The dashed vertical lines separate the four visit days
(indicated with filled symbols: circle for Pre-test, square for Post-test, triangle for 1-day
retention test, and diamond for 1-month retention test).
83
84
Figure 5.3.2. Individual MT at Target 5 in the age-matched control group (n = 10). Red,
green, cyan, blue, and purple represent individual trial-by-trial MT. Black solid lines
represent the model fitting based on each participant’s movement time; widely spaced
dash lines represent the single exponential fitting; narrowly spaced dash lines represent
the amount of fatigue using the equation (1). The dashed vertical lines separate the four
visit days (indicated with filled symbols: circle for Pre-test, square for Post-test, triangle
for 1-day retention test, and diamond for 1-month retention test).
85
F
b
m
re
o
am
tr
igure 5.3.3.
lue, and pur
model fitting
epresent the
f fatigue usin
mong five v
riangle for 1-
Individual M
rple represe
g based on
single expo
ng the equat
visits (indica
-day retentio
MT at Targe
ent individua
each partic
onential fittin
tion (1). The
ated with fill
on test, and d
86
et 5 in the s
al trial-by-tr
ipant’s mov
ng; narrowly
e dashed vert
led symbols
diamond for
stroke group
rial MT. Bla
vement time
y spaced das
tical lines se
: circle for P
1-month ret
p (n = 16). R
ack solid lin
e; widely sp
sh lines repr
eparate the la
Pre-test, squ
tention test).
Red, green,
nes represen
paced dash
resent the am
ast four visit
uare for Post
.
cyan,
nt the
lines
mount
t days
t-test,
5
in
p
W
du
M
tr
0
A
(p
re
w
.3.3. Relatio
nformation
There w
arameters (i
Wechsler’s m
uration (only
MT at Pre1-
raining targe
.05) (Table
,, A C and D
p > 0.05)
elationship w
with
C
and D
onship bet
parameters
were no st
i.e., , A tau
memory scor
y stroke gro
-test) showe
et number,
k
5.3.1). In t
k
D (training t
(Table 5.3.
with A and
3
D (p < 0.01
tween indi
s
tatistically s
,, u CD ) an
res (digit, fig
oup)) in all 3
ed a signific
k
= 2~5) (p
the stroke g
target numbe
1). The UE
k
D (i.e., trai
1), with the e
87
ividual mo
significant
nd demogra
gural), side
35 participan
cant relation
< 0.05), wit
group, Pre1
er,
k
= 1~5
E Fugl-Mey
ining target
exception of
odel param
correlation
aphic inform
affected (on
nts. In the co
nship with A
th the except
showed a
5) (p < 0.00
yer (FM) s
number,
k
f tau (p > 0.0
meters and
between in
mation param
nly stroke gr
ontrol group
A (p < 0.0
tion of tau,
significant r
1), with the
cores show
= 1, 2, 4, 5
05) (Table 5
d demogra
ndividual m
meters (i.e.,
roup), and s
p, only Pre1
01) and
k
D
C
, and
1
D
relationship
exception o
wed a signif
5) (p < 0.01)
.3.1).
aphic
model
age,
stroke
(i.e.,
(i.e.,
(p >
with
of tau
ficant
), and
88
Table 5.3.1. Correlation between individual model parameters and Pre1 (MT before
training) in the control and stroke groups and the UE FM scores in the stroke group. * p <
0.05, ** p < 0.01, *** p < 0.001.
5.3.4. Prediction of long-term (1-month) performance
The results of the equation (1) in the control group showed a significant linear
relationship between the ratio of “amplitude of the learning curve over initial
performance” (i.e.,
120
100
i
tau
ii
k
AA e
AD
) and the ratio of “MT between the Pre1-test and
the 1-month retention test over Pre1” (i.e.,
Pr 1 1
100
Pr 1
e month
e
) (
2
R = 0.59, p = 0.0001)
(e.g., Figure 5.3.4(A, Left) shows the results at Target 5). There was a significant linear
relationship between the ratio of “Performance at the end of training over initial
performance” (i.e.,
120
( 120)
100
i
tau
ii
k
AA e C
AD
) and the ratio of “MT between the
Pre1-test and the 1-month retention test over Pre1” (
2
R = 0.56, p = 0.0002) (e.g., Figure
5.3.4(A, Right) shows the results at Target 5). The results of the equation (2) in the
control group showed a significant linear relationship between the ratio of “amplitude of
the learning curve over initial performance” and the ratio of “MT between the Pre1-test
and the 1-month retention test over Pre1” (
2
R = 0.60, p = 0.0001) (e.g., Figure 5.3.4(B)
shows the results at Target 5).
F
F
re
“a
th
b
ra
li
re
p
igure 5.3.4.
irst row: (A
esults of equ
amplitude of
he Pre1-test
etween the r
atio of “MT
ine represent
The resu
elationship
erformance”
Prediction o
A) as a resul
uation (2) wi
f the learnin
t and the 1
ratio of “Perf
between the
ts least-squar
ults of the
between th
” and the rat
of 1-month p
lt of the equ
ithout fatigu
ng curve ove
-month rete
formance at
e Pre1-test a
res fit to ind
equation (1
he ratio of
tio of “MT b
89
performance
uation (1) w
ue term. Left
er initial perf
ention test
the end of tr
and the 1-mo
dividual data
) in the str
f “amplitude
between the
e at Target 5
with fatigue
t panel: Rela
formance” a
over Pre1”
training over
onth retentio
a.
roke group
e of the l
Pre1-test an
5 in the contr
term. Secon
ationship bet
and the ratio
. Right pan
r initial perfo
on test over
showed a s
learning cu
nd the 1-mo
rol group (n
nd row: (B)
tween the rat
of “MT bet
nel: Relation
ormance” an
Pre1”. Red
significant l
urve over i
onth retention
n=19).
) as a
tio of
tween
nship
nd the
solid
linear
initial
n test
ov
T
o
1
th
si
in
re
T
F
F
ver Pre1” ( R
There was no
f training ov
-month reten
he results at
ignificant lin
nitial perfor
etention test
Target 5).
igure 5.3.5.
irst row: (A
2
R = 0.35, p
o significant
ver initial pe
ntion test ov
t Target 5).
near relation
rmance” and
t over Pre1”
Prediction o
A) as a resul
p = 0.0166) (
linear relati
erformance”
ver Pre1” ( R
The results
nship betwee
d the ratio
(
2
R = 0.29
of 1-month p
lt of the equ
90
(Figure 5.3.5
ionship betw
and the rati
2
R = 0.18, p
s of the equ
en the ratio o
of “MT be
9, p = 0.031
performance
uation (1) w
5(A, Left) sh
ween the ratio
io of “MT b
= 0.1064) (F
uation (2) in
of “amplitud
etween the
5) (Figure 5
e at Target 5
with fatigue
hows the res
o of “Perfor
between the
Figure 5.3.5
n the stroke
de of the lea
Pre1-test a
5.3.5(B) sho
5 in the stro
term. Secon
sults at Targ
rmance at th
Pre1-test an
(A, Right) s
group show
arning curve
and the 1-m
ows the resu
ke group (n
nd row: (B)
get 5).
e end
nd the
shows
wed a
e over
month
ults at
n=16).
