Conference PaperPDF Available

Using wearable sensors to measure motor abilities following stroke

Authors:
T. Hester
T. Hester
T. Hester
  • This person is not on ResearchGate, or hasn't claimed this research yet.
R. Hughes
R. Hughes
R. Hughes
  • This person is not on ResearchGate, or hasn't claimed this research yet.
D.M. Sherrill
D.M. Sherrill
D.M. Sherrill
  • This person is not on ResearchGate, or hasn't claimed this research yet.
B. Knorr
B. Knorr
B. Knorr
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Show all 7 authorsHide
Download full-text PDF
Read full-text
Download citation

Abstract and Figures

Motor abilities of stroke survivors are often severely affected. Post-stroke rehabilitation is guided by the use of clinical assessments of motor abilities. Clinical assessment scores can be predicted by models based on features extracted from the wearable sensor data. Wearable sensors would allow monitoring of subjects in the home and provide accurate assessments to guide the rehabilitation process. We propose the use of a wearable sensor system to assess the motor abilities of stroke victims. Preliminary results from twelve subjects show the ability of this system to predict clinical scores of motor abilities.
Figures - uploaded by Paolo Bonato
Author content
All figure content in this area was uploaded by Paolo Bonato
Content may be subject to copyright.
Content uploaded by Paolo Bonato
Author content
All content in this area was uploaded by Paolo Bonato
Content may be subject to copyright.
Using Wearable Sensors to Measure Motor Abilities following Stroke
Todd Hester
1
, Richard Hughes
1
, Delsey M. Sherrill
1
, Bethany Knorr
2
,
Metin Akay
3
, Joel Stein
1
, and Paolo Bonato
1,4
1
Dept of PM&R, Harvard Medical School, Spaulding Rehabilitation Hospital, Boston MA
2
Thayer School of Engineering, Dartmouth College, Hanover NH
3
Dept of Bioengineering, Arizona State University, Tempe AZ
4
The Harvard-MIT Division of Health Science and Technology, Cambridge MA
tahester@partners.org, rhughes1@partners.org, delsey.sherrill@gmail.com,
bethany.r.knorr@dartmouth.edu
, metin.akay@asu.edu, jstein@partners.org,
pbonato@partners.org
Abstract
Motor abilities of stroke survivors are often
severely affected. Post-stroke rehabilitation is guided
by the use of clinical assessments of motor abilities.
Clinical assessment scores can be predicted by models
based on features extracted from the wearable sensor
data. Wearable sensors would allow monitoring of
subjects in the home and provide accurate assessments
to guide the rehabilitation process. We propose the use
of a wearable sensor system to assess the motor
abilities of stroke victims. Preliminary results from
twelve subjects show the ability of this system to
predict clinical scores of motor abilities.
Keywords: Wearable Sensors, Stroke, Clinical
Assessment
1. Introduction
Approximately 700,000 people are affected by
stroke each year in the United States and about
275,000 die from stroke each year [1]. Strokes affect a
person’s cognitive, language, perceptual, sensory, and
motor abilities [2]. More than 1,100,000 Americans
have reported difficulties with functional limitations
following stroke [3]. Recovery from stroke is a long
process that continues beyond the hospital stay and
into the home setting. The rehabilitation process is
guided by clinical assessments of motor abilities.
Accurate assessment of motor abilities is important
in selecting the best therapies for stroke survivors.
These assessments are based on observations of
subjects’ motor behavior using standardized clinical
rating scales. The accuracy and consistency of
observational assessments may vary greatly across
clinicians [4]. Wearable sensors could be used to
provide more accurate measures or could be used in
addition to observational clinical tools. Wearable
systems have the ability to measure motor behavior at
home and for longer periods than could be observed in
a clinical setting. Accelerometers can capture specific
patterns of movement relating to motor disabilities. We
propose that wearable systems can be used to predict
clinical scores of motor abilities and we present an
initial analysis of data demonstrating an association
between accelerometer data and clinical scores.
2. Methods
Twelve subjects who had a stroke within the past 2
to 24 months were recruited for the study. Each subject
was evaluated by a clinician using standardized clinical
motor performance scales, including the Fugl-Meyer
Assessment of Sensorimotor Recovery after Stroke,
Chedoke-McMaster Stroke Assessment, Wolf Motor
Performance Test, and the Reaching Performance
Scale. These scales measure dimensions of upper limb
motor behavior including movement quality, stage of
motor recovery, use of compensatory movement
strategies, and the ability to perform functional tasks.
All testing was performed at Spaulding Rehabilitation
Hospital. Subjects provided informed consent
approved by the hospital’s research review board.
Accelerometers were attached to the affected arm and
the trunk (Figure 1).
