Wearable Sensor to Monitor Quality of Upper Limb Task Practice for Stroke Survivors at Home.

Many stroke survivors experience persistent upper extremity impairment that limits performance in activities of daily living. Upper limb recovery requires high repetitions of task-specific practice. Stroke survivors are often prescribed task practices at home to supplement rehabilitation therapy. A poor quality of task practices, such as the use of compensatory movement patterns, results in maladaptive neuroplasticity and suboptimal motor recovery. There currently lacks a tool for the remote monitoring of movement quality of stroke survivors' task practices at home. The objective of this study was to evaluate the feasibility of classifying movement quality at home using a wearable IMU. Nineteen stroke survivors wore an IMU sensor on the paretic wrist and performed four functional upper limb tasks in the lab and later at home while videorecording themselves. The lab data served as reference data to classify home movement quality using dynamic time warping. Incorrect and correct movement quality was labeled by a therapist. The home task practice movement quality was classified with an accuracy of 92% and F1 score of 0.95 for all tasks combined. Movement types contributing to misclassification were further investigated. The results support the feasibility of a home movement quality monitoring system to assist with upper limb rehabilitation post stroke.

White Matter Integrity and Chronic Poststroke Upper Limb Function: An ENIGMA Stroke Recovery Analysis.

BACKGROUND: Integrity of the corticospinal tract (CST) is an important biomarker for upper limb motor function following stroke. However, when structurally compromised, other tracts may become relevant for compensation or recovery of function. METHODS: We used the ENIGMA Stroke Recovery data set, a multicenter, retrospective, and cross-sectional collection of patients with upper limb impairment during the chronic phase of stroke to test the relevance of tracts in individuals with less and more severe (laterality index of CST fractional anisotropy ≥0.25) CST damage in an observational study design. White matter integrity was quantified using fractional anisotropy for the CST, the superior longitudinal fascicle, and the callosal fibers interconnecting the primary motor cortices between hemispheres. Optic radiations served as a control tract as they have no a priori relevance for the motor system. Pearson correlation was used for testing correlation with upper limb motor function (Fugl-Meyer upper extremity). RESULTS: From 1235 available data sets, 166 were selected (by imaging, Fugl-Meyer upper extremity, covariates, stroke location, and stage) for analyses. Only individuals with severe CST damage showed a positive association of fractional anisotropy in both callosal fibers interconnecting the primary motor cortices (r[21]=0.49; P=0.025) and superior longitudinal fascicle (r[21]=0.51; P=0.018) with Fugl-Meyer upper extremity. CONCLUSIONS: Our data support the notion that individuals with more severe damage of the CST depend on residual pathways for achieving better upper limb outcome than those with less affected CST.

Associating Functional Neural Connectivity and Specific Aspects of Sensorimotor Control in Chronic Stroke.

Hand sensorimotor deficits often result from stroke, limiting the ability to perform daily living activities. Sensorimotor deficits are heterogeneous among stroke survivors. Previous work suggests a cause of hand deficits is altered neural connectivity. However, the relationships between neural connectivity and specific aspects of sensorimotor control have seldom been explored. Understanding these relationships is important for developing personalized rehabilitation strategies to improve individual patients' specific sensorimotor deficits and, thus, rehabilitation outcomes. Here, we investigated the hypothesis that specific aspects of sensorimotor control will be associated with distinct neural connectivity in chronic stroke survivors. Twelve chronic stroke survivors performed a paretic hand grip-and-relax task while EEG was collected. Four aspects of hand sensorimotor grip control were extracted, including reaction time, relaxation time, force magnitude control, and force direction control. EEG source connectivity in the bilateral sensorimotor regions was calculated in α and ß frequency bands during grip preparation and execution. Each of the four hand grip measures was significantly associated with a distinct connectivity measure. These results support further investigations into functional neural connectivity signatures that explain various aspects of sensorimotor control, to assist the development of personalized rehabilitation that targets the specific brain networks responsible for the individuals' distinct sensorimotor deficits.