) as a
91
results of the equation (2) without fatigue term. Left panel: Relationship between the ratio
of “amplitude of the learning curve over initial performance” and the ratio of “MT
between the Pre1-test and the 1-month retention test over Pre1”. Right panel:
Relationship between the ratio of “Performance at the end of training over initial
performance” and the ratio of “MT between the Pre1-test and the 1-month retention test
over Pre1”. Red solid line represents least-squares fit to individual data.
In both the control and stroke groups, the
120
,
100
i
tau
ii
iik
AA e
AD
, ratio of “amplitude of
the learning curve over initial performance”, and the
120
,
( 120)
i
tau
ii i
iik
AA e C
AD
, ratio of
“Performance at the end of training over initial performance”, showed a significant linear
relationship with the change in the MT between the Pre1-test and the 1-month retention
test (p < 0.05).
Figure 5.3.6 showed the two examples for comparison of two models with (i.e.,
equation (1)) and without fatigue term (i.e., equation (2)) in the stroke group. In the left
column of Figure 5.3.6, S1 and S15 in the stroke group showed a decreased exponential
curve and increased fatigue amount, whereas in the right column S1 and S15 showed zero
change of fatigue amount, as well as small and zero change of exponential curve,
respectively in the MT during training.
F
te
(1
m
si
5
b
le
st
igure 5.3.6.
erm effect at
1). Right pa
model fitting
ingle expone
.3.5. Learni
The one
etween the
earning deca
troke group
Comparison
t Target 5 in
anel: Fitting
g based on e
ential fitting;
ing rate in t
-way analys
control gro
ay rates, tau
(n = 16) com
n of two exam
n the stroke
results usin
each particip
; narrowly sp
the control a
is of varianc
up (n = 19
, showed a s
mpared to the
92
mples (i.e., S
group. Left
ng the equat
pant’s MT;
paced dash l
and stroke g
ce (ANOVA
) and stroke
significantly
e control gro
S1 and S15)
panel: Fittin
tion (2). Bla
widely spac
lines represe
groups
A) was used t
e group. Th
y higher log
oup (n = 19)
) with and w
ng results us
ack solid lin
ced dash lin
ent the amou
to compare t
he log trans
transformat
(p = 0.005)
without the fa
sing the equ
nes represen
nes represen
unt of fatigue
the learning
formation o
tion of tau o
.
atigue
uation
nt the
nt the
e.
rates
of the
of the
93
5.4. Discussion
With a data-driven approach, a novel nonlinear statistical model with mixed effect
were developed based on MTs as results of short-duration and intensive reach training in
19 non-disabled individuals and 16 individuals with chronic stroke. This chapter
demonstrated that the individual fitted curves obtained from the model with fatigue term
(i.e., equation (1)) significantly correlate with individual MTs during training in all
participants (p < 0. 00001) (Figure 5.3.1, 5.3.2, and 5.3.3). The learning decay rates, tau,
in the model show that the taus of the control group are significantly higher than the taus
of the stroke group. Based on these findings, we infer that the control group learned more
quickly. It also possible to reach performance in the stroke group, which in turn results in
less accurate movement, less smoothness, etc.
While the control group shows that the Pre1 (i.e., MT in Pre1-test) has a
significantly linear relationship with A and
234 5
,, , D DD D , the stroke group shows that
the Pre1 and FM scores have a significantly linear relationship with , A C and D (Table
5.3.1). In both groups, tau has no statistically significant relationship with Pre1 and FM.
Unlike the stroke group, the fatigue coefficient, C , in the control group is not significant
with Pre1 and FM (Table 5.3.1).
Since the fatigue term has as small effect in control group, both the ratio of
“amplitude of the learning curve over initial performance” and the ratio of “Performance
at the end of training over initial performance” have significant relationships with the
ratio of “MT between the Pre1-test and the 1-month retention test over Pre1” (Figure
5.3.4(A)). The results suggest that the model parameters obtained after the first day of
94
intensive reach training in the control group can provide reliable prediction of long-term
performance.
We also find that, unlike the control group, the fatigue term in the stroke group
affects the prediction of long-term performance. Figure 5.3.5(A, Left) shows that the ratio
of “amplitude of the learning curve over initial performance” using the equation (1) (i.e.,
with the fatigue term) significantly predicts the long-term performance (
2
R = 0.35, p =
0.0166). Figure 5.3.5(B) shows that the ratio of “amplitude of the learning curve over
initial performance” using the equation (2) (i.e., without the fatigue term) significantly
predicts long-term performance (
2
R = 0.29, p = 0.0315). Figure 5.3.4(A, Right),
however, shows that the ratio of “Performance at the end of training over initial
performance”, does not show a significant relationship with the change of the long-term
(1-month) performance (
2
R = 0.18, p = 0.1064). We interpret that while the “Amplitude
of the learning curve” in the performance using both models (equation (1) and (2))
predicts with long-term performance, the “Performance at the end of training” does not
predict long-term performance. For example, Figure 5.3.6 (right panel) showed that S15
showed zero change of exponential curve, i.e., S15 never learned during the intensive
reach training. However, based on the results of Figure 5.3.5, we can infer that S15
having a decreased exponential curve with increased fatigue amount during training in
Figure 5.3.6 (left panel) learned how to move quickly during the intensive reach training.
We suggest that the equation (1) can provide better long-term prediction than the
equation (2) in the stroke group. For example, S1 in Figure 5.3.5(A, Left) shows around
10% difference between the real data and the prediction with the model. And S1 in Figure
5.3.5(B) shows increased error (more than 10%) between the real data and the prediction
95
with the model. Moreover, the “Amplitude of learning curve” of the equation (1)
significantly predicts long-term performance change with higher
2
R unlike the equation
(2) without the fatigue term.