Sensor data was recorded using the Vitaport 3
(Temec BV, The Netherlands) ambulatory recorder,
which was worn on the waist. Subjects performed
multiple repetitions of tasks requiring reaching and
prehension, selected from the clinical scales. The tasks
included reaching to close and distant objects, placing
the hand or forearm from lap to a table, pushing and
pulling a weight across a table, drinking from a
beverage can, lifting a pencil, flipping a card, and
turning a key.
The accelerometer data was digitally low-pass
filtered in Matlab with a cutoff frequency of 15 Hz to
remove high frequency noise. Both this low-pass
filtered and a high-pass filtered version of the data
were utilized in the analysis. The high-pass filtered
version of the data was derived in an attempt to isolate
actual acceleration components from gross postural
adjustments. A 1 Hz cut-off frequency was used.
The signals were marked manually during testing
using a 5 V pulse and marks were checked via visual
inspection of the data. The marks were used to
segment the data by task through an automated
software procedure based on threshold detection of the
manual markings. Manipulation tasks such as card
flipping were also segmented within the task using a
touch sensor. They were segmented into a reaching
epoch, a manipulation epoch, and a release/return
epoch. Subjects performed between 10 and 20
repetitions of each task, resulting in an average of 109
segments for each subject. The epochs ranged from
0.30 s to 23.3 s, with a mean length of 2.54 s and a
standard deviation of 1.90 s. The following features
were extracted from each epoch of accelerometer data
for later analysis:
• Mean value of the low-pass filtered data
• Root-Mean-Square value (this feature and the
following ones were all derived from the
high-pass filtered version of the accelerometer
data)
• Dominant frequency
• Ratio of energy in 0.2 Hz bin around the
dominant frequency to total energy (measure
of periodicity)
• Range of autocovariance
• Root-Mean-Square value of the derivative of
acceleration (i.e. jerk time series)
• Dominant frequency of the jerk time series
• Ratio of energy in 0.2 Hz bin around the
dominant frequency of jerk to total energy
(measure of periodicity)
• Peak velocity
• Jerk metric (i.e. the RMS jerk normalized by
the peak velocity
• Approximate entropy (nonlinear measure of
complexity)
• Correlation at zero lag between selected pairs
of accelerometer time series
• Peak correlation within a 1 s window between
selected pairs of accelerometer time series
• Lag time of the peak correlation between
selected pairs of accelerometer time series
Features from each task were imported into the
Waikato Environment for Knowledge Analysis
(WEKA) for exploratory analysis [5]. Initially, we
used WEKA to look at scatterplots of features and
clinical scores to assess the suitability of the current
sensors, tasks, and features to predict the subjects’
clinical scores. Next, we built linear regression models
in WEKA to explore their ability to predict the clinical
scores as well as to examine which feature sets were
useful in predicting the scores. Features for the linear
regression models were selected by the M5 method,
which performs a backward stepwise regression using
the Akaike information criterion [6]. The models
provided feature sets for each clinical score. Then,
linear regression models were built in Matlab to
predict clinical scores using a leave-one-subject-out
method (i.e. clinical scores for each subject were
predicted based on a model built with data from all
other subjects). All of the features were normalized to
have a mean of zero and a standard deviation of one.
Table 1 lists the clinical scores predicted by the model.
Scores were predicted based on analysis of features
from 16 different segments from 8 tasks. We looked at
the forearm to table, hand to table, pushing a weight,
and retrieving a weight tasks in their entirety and at the
Figure 1 Sensor setup and orientation of the axes
of the accelerometers.
can lifting, pencil lifting, card-flipping, and key-
turning tasks in three segments each.
3. Results
Figure 2 shows a scatterplot of the peak correlation
within a 1 s window between the index finger and
hand accelerometers and the peak velocity of the
thumb accelerometer during the manipulation epoch of
the can lifting task. The colors represent the scores on
the Chedoke-McMaster Hand Stage. The figure shows
that higher peak velocities for the thumb accelerometer
data correlate to higher clinical scores, while higher
correlations between the index finger and hand data
correlate with lower scores.
Models predicting all seven clinical scores with
features from all 16 different task segments were built.
Table 1 shows the root mean square error of the best
model for each clinical score along with the range for
each clinical score. The models were most successful
in predicting the Chedoke-McMaster Hand Stage and
the shoulder and elbow portion of the Fugl-Meyer
scale, with relative errors close to 10%. The models
were less successful in predicting other clinical scores.