Other variables may also be good predictors of long-term gain. Thus, we tested the
initial number of peaks as a dependent variable in the model. As a result, the initial
number of peaks (13 out of 16 participants in the stroke group) does not exceed 4 and do
not fit using the model including exponential decay, fatigue, and intercept terms (i.e.,
equation (1)). However, it is still unknown what variable is most effective. In addition, to
examine covariates in stroke group included baseline impairment levels as measured by
UE FM test, baseline reaching performance, age, stroke onset, Wechsler memory scores.
However extended nonlinear statistical model can include detailed joint kinematics of
UE, MRI and fMRI-derived neural covariates, such as lesion location, volume,
percentage of intact corticospinal tract, levels of ipsi- and contra-lesional activity,
laterality indices, and intra- and inter-connectivity measures of resting state networks will
improve predictions of long-term behavioral recovery. This approach lead to identify
patient-specific covariates that can explain the inter-patient variability typically observed
in the treatment response of these patients. For instance, predictions from initial UE FM
are improved when motor cortical activity measured via fMRI is added in regression
models (Cramer et al., 2007; Zarahn et al., 2011).
In conclusion, the present results suggest that the nonlinear statistical model with
mixed effect support to predict the long-term performance changes in individuals with
mild to moderate impairments who performed two sessions of unassisted intensive reach
training. The model will be applicable to provide an efficiently individualized training
96
schedule based on a clinical software that suggests timing, dosage, and content of therapy
from early clinical data, kinematic performance, and routine scans. Such an approach will
transform neuro-rehabilitation programs because clinicians, patients, and insurance
companies will be able to determine effective treatments while reducing costs. In
addition, the nonlinear statistical model with mixed effect can be extensively adapted to
other neurological disorders that lead to motor impairments, such as traumatic brain
injury or Parkinson’s disease. Finally, this work will provide future neuroscientists,
clinicians, and engineers’ strategies to develop effective therapeutic strategies grounded
in 1) an understanding of the relationship between mathematical components in
performance and learning, 2) motor learning mechanism, and 3) state of the art statistical
learning methods.
Acknowledgements
This work was supported in part by NSF BCS 1031899, NIH R01 HD065438, and MC-
IIF 299687.
A
Appendix C
97
98
Figure C1. Individual MT at Target 1 in the young control group (n = 9). Red, green,
cyan, blue, and purple represent individual trial-by-trial MT. Black solid lines represent
the model fitting based on each participant’s movement time; widely spaced dash lines
represent the single exponential fitting; narrowly spaced dash lines represent the amount
of fatigue using the equation (1). The dashed vertical lines separate the four visit days
(indicated with filled symbols: circle for Pre-test, square for Post-test, triangle for 1-day
retention test, and diamond for 1-month retention test).
F
cy
igure C2. In
yan, blue, an
ndividual MT
nd purple re
T at Target 1
epresent indi
99
1 in age-mat
ividual trial-
tched contro
-by-trial MT
ol group (n =
T. Black soli
= 10). Red, g
id lines repr
green,
resent
100
the model fitting based on each participant’s movement time; widely spaced dash lines
represent the single exponential fitting; narrowly spaced dash lines represent the amount
of fatigue using the equation (1). The dashed vertical lines separate the four visit days
(indicated with filled symbols: circle for Pre-test, square for Post-test, triangle for 1-day
retention test, and diamond for 1-month retention test).
101
F
an
fi
si
u
fi
fo
igure C3. In
nd purple re
itting based o
ingle expon
sing the equ
ive visits (in
or 1-day rete
ndividual MT
epresent ind
on each part
ential fitting
uation (1). T
ndicated with
ention test, a
T at Target 1
ividual trial
ticipant’s mo
g; narrowly
The dashed v
h filled symb
and diamond
102
1 in the strok
-by-trial MT
ovement tim
spaced dash
vertical lines
bols: circle f
for 1-month
ke group (n =
T. Black sol
me; widely sp
h lines repr
s separate th
for Pre-test,
h retention te
= 16). Red, g
lid lines rep
paced dash li
resent the am
he last four v
square for P
est).
green, cyan,
present the m
ines represen
mount of fa
visit days am
Post-test, tri
blue,
model
nt the
atigue
mong
angle
F
w
A
as
“A
fi
re
am
igure C4. A
with fatigue t
10 10,5
A D is
s the “Perf
Amplitude o
itting based
epresents the
mount of fat
An example
term (i.e., e
defined as t
formance at
of learning c
on each p
e single exp
tigue using th
of a fitted c
quation (1))
he “initial p
the end o
curve” each
participant’s
ponential fitt
he equation
103
curve obtain
) at Target 5
performance”
f training”;
trained targe
movement
ting; narrow
(1).
ned from th
5 of subject
”;
10
120
10
tau
Ae
and
10
AA
et. Black so
time; wide
wly spaced da
e nonlinear
10 in the st
0
0
10
120 C
10
120
10
tau
A e
is
olid line repr
ely spaced d
ash green li
statistical m
troke group
10,5
D is de
s defined a
resents the m
dash purple
ne represent
model
. The
efined
s the
model
e line
ts the
F
gr
ch
b
te
le
eq
igure C5. C
roup. A: Le
hange in MT
etween “Per
est and the
earning curv
quation (2).
Comparison
eft: Relation
T between th
rformance at
1-month ret
ve” and chan
between th
nship betwe
he Pre1-test
t the end of
tention test.
ge in MT be
104
e equation
een change
t and the 1-m
f training” an
B: Relation
etween Pre1-
(1) and the
in “Amplitu
month retent
nd change in
nship betwee
-test and 1-m
equation (2
tude of lear
tion test. Rig
n the MT b
en change i
month retent
2) in the co
rning curve”
ght: Relation
etween the
in “Amplitud
tion test usin
ontrol
” and
nship
Pre1-
de of
ng the
F
A
M
“P
1
an
igure C6. Co
A: Left: Rela
MT between
Performance
-month reten
nd change in
omparison b
ationship be
the Pre1-te
e at the end
ntion test. B
n MT betwee
between the e
tween chang
st and the 1
of training”
: Relationsh
en Pre1-test
105
equation (1)
ge in “Amp
1-month rete
and change
hip between c
and 1-month
) and the equ
plitude of lea
ention test.
in the MT b
change in “A
h retention t
uation (2) in
arning curve
Right: Relat
between the
Amplitude o
test using the
the stroke g
e” and chan
tionship bet
Pre1-test an
of learning cu
e equation (2
group.
nge in
tween
nd the
urve”
2).
106
CHAPTER 6: Conclusion and future work
This dissertation work focuses on the design and development of ART system for
stroke UE rehabilitation and assessment of short-duration intensive training effect with
the ART system in non-disabled and post-stroke individuals. In addition, this dissertation
describes how non-disabled individuals learn to move fast and the effect of intensive
reach training in individuals with chronic stroke scientifically and clinically. Furthermore,
the dissertation proposes a novel nonlinear statistical model with mixed effect to predict
long-term gains in performance due to training in individuals post-stroke. It is expected
that the prediction provides important considerations in optimal training strategy for UE
stroke rehabilitation.
The studies presented in Chapter 3 and 4 show that the short-duration and intensive
reach training with the ART system induces significant and durable improvement in
performance and generalization effect in non-disabled and post-stroke individuals.
Especially the stroke group also showed significant clinical improvement and transfer
effect. This is indeed evidence that the ART system might be appropriate for UE
rehabilitation in individuals with chronic stroke and mild to moderate impairments.
107
The ART system provides training including relatively simple reach movements
with 35 targets -- distributed in 20 degrees increments between 30° and 150° and 5 cm
increments from 10 cm to 30 cm away from the home position (Figure 2.2.3). Future
studies should include more targets on extended ART workspace and more functional
activities such as grasping with different size objects, writing, or other movements
including wrist and fingers. Moreover, the present ART system has the fixed diameter of
each target (i.e., 3 cm). Based on Fitts’ law, the ART system can adjust the task difficulty
by changing the target size based on a participant’s performance. In addition, the current
ART system used two magnetic sensors to capture the positions of the elbow and index
finger. We assumed that a seat belt fully restrain trunk movements. Future studies should
develop ART system with more magnetic sensors to capture multiple body segments (e.g.,
trunk, shoulder, wrist, fingers). For this, the ART system will need additional software
and hardware considerations. In addition, this dissertation work had limited target
population (i.e., individuals with chronic stroke and mild to moderate impairments).
Therefore, the future study should have the extended target population such as
individuals with severe to moderate impairments and/or with sub-acute. For including
individuals with severe to moderate impairments and/or with sub-acute, the ART system
will need to include passive arm support.
The prediction using nonlinear statistical model with mixed effect presented in
Chapter 5 can be used to develop better rehabilitation strategies by defining an specific
individualized training schedule that is optimally adjustable the doses of training, spacing,
and feedback frequency effects. Moreover, this approach will derive a better
understanding of the motor learning mechanism in each individual with stroke. Therefore
108
the future ART system might help individuals with stroke promote their motor
performance greater and faster.
The present ART system seems promising to be effectively used for stroke patients
to guide arm reach rehabilitation exercises. It is expected that the ART system can be
used at clinics and/or home and help individuals with stroke to improve a level of their
independence and to maintain their quality of life. This dissertation work might
contribute to provide information of the future systematic rehabilitation design that
supports clinical and home-based rehabilitation testing and training. In addition, this in-
depth analyses and understanding of reach training for stroke rehabilitation might offer
insight and inspiration of optimal training strategy and develop advanced treatment
modalities for UE therapy. Moreover, the thorough understanding of clinical and
scientific findings might promote the motor performance greater and faster in individuals
with stroke as well as in patients with various neuromuscular disorders, by extension, in
sports discipline.
109
REFERENCES
Aisen, M. L., Krebs, H. I., Hogan, N., McDowell, F., & Volpe, B. T. (1997). The effect
of robot-assisted therapy and rehabilitative training on motor recovery following
stroke. Archives of neurology, 54(4), 443-446.
Beer, R. F., Dewald, J. P., Dawson, M. L., & Rymer, W. Z. (2004). Target-dependent
differences between free and constrained arm movements in chronic hemiparesis.
Experimental Brain Research, 156(4), 458-470.
Beer, R. F., Dewald, J. P., & Rymer, W. Z. (2000). Deficits in the coordination of
multijoint arm movements in patients with hemiparesis: evidence for disturbed
control of limb dynamics. Experimental brain research, 131(3), 305-319.
Birkenmeier, R. L., Prager, E. M., & Lang, C. E. (2010). Translating animal doses of
task-specific training to people with chronic stroke in 1-hour therapy sessions: a
proof-of-concept study. Neurorehabilitation and neural repair.
Blennerhassett, J., & Dite, W. (2004). Additional task-related practice improves mobility
and upper limb function early after stroke: a randomised controlled trial.
Australian Journal of Physiotherapy, 50(4), 219-224.
110
Cahill, L., McGaugh, J. L., & Weinberger, N. M. (2001). The neurobiology of learning
and memory: some reminders to remember. Trends in neurosciences, 24(10), 578-
581.
Carel, C., Loubinoux, I., Boulanouar, K., Manelfe, C., Rascol, O., Celsis, P., & Chollet,
F. (2000). Neural Substrate for the Effects of Passive Training on Sensorimotor
Cortical Representation: A Study With Functional Magnetic Resonance
Imaging in Healthy Subjects. Journal of Cerebral Blood Flow & Metabolism,
20(3), 478-484.
Carr, J. H., & Shepherd, R. B. (2003). Stroke rehabilitation: guidelines for exercise and
training to optimize motor skill: Butterworth-Heinemann Medical.
Chen, H.-M., Chen, C. C., Hsueh, I.-P., Huang, S.-L., & Hsieh, C.-L. (2009). Test-retest
reproducibility and smallest real difference of 5 hand function tests in patients
with stroke. Neurorehabilitation and neural repair.
Cirstea, M., & Levin, M. (2007). Improvement of arm movement patterns and endpoint
control depends on type of feedback during practice in stroke survivors.
Neurorehabilitation and neural repair.
Cirstea, M., & Levin, M. F. (2000). Compensatory strategies for reaching in stroke.
Brain, 123(5), 940-953.
Cirstea, M., Mitnitski, A., Feldman, A., & Levin, M. (2003). Interjoint coordination
dynamics during reaching in stroke. Experimental Brain Research, 151(3), 289-
300.
111
Cirstea, M., Ptito, A., & Levin, M. (2003). Arm reaching improvements with short-term
practice depend on the severity of the motor deficit in stroke. Experimental brain
research, 152(4), 476-488.
Coderre, A. M., Zeid, A. A., Dukelow, S. P., Demmer, M. J., Moore, K. D., Demers, M.
J., . . . Norman, K. E. (2010). Assessment of upper-limb sensorimotor function of
subacute stroke patients using visually guided reaching. Neurorehabilitation and
neural repair, 24(6), 528-541.
Conner, J. M., Culberson, A., Packowski, C., Chiba, A. A., & Tuszynski, M. H. (2003).