Table 2 shows the RMS error of the prediction of
the shoulder and elbow portion of the Fugl-Meyer
scale and the Chedoke-McMaster Hand Stage using
each task. The best predictors of the Hand Stage were
models built with features extracted from the forearm
to table task and the manipulation segments of the can
lifting and card flipping tasks. The best predictors of
the shoulder and elbow portion of the Fugl-Meyer
were the “reaching” segments of the manipulation
tasks. The worst predictor of both clinical scores was
the “releasing” segment of the pencil lifting task.
Table 1. Errors in predicting clinical scores
Clinical Score
RMS
Error
Score
Range
Range for
Subjects
Tested
Chedoke-
McMaster Hand
Stage
0.42 1-7 3-5
Chedoke-
McMaster Arm
Stage
1.27 1-7 3-7
Wolf Test –
Median Time
1.30 0-120 1.57-9.66
Fugl-Meyer
Shoulder-Elbow
2.35 0-30 19-30
Fugl-Meyer
Shoulder-Elbow
3.32 0-24 4-22
Fugl-Meyer Total
Score
10.01 0-66 29-63
Table 2. RMS Errors in predicting clinical scores for
different tasks.
Task
Chedoke-
McMaster
Hand Stage
Fugl-Meyer
Shoulder-
Elbow
Can - Segment 1 0.58 3.66
Can - Segment 2 0.50 6.02
Can - Segment 3 0.67 5.13
Card - Segment 1 0.58 2.35
Card - Segment 2 0.50 3.74
Card - Segment 3 0.67 5.23
Forearm To Table 0.67 4.90
Hand To Table 0.42 6.23
Key - Segment 1 0.86 2.99
Key - Segment 2 0.90 3.75
Key - Segment 3 0.67 4.80
Pencil - Segment 1 0.62 3.10
Pencil - Segment 2 0.67 5.22
Pencil - Segment 3 0.93 7.76
Push Weight 0.71 6.19
Retrieve Weight 0.74 5.16
Figure 2 Scatterplot of peak correlation between
accelerometer data from index finger and hand and
peak velocity derived from thumb accelerometer
data in comparison to Chedoke-McMaster Hand
Stage scores. A line separates well samples
associated with a score of 3 from samples
associated with a score of 5. Samples for a score o
f
4 are in between and overlap with the rest of the
data.
Table 3 shows the linear regression model used to
predict the shoulder and elbow portion of the Fugl-
Meyer score using features from the reaching segment
of the card flipping task. Table 4 shows the actual
scores, predicted scores, and the standard deviation of
the predicted scores for the shoulder and elbow portion
of the Fugl-Meyer scale using the model shown in
Table 3. Five of the predicted scores were within 1
point of the actual score, and the closest was within
0.02. The worst prediction was 35.81 for a subject with
a clinical score of 30.
Table 3. Linear regression model for prediction of the
shoulder and elbow portion of the Fugl-Meyer score
based on the reaching segment of the card flipping task
Coefficient Feature
1.18 * Mean of Forearm X acc
1.81 * Mean of Forearm Y acc
0.65 * Mean of Upper Arm X acc
1.50 * Mean of Upper Arm Y acc
-0.86 * Mean of sternum acc
0.99 * RMS of forearm X acc
1.66 * RMS of forearm Y acc
0.31 * RMS of Upper Arm X acc
-1.31 * RMS of Upper Arm Y acc
-0.16 * RMS of sternum acc
-0.28 * Peak Corr. of Thumb and Hand
Table 4. Scores for the shoulder and elbow portion of
the Fugl-Meyer score
Subject
Actual
Score
Predicted
Score
STD of
Prediction
A 23 22.15 0.86
B 26 26.50 1.40
C 19 20.66 0.40
D 26 22.00 0.63
E 30 35.81 2.47
F 30 29.98 2.83
G 24 24.97 0.18
H 24 22.63 0.99
I 30 26.04 1.57
J 30 28.79 0.75
K 20 21.16 1.11
L 27 27.40 1.38
4. Discussion and Conclusion
The results of the linear regression models have
been promising so far. Our models predicted two
clinical scores within 10% of the average score. Our
sensor system showed the ability to detect specific
movement patterns related to clinical scores of
movement ability. For example, a negative coefficient
was associated with the root mean square value of the
sternum accelerometer channel in most of the linear
regression models, indicating that the method was able
to detect compensatory trunk movements related to
lower clinical scores. The jerk metric and the root
mean square of the jerk time series usually had a high
coefficient in the models, showing that the system
determined that smooth movement was significantly
related to the clinical scores. Scores such as the median
time on the Wolf Motor Function Test may be difficult
to predict because of non-linear relationships between
post-stroke motor ability and performance scores on
this test. To predict these scores more accurately, it
may be necessary to include non-linear parameters or
create a non-linear model to predict the scores. We are
currently collecting data from more subjects, which
will allow us to improve the linear regression models
and explore the use of nonlinear models. The small
size of the current dataset limits the number of features
we can use in the models. Including more features is
expected to improve the model. Collecting data from
subjects with a wider range of motor abilities and
clinical scores will allow us to develop a more accurate
linear regression model.