Lesions of the basal forebrain cholinergic system impair task acquisition and
abolish cortical plasticity associated with motor skill learning. Neuron, 38(5),
819-829.
Coster, W. J., Haley, S. M., Andres, P. L., Ludlow, L. H., Bond, T. L., & Ni, P.-s. (2004).
Refining the conceptual basis for rehabilitation outcome measurement: personal
care and instrumental activities domain. Medical care, 42(1), I-62.
Cramer, S. C., Parrish, T. B., Levy, R. M., Stebbins, G. T., Ruland, S. D., Lowry, D. W., .
. . Savage, C. R. (2007). Predicting functional gains in a stroke trial. Stroke, 38(7),
2108-2114.
de Groot, M. H., Phillips, S. J., & Eskes, G. A. (2003). Fatigue associated with stroke and
other neurologic conditions: implications for stroke rehabilitation. Archives of
physical medicine and rehabilitation, 84(11), 1714-1720.
DeJong, S. L., Schaefer, S. Y., & Lang, C. E. (2012). Need for Speed Better Movement
Quality During Faster Task Performance After Stroke. Neurorehabilitation and
neural repair, 26(4), 362-373.
112
Demaerschalk, B. M., Hwang, H.-M., & Leung, G. (2010). US cost burden of ischemic
stroke: a systematic literature review. The American journal of managed care,
16(7), 525-533.
Dewald, J. P., Pope, P. S., Given, J. D., Buchanan, T. S., & Rymer, W. Z. (1995).
Abnormal muscle coactivation patterns during isometric torque generation at the
elbow and shoulder in hemiparetic subjects. Brain, 118(2), 495-510.
Dewald, J. P., Sheshadri, V., Dawson, M. L., & Beer, R. F. (2001). Upper-limb
discoordination in hemiparetic stroke: implications for neurorehabilitation. Topics
in stroke rehabilitation, 8(1), 1-12.
Dhamoon, M. S., Moon, Y. P., Paik, M. C., Boden-Albala, B., Rundek, T., Sacco, R. L.,
& Elkind, M. S. (2009). Long-term functional recovery after first ischemic stroke
the northern manhattan study. Stroke, 40(8), 2805-2811.
Dipietro, L., Krebs, H., Volpe, B., Stein, J., Bever, C., Mernoff, S., . . . Hogan, N. (2012).
Learning, not adaptation, characterizes stroke motor recovery: evidence from
kinematic changes induced by robot-assisted therapy in trained and untrained task
in the same workspace. Neural Systems and Rehabilitation Engineering, IEEE
Transactions on, 20(1), 48-57.
Dobkin, B. H. (2005). Rehabilitation after stroke. New England Journal of Medicine,
352(16), 1677-1684.
Dromerick, A., Lang, C., Birkenmeier, R., Wagner, J., Miller, J., Videen, T., . . .
Edwards, D. (2009). Very early constraint-induced movement during stroke
rehabilitation (VECTORS) A single-center RCT. Neurology, 73(3), 195-201.
113
Duff, M., Yinpeng, C., Attygalle, S., Herman, J., Sundaram, H., Gang, Q., . . . Rikakis, T.
(2010). An Adaptive Mixed Reality Training System for Stroke Rehabilitation.
Neural Systems and Rehabilitation Engineering, IEEE Transactions on, 18(5),
531-541. doi: 10.1109/TNSRE.2010.2055061
Duncan, P. W., Goldstein, L. B., Matchar, D., Divine, G. W., & Feussner, J. (1992).
Measurement of motor recovery after stroke. Outcome assessment and sample
size requirements. Stroke, 23(8), 1084-1089.
Duncan, P. W., Wallace, D., Lai, S. M., Johnson, D., Embretson, S., & Laster, L. J.
(1999). The Stroke Impact Scale Version 2.0 evaluation of reliability, validity,
and sensitivity to change. Stroke, 30(10), 2131-2140.
Ellis, M. D., Sukal-Moulton, T., & Dewald, J. P. (2009). Progressive shoulder abduction
loading is a crucial element of arm rehabilitation in chronic stroke.
Neurorehabilitation and neural repair, 23(8), 862-869.
Fitts, P. M. (1954). The information capacity of the human motor system in controlling
the amplitude of movement. Journal of experimental psychology, 47(6), 381.
Fitts, P. M., & Peterson, J. R. (1964). Information capacity of discrete motor responses.
Journal of experimental psychology, 67(2), 103.
Folstein, M. F., Folstein, S. E., & McHugh, P. R. (1975). “Mini-mental state”: a practical
method for grading the cognitive state of patients for the clinician. Journal of
psychiatric research, 12(3), 189-198.
Fugl-Meyer, A. R., Jääskö, L., Leyman, I., Olsson, S., & Steglind, S. (1974). The post-
stroke hemiplegic patient. 1. a method for evaluation of physical performance.
Scandinavian journal of rehabilitation medicine, 7(1), 13-31.
114
Fullerton, K., McSherry, D., & Stout, R. (1986). Albert's test: a neglected test of
perceptual neglect. The Lancet, 327(8478), 430-432.
Gordon, J., Ghilardi, M. F., Cooper, S. E., & Ghez, C. (1994). Accuracy of planar
reaching movements. Experimental Brain Research, 99(1), 112-130.
Guigon, E., Baraduc, P., & Desmurget, M. (2007). Computational motor control:
redundancy and invariance. Journal of neurophysiology, 97(1), 331-347.
Harris-Love, M. L., Morton, S. M., Perez, M. A., & Cohen, L. G. (2011). Mechanisms of
Short-Term Training-Induced Reaching Improvement in Severely Hemiparetic
Stroke Patients A TMS Study. Neurorehabilitation and neural repair, 25(5), 398-
411.
Heidenreich, P. A., Trogdon, J. G., Khavjou, O. A., Butler, J., Dracup, K., Ezekowitz, M.
D., . . . Khera, A. (2011). Forecasting the future of cardiovascular disease in the
United States a policy statement from the American heart association.
Circulation, 123(8), 933-944.
Hesse, S., Werner, C., Pohl, M., Rueckriem, S., Mehrholz, J., & Lingnau, M. (2005).
Computerized arm training improves the motor control of the severely affected
arm after stroke a single-blinded randomized trial in two centers. Stroke, 36(9),
1960-1966.
Hung, Y.-C., Kaminski, T., Fineman, J., Monroe, J., & Gentile, A. (2008). Learning a
multi-joint throwing task: a morphometric analysis of skill development.
Experimental brain research, 191(2), 197-208.
Jeffery, D. R., & Good, D. C. (1995). Rehabilitation of the stroke patient. Current
Opinion in Neurology, 8(1), 62-68.