Acknowledgments
This study was supported by the grant entitled
"Field Measures of Functional Tasks for CIT
Intervention", #R21HD045873-01, NIH-NICHD.
References
[1] American Heart Association, Heart Disease and Stroke
Statistics – 2005 Update, American Heart Association,
Dallas, TX, 2004.
[2] National Institute of Neurological Disorders and Stroke
(NINDS), Stroke: Hope Through Research, NINDS,
Washington, DC, 2004.
[3] Center for Disease Control, Morbidity and Mortality
Weekly Report, CDC, Atlanta, GA, 2001.
[4] V. Pomeroy, A. Pramanik, L. Sykes, J. Richards, and E.
Hill, “Agreement between physiotherapists on quality
of movement rated via videotape”, Clinical
Rehabilitation, May 2003, pp. 264-272.
[5] IH Witten and E. Frank, Data Mining: Practical
machine learning tools and techniques. Morgan
Kaufman, San Francisco, 2005.
[6] H Akaike, “A New Look at Statistical Model
Identification." IEEE Transactions on Automatic
Control, IEEE, 1974, pp. 716-722.
... 04 Thanawattano et al., 2015;Rigas et al., 2016;Bravo, 2017;Delrobaei et al., 2018;Lukšys et al., 2018;Zhu and Miller, 2020;Schaefer et al., 2021;Chan et al., 2022). Twelve studies included patients with essential tremor (Heldman et al., 2011b;Šprdlík et al., 2011;Gallego et al., 2012;Budini et al., 2014;Heo et al., 2015;Samotus et al., 2016;Chan et al., 2018;López-Blanco et al., 2018;Benito-LeónSerrano et al., 2019;Kwon et al., 2020;McGurrin et al., 2021;Ali et al., 2022) and 11 included adults post-stroke (Kamper et al., 2002;Knorr, 2005;Hester, 2006;Thies et al., 2009;Parnandi et al., 2010;Patel et al., 2010;Del Din et al., 2011;Zhang et al., 2012;Otten et al., 2015;van Meulen et al., 2015;Repnik et al., 2018), whereas six included adults with MS (Ketteringham, 2011;Carpinella et al., 2014;Carpinella et al., 2015;Teufl et al., 2017;Western et al., 2019;Teufl et al., 2021). One study included adults with HD and eight studies included children or adults with ataxia (Bennasar et al., 2018;Martinez-Manzanera et al., 2018;Krishna et al., 2019;Kashyap et al., 2020;Nguyen et al., 2020;Tran et al., 2020;Dominguez-Vega et al., 2021;Oubre et al., 2021;Gupta, 2022). ...
... 05 Type, number and location of sensors Table S2 yield an overview of type and number of sensors used, and their respective location for all included studies. From the 101 identified studies, 24 studies used an accelerometer (Makabe and Sakamoto, 2000;Hoff et al., 2001;Keijsers et al., 2003;Bonato et al., 2004;Legros et al., 2004;Knorr, 2005;Hester, 2006;Okuno et al., 2006;Patel et al., 2009;Yokoe et al., 2009;Cole et al., 2010;Patel et al., 2010;Del Din et al., 2011;Griffiths et al., 2012;Budini et al., 2014;Rahimi et al., 2015;Samotus et al., 2016;Bravo, 2017;Teufl et al., 2017;Bennasar et al., 2018;Kim et al., 2018;Habets et al., 2021;Schaefer et al., 2021;Gupta, 2022), 13 studies measured motion with a gyroscope or angular sensor (Koop et al., 2006;Salarian et al., 2007;Espay et al., 2011;Hoffman and McNames, 2011;Jun et al., 2011;Kim et al., 2011;Lee et al., 2015b;Heo et al., 2015;Summa et al., 2017;Kwon et al., 2018;López-Blanco et al., 2018;Kwon et al., 2020), four studies collected motion data using an orientation sensor, motion sensor or magnetic motion tracker (Sanger, 2006;Chelaru et al., 2010;di Biase et al., 2018;Kamper et al., 2002), 58 studies used IMUs including an accelerometer and gyroscope and three studies included IMUs but only used the acceleration signal for further analysis (Chan et al., 2018;Benito-LeónSerrano et al., 2019;Ali et al., 2022), while one study used IMUs but only processed angular velocity signals (Thanawattano et al., 2015). ...
... Four studies used five sensors on the upper limbs, with sensor placement on hand, wrist, upper arm, shoulder and sternum in MS (Ketteringham, 2011;Western et al., 2019), thumb, index finger, hand, forearm, upper arm and sternum for Di Biase et al. and hand forearm, upper arm, head and back for Sanger et al. (note that both studies are based on magnetic or orientation sensors) (Sanger, 2006;di Biase et al., 2018). Seven studies used six sensors, where the sensor placement was thumb, index finger, hand, forearm, upper arm and sternum (Hester, 2006;Patel et al., 2010). Two studies placed the sensors on hands, forearms and upper arms (Chan et al., 2018;Lukšys et al., 2018), while Krishna et al. used one sensor but subsequently fixed it on both hands, wrists and ankles, thus including six sensor signals in the analysis (Krishna et al., 2019). ...
Assessment of movement disorders using wearable sensors during upper limb tasks: A scoping review
Article
Full-text available
  • Jan 2023
Background: Studies aiming to objectively quantify movement disorders during upper limb tasks using wearable sensors have recently increased, but there is a wide variety in described measurement and analyzing methods, hampering standardization of methods in research and clinics. Therefore, the primary objective of this review was to provide an overview of sensor set-up and type, included tasks, sensor features and methods used to quantify movement disorders during upper limb tasks in multiple pathological populations. The secondary objective was to identify the most sensitive sensor features for the detection and quantification of movement disorders on the one hand and to describe the clinical application of the proposed methods on the other hand. Methods: A literature search using Scopus, Web of Science, and PubMed was performed. Articles needed to meet following criteria: 1) participants were adults/children with a neurological disease, 2) (at least) one sensor was placed on the upper limb for evaluation of movement disorders during upper limb tasks, 3) comparisons between: groups with/without movement disorders, sensor features before/after intervention, or sensor features with a clinical scale for assessment of the movement disorder. 4) Outcome measures included sensor features from acceleration/angular velocity signals. Results: A total of 101 articles were included, of which 56 researched Parkinson’s Disease. Wrist(s), hand(s) and index finger(s) were the most popular sensor locations. Most frequent tasks were: finger tapping, wrist pro/supination, keeping the arms extended in front of the body and finger-to-nose. Most frequently calculated sensor features were mean, standard deviation, root-mean-square, ranges, skewness, kurtosis/entropy of acceleration and/or angular velocity, in combination with dominant frequencies/power of acceleration signals. Examples of clinical applications were automatization of a clinical scale or discrimination between a patient/control group or different patient groups. Conclusion: Current overview can support clinicians and researchers in selecting the most sensitive pathology-dependent sensor features and methodologies for detection and quantification of upper limb movement disorders and objective evaluations of treatment effects. Insights from Parkinson’s Disease studies can accelerate the development of wearable sensors protocols in the remaining pathologies, provided that there is sufficient attention for the standardisation of protocols, tasks, feasibility and data analysis methods.
View
... Total time was the time (seconds) to complete the entire paragraph, whereas writing time was the time spent with the pen on paper, and percentage writing was the time spent on paper with respect to total time. The literature reports different measures for smoothness, for example, number of segments (Rosenblum, Dvorkin et al., 2006), root-mean-square value (Hester et al., 2006), signal to noise velocity fluctuations (Danna et al., 2013), and mean arrest period ratio (MAPR) and jerk (Rohrer et al., 2002). As the present study considered the kinematics across sentences, we measured smoothness with the number of lifts and MAPR (see for similar methods, Hester et al., 2006;Hogan & Sternad, 2009;Rohrer et al., 2002). ...
... The literature reports different measures for smoothness, for example, number of segments (Rosenblum, Dvorkin et al., 2006), root-mean-square value (Hester et al., 2006), signal to noise velocity fluctuations (Danna et al., 2013), and mean arrest period ratio (MAPR) and jerk (Rohrer et al., 2002). As the present study considered the kinematics across sentences, we measured smoothness with the number of lifts and MAPR (see for similar methods, Hester et al., 2006;Hogan & Sternad, 2009;Rohrer et al., 2002). A lift of the pen was measured when the pen was lifted from the paper, identified as a clear drop of the tablet signal. ...
... Yet, this comes with a limitation, as it was not clear beforehand which of the measures can or are most relevant in distinguishing satisfactory and unsatisfactory handwriting in children who are still developing their handwriting skills. For example, MAPR, which was used as an indicator of writing smoothness, was included based on its sensitivity in examining handwriting in adult patients with Parkinson's and stroke (Hester et al., 2006;Hogan & Sternad, 2009;Rohrer et al., 2002). However, we found no literature examining children that could serve as a direct comparison. ...