115
Kamper, D. G., McKenna-Cole, A. N., Kahn, L. E., & Reinkensmeyer, D. J. (2002).
Alterations in reaching after stroke and their relation to movement direction and
impairment severity. Archives of physical medicine and rehabilitation, 83(5), 702-
707.
Karni, A. (1995). When practice makes perfect. The Lancet, 345(8946), 395.
Keetch, K. M., Schmidt, R. A., Lee, T. D., & Young, D. E. (2005). Especial skills: their
emergence with massive amounts of practice. Journal of experimental
psychology: human perception and performance, 31(5), 970.
Keith, R. A., & Cowell, K. S. (1987). Time use of stroke patients in three rehabilitation
hospitals. Social science & medicine, 24(6), 529-533.
Kording, K. P., Tenenbaum, J. B., & Shadmehr, R. (2007). The dynamics of memory as a
consequence of optimal adaptation to a changing body. Nature neuroscience,
10(6), 779-786.
Krakauer, J. W., & Mazzoni, P. (2011). Human sensorimotor learning: adaptation, skill,
and beyond. Current opinion in neurobiology, 21(4), 636-644.
Krebs, H. I., Hogan, N., Aisen, M. L., & Volpe, B. T. (1998). Robot-aided
neurorehabilitation. Rehabilitation Engineering, IEEE Transactions on, 6(1), 75-
87.
Krebs, H. I., Krams, M., Agrafiotis, D. K., DiBernardo, A., Chavez, J. C., Littman, G. S.,
. . . Rykman, A. (2014). Robotic measurement of arm movements after stroke
establishes biomarkers of motor recovery. Stroke, 45(1), 200-204.
Kwakkel, G., Kollen, B. J., & Krakauer, J. W. Predicting activities after stroke.
116
Kwakkel, G., Kollen, B. J., & Krebs, H. I. (2007). Effects of robot-assisted therapy on
upper limb recovery after stroke: a systematic review. Neurorehabilitation and
neural repair.
Lackland, D. T., Roccella, E. J., Deutsch, A. F., Fornage, M., George, M. G., Howard,
G., . . . Lisabeth, L. D. (2014). Factors influencing the decline in stroke mortality
a statement from the american heart association/american stroke association.
Stroke, 45(1), 315-353.
Lang, C. E., MacDonald, J. R., Reisman, D. S., Boyd, L., Kimberley, T. J., Schindler-
Ivens, S. M., . . . Scheets, P. L. (2009). Observation of amounts of movement
practice provided during stroke rehabilitation. Archives of physical medicine and
rehabilitation, 90(10), 1692-1698.
Levin, M. F. (1996). Interjoint coordination during pointing movements is disrupted in
spastic hemiparesis. Brain, 119(1), 281-293.
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. Experimental brain research, 143(2), 171-180.
Lincoln, N. B., Willis, D., Philips, S., Juby, L., & Berman, P. (1996). Comparison of
rehabilitation practice on hospital wards for stroke patients. Stroke, 27(1), 18-23.
Lo, A. C., Guarino, P. D., Richards, L. G., Haselkorn, J. K., Wittenberg, G. F., Federman,
D. G., . . . Volpe, B. T. (2010). Robot-assisted therapy for long-term upper-limb
impairment after stroke. New England Journal of Medicine, 362(19), 1772-1783.
Lotze, M., Braun, C., Birbaumer, N., Anders, S., & Cohen, L. G. (2003). Motor learning
elicited by voluntary drive. Brain, 126(4), 866-872.
117
Lum, P. S., Burgar, C. G., Van der Loos, M., Shor, P. C., Majmundar, M., & Yap, R.
(2006). MIME robotic device for upper-limb neurorehabilitation in subacute
stroke subjects: A follow-up study. Journal of rehabilitation research and
development, 43(5), 631.
Lum, P. S., Mulroy, S., Amdur, R. L., Requejo, P., Prilutsky, B. I., & Dromerick, A. W.
(2009). Gains in upper extremity function after stroke via recovery or
compensation: Potential differential effects on amount of real-world limb use.
Topics in stroke rehabilitation, 16(4), 237-253.
Mackey, F., Ada, L., Heard, R., & Adams, R. (1996). Stroke rehabilitation: are highly
structured units more conducive to physical activity than less structured units?
Archives of physical medicine and rehabilitation, 77(10), 1066-1070.
Mani, S., Mutha, P. K., Przybyla, A., Haaland, K. Y., Good, D. C., & Sainburg, R. L.
(2013). Contralesional motor deficits after unilateral stroke reflect hemisphere-
specific control mechanisms. Brain, aws283.
Mani, S., Przybyla, A., Good, D. C., Haaland, K. Y., & Sainburg, R. L. (2014).
Contralesional arm preference depends on hemisphere of damage and target
location in unilateral stroke patients. Neurorehabilitation and neural repair,
28(6), 584-593.
Martin, T., Keating, J., Goodkin, H., Bastian, A., & Thach, W. (1996). Throwing while
looking through prisms II. Specificity and storage of multiple gaze—throw
calibrations. Brain, 119(4), 1199-1211.
118
Mathiowetz, V., Volland, G., Kashman, N., & Weber, K. (1985). Adult norms for the
Box and Block Test of manual dexterity. American Journal of Occupational
Therapy, 39(6), 386-391.
McCrea, P. H., & Eng, J. J. (2005). Consequences of increased neuromotor noise for
reaching movements in persons with stroke. Experimental brain research, 162(1),
70-77.
Michaelsen, S. M., & Levin, M. F. (2004). Short-Term Effects of Practice With Trunk
Restraint on Reaching Movements in Patients With Chronic Stroke A Controlled
Trial. Stroke, 35(8), 1914-1919.
Mirbagheri, M. M., Tsao, C. C., & Rymer, W. Z. (2008). Changes of elbow kinematics
and kinetics during 1 year after stroke. Muscle & nerve, 37(3), 387-395.
Molier, B. I., Van Asseldonk, E. H., Hermens, H. J., & Jannink, M. J. (2010). Nature,
timing, frequency and type of augmented feedback; does it influence motor
relearning of the hemiparetic arm after stroke? A systematic review. Disability &
Rehabilitation, 32(22), 1799-1809.
Mozaffarian, D., Benjamin, E. J., Go, A. S., Arnett, D. K., Blaha, M. J., Cushman, M., . .
. Howard, V. J. (2015). Executive Summary: Heart Disease and Stroke
Statistics—2015 Update A Report From the American Heart Association.
Circulation, 131(4), 434-441.