The Effects of Number and Separation of Support Lines on the Size, Velocity, and Smoothness of Handwriting
Article
  • Apr 2022
We examined the effects of number of and separation between support lines on handwriting characteristics of primary school students with satisfactory and unsatisfactory handwriting. Students (mean age 7.9 years) copied a text on paper with a baseline and with two or four support lines with a separation of 3 or 4 mm between the central lines. Handwriting size, velocity, and smoothness were determined for the four conditions relative to baseline. Children with unsatisfactory handwriting wrote larger and had more lifts during baseline condition. Writing between support lines, especially with small separation, immediately reduced the size of handwriting, but also adversely affected velocity and smoothness. Future research is needed to assess long-term effects.
View
... New telemedicine systems already use emerging technologies, such as, 802.15 and 802. 16, which are characterized by distribution networks for medical information providing life-saving services [5]. ...
... The rehabilitation process is long and continues beyond the hospital. Wearable sensors (normally with accelerometers) can be used to monitor home-bound people (like motor behavior in the home and predict some important clinical conditions, such as motor disabilities with hands, arms, legs or other human body parts), and it provides accurate assessments to guide the rehabilitation process [16]. ...
View
... Thus, this study compared the outcomes of artificial neural network (ANN), Gaussian process, k-nearest neighbor (kNN), linear regression, and support vector regression (SVR) for the proposed method. These five machine learning algorithms were selected given their frequent use for wearable systems in previous studies [32][33][34][35][36]. In addition, ANN, Gaussian process, kNN, and SVR were expected to build regression models for nonlinear and complex data distribution [37][38][39][40]. ...
... Thus, the left foot was always fixed to the same spot between the footrests of the wheelchair during each trial. As shown in Table 1, these foot patterns included different combinations of anteroposterior (10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35)(36)(37)(38)(39)(40), and 41-55 %height) and mediolateral (10-20, 21-30, and 31-40 %height) foot distances. These foot positions were defined for collecting various foot position data. ...
Foot Position Measurement during Assistive Motion for Sit-to-Stand Using a Single Inertial Sensor and Shoe-Type Force Sensors
Article
Full-text available
Assistive motion for sit-to-stand causes lower back pain (LBP) among caregivers. Considering previous studies that showed that foot position adjustment could reduce lumbar load during assistive motion for sit-to-stand, quantitative monitoring of and instructions on foot position could contribute toward reducing LBP among caregivers. The present study proposes and evaluates a new method for the quantitative measurement of foot position during assistive motion for sit-to-stand using a few wearable sensors that are not limited to the measurement area. The proposed method measures quantitative foot position (anteroposterior and mediolateral distance between both feet) through a machine learning technique using features obtained from only a single inertial sensor on the trunk and shoe-type force sensors. During the experiment, the accuracy of the proposed method was investigated by comparing the obtained values with those from an optical motion capture system. The results showed that the proposed method produced only minor errors (less than 6.5% of body height) when measuring foot position during assistive motion for sit-to-stand. Furthermore, Bland–Altman plots suggested no fixed errors between the proposed method and the optical motion capture system. These results suggest that the proposed method could be utilized for measuring foot position during assistive motion for sit-to-stand.
View
... For example, Raihan's study employed convolutional neural networks to predict the summary score of rehabilitation exercise sessions [9]. Similarly, various studies use motion-related sensors with data-driven techniques to calculate scores for assessments [11][12][13][14][15]. For instance, Lee's study used time-frequency informed singular spectrum analysis to automatically assess the score of one grasping task in the Action Research Arm Test (ARAT) [11]. ...
Autonomous modeling of repetitive movement for rehabilitation exercise monitoring
Article
Full-text available
  • Jul 2022
  • BMC MED INFORM DECIS
Background Insightful feedback generation for daily home-based stroke rehabilitation is currently unavailable due to the inefficiency of exercise inspection done by therapists. We aim to produce a compact anomaly representation that allows a therapist to pay attention to only a few specific sections in a long exercise session record and boost their efficiency in feedback generation. Methods This study proposes a data-driven technique to model a repetitive exercise using unsupervised phase learning on an artificial neural network and statistical learning on principal component analysis (PCA). After a model is built on a set of normal healthy movements, the model can be used to extract a sequence of anomaly scores from a movement of the same prescription. Results The method not only works on a standard marker-based motion capture system but also performs well on a more compact and affordable motion capture system based-on Kinect V2 and wrist-worn inertial measurement units that can be used at home. An evaluation of four different exercises shows its potential in separating anomalous movements from normal ones with an average area under the curve (AUC) of 0.9872 even on the compact motion capture system. Conclusions The proposed processing technique has the potential to help clinicians in providing high-quality feedback for telerehabilitation in a more scalable way.
View
... To tackle these issues, researchers have developed applications to assess rehabilitation outcomes using novel technologies namely "wearable sensors" [28], which provide a high level of portability and low price giving researchers and therapists a plethora of possibilities and solutions [29]. Indeed, wearable sensors allow patients to execute their exercises at home relieving them of the drain of transportation. ...
Wearable sensors and machine learning in post-stroke rehabilitation assessment: A systematic review
Article
Full-text available
  • Jan 2022
  • BIOMED SIGNAL PROCES
A cerebrovascular accident or stroke is the second commonest cause of death in the world. If it is not fatal, it can result in paralysis, sensory impairment and significant disability. Rehabilitation plays an important role to help survivors relearn lost skills and assist them to regain independence and thus ameliorate their quality of life. With the development of technology, researchers have come up with new solutions to assist clinicians in monitoring and assessing their patients; as well as making physiotherapy available to all. The objective of this review is to assess the recent developments made in the field of post-stroke rehabilitation using wearable devices for data collection and machine learning algorithms for the exercises’ evaluation. To do so, PRISMA guidelines for systematic reviews were followed. Scopus, Lens, PubMed, ScienceDirect and Microsoft academic were electronically searched. Peer-reviewed papers using sensors in post-stroke rehabilitation were included, for the period between 2015 to August 2021. Thirty-three publications that used wearable sensors for patients’ assessment were included. Based on that, we have proposed a taxonomy that divided the assessment systems into three categories namely activity recognition, movement classification, and clinical assessment emulation. Moreover, The most commonly employed sensors as well as the most targeted body–limbs, outcome measures, and study designs are reviewed, in addition to the examination of the machine learning approaches starting from the feature engineering to the classification done. Finally, limitations and potential study directions in the field are presented.
View
NSF DARE-transforming modeling in neurorehabilitation: a patient-in-the-loop framework
Article
Full-text available
  • Feb 2024
  • J NEUROENG REHABIL
In 2023, the National Science Foundation (NSF) and the National Institute of Health (NIH) brought together engineers, scientists, and clinicians by sponsoring a conference on computational modelling in neurorehabiilitation. To facilitate multidisciplinary collaborations and improve patient care, in this perspective piece we identify where and how computational modelling can support neurorehabilitation. To address the where, we developed a patient-in-the-loop framework that uses multiple and/or continual measurements to update diagnostic and treatment model parameters, treatment type, and treatment prescription, with the goal of maximizing clinically-relevant functional outcomes. This patient-in-the-loop framework has several key features: (i) it includes diagnostic and treatment models, (ii) it is clinically-grounded with the International Classification of Functioning, Disability and Health (ICF) and patient involvement, (iii) it uses multiple or continual data measurements over time, and (iv) it is applicable to a range of neurological and neurodevelopmental conditions. To address the how, we identify state-of-the-art and highlight promising avenues of future research across the realms of sensorimotor adaptation, neuroplasticity, musculoskeletal, and sensory & pain computational modelling. We also discuss both the importance of and how to perform model validation, as well as challenges to overcome when implementing computational models within a clinical setting. The patient-in-the-loop approach offers a unifying framework to guide multidisciplinary collaboration between computational and clinical stakeholders in the field of neurorehabilitation.