Muellbacher, W., Ziemann, U., Boroojerdi, B., Cohen, L., & Hallett, M. (2001). Role of
the human motor cortex in rapid motor learning. Experimental Brain Research,
136(4), 431-438.
119
Nef, T., & Riener, R. (2005). ARMin-design of a novel arm rehabilitation robot. Paper
presented at the Rehabilitation Robotics, 2005. ICORR 2005. 9th International
Conference on.
Nudo, R. J., Wise, B. M., SiFuentes, F., & Milliken, G. W. (1996). Neural substrates for
the effects of rehabilitative training on motor recovery after ischemic infarct.
Science, 272(5269), 1791-1794.
Oldfield, R. C. (1971). The assessment and analysis of handedness: the Edinburgh
inventory. Neuropsychologia, 9(1), 97-113.
Oujamaa, L., Relave, I., Frooger, J., Mottet, D., & Pelessier, J. (2009). Rehabititation of
Arm Function After Stroke.
Page, S. J., Sisto, S., Levine, P., & McGrath, R. E. (2004). Efficacy of modified
constraint-induced movement therapy in chronic stroke: a single-blinded
randomized controlled trial. Archives of physical medicine and rehabilitation,
85(1), 14-18.
Party, I. S. W. (2008). National clinical guideline for stroke: London: Royal College of
Physicians.
Pavlides, C., Miyashita, E., & Asanuma, H. (1993). Projection from the sensory to the
motor cortex is important in learning motor skills in the monkey. Journal of
Neurophysiology, 70(2), 733-741.
Plautz, E. J., Milliken, G. W., & Nudo, R. J. (2000). Effects of Repetitive Motor Training
on Movement Representations in Adult Squirrel Monkeys: Role of Use versus
Learning. Neurobiology of Learning and Memory, 74(1), 27-55.
120
Reinkensmeyer, D. J., Cole, A. M., Kahn, L. E., & Kamper, D. G. (2002). Directional
control of reaching is preserved following mild/moderate stroke and stochastically
constrained following severe stroke. Experimental brain research, 143(4), 525-
530.
Reinkensmeyer, D. J., Emken, J. L., & Cramer, S. C. (2004). Robotics, motor learning,
and neurologic recovery. Annu. Rev. Biomed. Eng., 6, 497-525.
Reinkensmeyer, D. J., Kahn, L. E., Averbuch, M., McKenna-Cole, A., Schmit, B. D., &
Rymer, W. Z. (2000). Understanding and treating arm movement impairment
after chronic brain injury: progress with the ARM guide. Journal of rehabilitation
research and development, 37(6), 653-662.
Reinkensmeyer, J. (2001). Comparison of robot-assisted reaching to free reaching in
promoting recovery from chronic stroke. Paper presented at the Integration of
Assistive Technology in the Information Age: ICORR’2001, 7th International
Conference on Rehabilitation Robotics.
Ringleb, P., Bousser, M. G., Ford, G., Bath, P., Brainin, M., Caso, V., . . . Csiba, L.
(2010). Ischaemic Stroke and Transient Ischaemic Attack. European Handbook of
Neurological Management, Second Edition, Volume 1, Second Edition, 101-158.
Roby ‐Brami, A., Feydy, A., Combeaud, M., Biryukova, E., Bussel, B., & Levin, M.
(2003). Motor compensation and recovery for reaching in stroke patients. Acta
neurologica scandinavica, 107(5), 369-381.
Rohrer, B., Fasoli, S., Krebs, H. I., Hughes, R., Volpe, B., Frontera, W. R., . . . Hogan, N.
(2002). Movement smoothness changes during stroke recovery. The Journal of
neuroscience, 22(18), 8297-8304.
121
Sanger, T. D. (2004). Failure of Motor Learning for Large Initial Errors. Neural
Computation, 16(9), 1873-1886. doi: 10.1162/0899766041336431
Schaefer, S. Y., Patterson, C. B., & Lang, C. E. (2013). Transfer of Training Between
Distinct Motor Tasks After Stroke Implications for Task-Specific Approaches to
Upper-Extremity Neurorehabilitation. Neurorehabilitation and neural repair,
1545968313481279.
Shadmehr, R., & Mussa-Ivaldi, F. A. (1994). Adaptive representation of dynamics during
learning of a motor task. The Journal of neuroscience, 14(5), 3208-3224.
Shmuelof, L., Krakauer, J. W., & Mazzoni, P. (2012). How is a motor skill learned?
Change and invariance at the levels of task success and trajectory control. Journal
of neurophysiology, 108(2), 578-594.
Smith, M. A., Ghazizadeh, A., & Shadmehr, R. (2006). Interacting adaptive processes
with different timescales underlie short-term motor learning. PLoS biology, 4(6),
e179.
Soechting, J. F., Buneo, C. A., Herrmann, U., & Flanders, M. (1995). Moving effortlessly
in three dimensions: does Donders' law apply to arm movement? The Journal of
neuroscience, 15(9), 6271-6280.
Sukal, T. M., Ellis, M. D., & Dewald, J. P. (2007). Shoulder abduction-induced
reductions in reaching work area following hemiparetic stroke: neuroscientific
implications. Experimental brain research, 183(2), 215-223.
Taub, E., Uswatte, G., & Elbert, T. (2002). New treatments in neurorehabiliation founded
on basic research. Nature Reviews Neuroscience, 3(3), 228-236.
122
Thielman, G. T., Dean, C. M., & Gentile, A. (2004). Rehabilitation of reaching after
stroke: task-related training versus progressive resistive exercise. Archives of
physical medicine and rehabilitation, 85(10), 1613-1618.
Thoroughman, K. A., & Shadmehr, R. (2000). Learning of action through adaptive
combination of motor primitives. Nature, 407(6805), 742-747.
Trombly, C. A. (1992). Deficits of reaching in subjects with left hemiparesis: a pilot
study. American Journal of Occupational Therapy, 46(10), 887-897.
van Beers, R. J., Haggard, P., & Wolpert, D. M. (2004). The role of execution noise in
movement variability. Journal of Neurophysiology, 91(2), 1050-1063.
van de Port, I. G., Kwakkel, G., van Wijk, I., & Lindeman, E. (2006). Susceptibility to
Deterioration of Mobility Long-Term After Stroke A Prospective Cohort Study.
Stroke, 37(1), 167-171.
van Dokkum, L., Hauret, I., Mottet, D., Froger, J., Métrot, J., & Laffont, I. (2013). The
contribution of kinematics in the assessment of upper limb motor recovery early
after stroke. Neurorehabilitation and neural repair, 1545968313498514.
Volpe, B., Krebs, H., Hogan, N., Edelsteinn, L., Diels, C., & Aisen, M. (1999). Robot
training enhanced motor outcome in patients with stroke maintained over 3 years.
Neurology, 53(8), 1874-1874.
Winstein, C., Wing, A., & Whitall, J. (2003). Motor control and learning principles for
rehabilitation of upper limb movements after brain injury. Handbook of
neuropsychology, 9, 79-138.
123
Winstein, C. J., Pohl, P. S., Cardinale, C., Green, A., Scholtz, L., & Waters, C. S. (1996).
Learning a partial-weight-bearing skill: effectiveness of two forms of feedback.
Physical Therapy, 76(9), 985-993.
Winstein, C. J., Rose, D. K., Tan, S. M., Lewthwaite, R., Chui, H. C., & Azen, S. P.
(2004). A randomized controlled comparison of upper-extremity rehabilitation
strategies in acute stroke: a pilot study of immediate and long-term outcomes.
Archives of physical medicine and rehabilitation, 85(4), 620-628.
Wolf, S. L., Winstein, C. J., Miller, J. P., Taub, E., Uswatte, G., Morris, D., . . .
Investigators, E. (2006). Effect of constraint-induced movement therapy on upper
extremity function 3 to 9 months after stroke: the EXCITE randomized clinical
trial. Jama, 296(17), 2095-2104.
Wolpert, D. M., Diedrichsen, J., & Flanagan, J. R. (2011). Principles of sensorimotor
learning. Nature Reviews Neuroscience, 12(12), 739-751.
Wulf, G., & Lewthwaite, R. (2010). Effortless motor learning? An external focus of
attention enhances movement effectiveness and efficiency. Effortless attention: A
new perspective in attention and action, 75-101.
Zarahn, E., Alon, L., Ryan, S. L., Lazar, R. M., Vry, M.-S., Weiller, C., . . . Krakauer, J.
W. (2011). Prediction of motor recovery using initial impairment and fMRI 48 h
poststroke. Cerebral Cortex, 21(12), 2712-2721.
Abstract (if available)
Abstract
Many individuals with stroke who have impaired movement patterns benefit from programs that provide intensive individualized rehabilitation under the supervision of a therapist. Some individuals post-stroke, however, cannot receive the ideal number of rehabilitation sessions due to therapy cost. Since most daily activities involve the upper extremity, the recovery of reaching abilities is critical to maintain independent living and quality of life. ❧ This dissertation work focuses on the design and development of an arm reach training (ART) system consisting of relatively inexpensive, reliable hardware and software to be used in testing and training sessions. The ART system without any assistance from a therapist provides intensive reach training with adaptive visuo-auditory feedback to enhance the motor performance of an affected arm of individuals post-stroke. ❧ Second, to better understand change of reaching performance through short-duration intensive training with the ART system, this dissertation investigates how non-disabled individuals learn to move quickly. We identify specific characteristics of motor skill learning and the learning patterns by analyzing both spatial and temporal kinematics. ❧ Third, this dissertation evaluates the long-term and generalization effects of short-duration and intensive reach training in individuals with chronic stroke and mild to moderate impairments and in age-matched non-disabled right-handed individuals. We investigate change of clinical scores after training as well as the learning patterns, by again analyzing both spatial and temporal kinematics in stroke. ❧ Finally, this dissertation develops a nonlinear statistical model with mixed effect to predict long-term performance change due to training in individuals with chronic stroke. We test the model on three types of participants (i.e., stroke group, age-matched control group, and young control group). We find that the model parameters can be indicators to predict individualized long-term performance in the three groups and that the control group learns more quickly than the stroke group. ❧ We suggest that the results of this dissertation can inform future designs of stroke upper extremity rehabilitation systems for clinical and in-home use. The results may also inspire efforts to improve the rehabilitation of other neuromuscular disorders. Other applications, such as sports rehabilitation and training may benefit from the results of this dissertation.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Arm choice post-stroke
PDF
Relationship between brain structure and motor behavior in chronic stroke survivors
PDF
Planning of unconstrained reach actions after unilateral sensorimotor stroke
PDF
Computational model of stroke therapy and long term recovery
PDF
Hemisphere-specific deficits in the control of bimanual movements after stroke
PDF
Deficits and rehabilitation of upper-extremity multi-joint movements in individuals with chronic stroke
PDF
Modeling motor memory to enhance multiple task learning
PDF
The effects of fast walking, biofeedback, and cognitive impairment on post-stroke gait
PDF
Hemispheric specialization of reach-to-grasp actions
PDF
Design of adaptive automated robotic task presentation system for stroke rehabilitation
PDF
Understanding reactive balance control strategies in non-disabled and post-stroke gait
PDF
Experimental and computational explorations of different forms of plasticity in motor learning and stroke recovery
PDF
The brain and behavior of motor learning: the what, how and where
PDF
Using ecological momentary assessment to study the impact of social-cognitive factors on paretic hand use after stroke
PDF
Relationship between interhemispheric inhibition and bimanual coordination in musicians
PDF
Virtual surgeries as a tool for studying motor learning
PDF
Behavioral, muscular and dynamical changes in low force dexterous manipulation during development and aging
PDF
Reaching decisions in dynamic environments
PDF
The structure of kinematic variability during hopping: a window into motor control factors in overuse injury
PDF
A non-invasive and intuitive command source for upper limb prostheses
Asset Metadata
Creator
Park, Hyeshin
(author)
Core Title
Learning reaching skills in non-disabled and post-stroke individuals
School
School of Dentistry
Degree
Doctor of Philosophy
Degree Program
Biokinesiology
Publication Date
07/31/2015
Defense Date
05/06/2015
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
arm movements,generalization,long-term retention,motor skill learning,OAI-PMH Harvest,prediction of training effect,reach training,stroke rehabilitation
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Schweighofer, Nicholas (
committee chair
), Gordon, James G. (
committee member
), Gregor, Robert J. (
committee member
), Valero-Cuevas, Francisco (
committee member
), Winstein, Carolee J. (
committee member
)
Creator Email
hyeshinp@usc.edu,resungod@gmail.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-621375
Unique identifier
UC11303079
Identifier
etd-ParkHyeshi-3773.pdf (filename),usctheses-c3-621375 (legacy record id)
Legacy Identifier
etd-ParkHyeshi-3773.pdf
Dmrecord
621375
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Park, Hyeshin
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
Tags
arm movements
generalization
long-term retention
motor skill learning
prediction of training effect
reach training
stroke rehabilitation