View
Wearable Technology for Smart Manufacturing in Industry 5.0
Chapter
  • Jun 2023
The innovation of wearable Internet of Things devices has fuelled the transition from Industry 4.0 to Industry 5.0. Increasing resource efficiency, safety, and economic efficiency are some of the main goals of Industry 5.0. Herein, wearable Internet of Things devices is parallel to humans to optimize human tasks and meet a new Industry’s requirements. Integrating artificial intelligence algorithms and IoT into wearable technologies and the progress of sensors has created significant innovations in many fields, such as manufacturing, health, sports, etc.. However, wearable technologies have faced challenges and difficulties such as security, privacy, accuracy, latency, and connectivity. More specifically, the increasingly massive and complex data volume has dramatically influenced the improvement of the limits. However, these challenges have created a new solution: the federated Learning algorithm. In recent years, federated learning has been implemented with deep learning and AI to enhance powerful computing with big data, stable accuracy, and ensure the security of edge devices. In this chapter, the first objective is to survey the applications of wearable Internet of Things devices in industrial sectors, particularly in manufacturing. Second, the challenges of wearable Internet of Things devices are discussed. Finally, this chapter provides case studies applying machine learning, deep learning, and federated learning in fall and fatigue classification. These cases are the two most concerning work efficiency and safety topics in Smart Manufacturing 5.0.KeywordsWearable technologySmart ManufacturingIndustry 5.0
View
View
Assessment of upper limb movement disorders using wearable sensors during functional tasks: a systematic review
Preprint
Full-text available
  • Oct 2022
Background Studies aiming to objectively quantify upper limb movement disorders during functional tasks using wearable sensors have recently increased, but there is a wide variety in described measurement and analyzing methods, hampering standardization of methods in research and clinics. Therefore, the primary objective of this review was to provide an overview of sensor set-up and type, included tasks, sensor features and methods used to quantify movement disorders during upper limb tasks in multiple pathological populations. The secondary objective was to select the most sensitive sensor features for symptom detection and quantification and discuss application of the proposed methods in clinical practice. Methods A literature search using Scopus, Web of Science, and PubMed was performed. Articles needed to meet following criteria: (1) participants were adults/children with a neurological disease, (2) (at least) one sensor was placed on the upper limb for evaluation of movement disorders during functional tasks, (3) comparisons between: groups with/without movement disorders, sensor features before/after intervention, or sensor features with a clinical scale for assessment of the movement disorder. (4) Outcome measures included sensor features from acceleration/angular velocity signals. Results A total of 101 articles were included, of which 56 researched Parkinson’s Disease. Wrist(s), hand and index finger were the most popular sensor locations. The most frequent tasks for assessment were: finger tapping, wrist pro/supination, keeping the arms extended in front of the body and finger-to-nose. The most frequently calculated sensor features were mean, standard deviation, root-mean-square, ranges, skewness, kurtosis and entropy of acceleration and/or angular velocity, in combination with dominant frequencies and power of acceleration signals. Examples of clinical applications were automatization of a clinical scale or discrimination between a patient/control group or different patient groups. Conclusion Current overview can support clinicians and researchers to select the most sensitive pathology-dependent sensor features and measurement methodologies for detection and quantification of upper limb movement disorders and for the objective evaluations of treatment effects. The insights from Parkinson’s Disease studies can accelerate the development of wearable sensors protocols in the remaining pathologies, provided that there is sufficient attention for the standardisation of protocols, tasks, feasibility and data analysis methods.
View
Agreement between physiotherapists on quality of movement rated via videotape
Article
Full-text available
  • Jun 2003
Although achieving quality of movement after stroke is an important aim of physiotherapy it is rarely measured objectively or described explicitly. To test whether physiotherapists agree on a composite measure of quality of movement. SETTING; A movement analysis laboratory Ten stroke patients and 10 healthy age-matched volunteers. Prospective correlational. All subjects were videofilmed performing three trials of six standardized functional tasks. Two videotapes were made, each with a different randomized order of appearance of the trials. Ten senior physiotherapists independently rated the videotapes twice using a 100-mm visual analogue scale. Analysis of variance models were fitted to transformed data. Estimates of components of variance were calculated and presented as a percentage of the total variance for differences, within subjects (intra-subject), between raters (inter-rater) and within raters (intra-rater). An acceptable percentage was set at less than 10%. The percentage of intra-subject variance ranged from 1% (pick up box and walking) to 9% (step on block). The percentage of inter-rater variance ranged from 18% (pick up pencil) to 38% (sit to stand). The percentage of intra-rater variance was less than 1% for all tasks. Although physiotherapists disagreed with each other on quality of movement they were more consistent in their own scoring.
View
View
View
View
Hope through research
Article
  • Feb 1987
Schizophrenia is the most under-researched major disease in the Western world.
View
A New Look At The Statistical Model Identification
Article
  • Jan 1975
The history of the development of statistical hypothesis testing in time series analysis is reviewed briefly and it is pointed out that the hypothesis testing procedure is not adequately defined as the procedure for statistical model identification. The classical maximum likelihood estimation procedure is reviewed and a new estimate minimum information theoretical criterion (AIC) estimate (MAICE) which is designed for the purpose of statistical identification is introduced. When there are several competing models the MAICE is defined by the model and the maximum likelihood estimates of the parameters which give the minimum of AIC defined by AIC = (-2)log-(maximum likelihood) + 2(number of independently adjusted parameters within the model). MAICE provides a versatile procedure for statistical model identification which is free from the ambiguities inherent in the application of conventional hypothesis testing procedure. The practical utility of MAICE in time series analysis is demonstrated with some numerical examples.
View
Heart Disease and Stroke Statistics
  • Jan 2004
American Heart Association, Heart Disease and Stroke Statistics – 2005 Update, American Heart Association, Dallas, TX, 2004.
  • Jan 2001
  • Disease Center
  • Control
Center for Disease Control, Morbidity and Mortality Weekly Report, CDC, Atlanta, GA, 2001.