Application of Deep Learning Algorithm to Monitor Upper Extremity Task Practice.

Upper extremity hemiplegia is a serious problem affecting the lives of many people post-stroke. Motor recovery requires high repetitions and quality of task-specific practice. Sufficient practice cannot be completed during therapy sessions, requiring patients to perform additional task practices at home on their own. Adherence to and quality of these home task practices are often limited, which is likely a factor reducing rehabilitation effectiveness post-stroke. However, home adherence is typically measured by self-reports that are known to be inconsistent with objective measurement. The objective of this study was to develop algorithms to enable the objective identification of task type and quality. Twenty neurotypical participants wore an IMU sensor on the wrist and performed four representative tasks in prescribed fashions that mimicked correct, compensatory, and incomplete movement qualities typically seen in stroke survivors. LSTM classifiers were trained to identify the task being performed and its movement quality. Our models achieved an accuracy of 90.8% for task identification and 84.9%, 81.1%, 58.4%, and 73.2% for movement quality classification for the four tasks for unseen participants. The results warrant further investigation to determine the classification performance for stroke survivors and if quantity and quality feedback from objective monitoring facilitates effective task practice at home, thereby improving motor recovery.

The ENIGMA Stroke Recovery Working Group: Big data neuroimaging to study brain-behavior relationships after stroke.

The goal of the Enhancing Neuroimaging Genetics through Meta-Analysis (ENIGMA) Stroke Recovery working group is to understand brain and behavior relationships using well-powered meta- and mega-analytic approaches. ENIGMA Stroke Recovery has data from over 2,100 stroke patients collected across 39 research studies and 10 countries around the world, comprising the largest multisite retrospective stroke data collaboration to date. This article outlines the efforts taken by the ENIGMA Stroke Recovery working group to develop neuroinformatics protocols and methods to manage multisite stroke brain magnetic resonance imaging, behavioral and demographics data. Specifically, the processes for scalable data intake and preprocessing, multisite data harmonization, and large-scale stroke lesion analysis are described, and challenges unique to this type of big data collaboration in stroke research are discussed. Finally, future directions and limitations, as well as recommendations for improved data harmonization through prospective data collection and data management, are provided.

Use of an EMG-Controlled Game as a Therapeutic Tool to Retrain Hand Muscle Activation Patterns Following Stroke: A Pilot Study.

BACKGROUND/PURPOSE: To determine the feasibility of training with electromyographically (EMG) controlled games to improve control of muscle activation patterns in stroke survivors. METHODS: Twenty chronic stroke survivors (>6 months) with moderate hand impairment were randomized to train either unilaterally (paretic only) or bilaterally over 9 one-hour training sessions. EMG signals from the unilateral or bilateral limbs controlled a cursor location on a computer screen for gameplay. The EMG muscle activation vector was projected onto the plane defined by the first 2 principal components of the activation workspace for the nonparetic hand. These principal components formed the x- and y-axes of the computer screen. RESULTS: The recruitment goal (n = 20) was met over 9 months, with no screen failure, no attrition, and 97.8% adherence rate. After training, both groups significantly decreased the time to move the cursor to a novel sequence of targets (P = 0.006) by reducing normalized path length of the cursor movement (P = 0.005), and improved the Wolf Motor Function Test (WMFT) quality score (P = 0.01). No significant group difference was observed. No significant change was seen in the WMFT time or Box and Block Test. DISCUSSION/CONCLUSIONS: Stroke survivors could successfully use the EMG-controlled games to train control of muscle activation patterns. While the nonparetic limb EMG was used in this study to create target EMG patterns, the system supports various means for creating target patterns per user desires. Future studies will employ training with the EMG-controlled games in conjunction with functional task practice for a longer intervention duration to improve overall hand function.Video Abstract available for more insights from the authors (see the Video, Supplemental Digital Content 1, available at: http://links.lww.com/JNPT/A379).

Use of imperceptible wrist vibration to modulate sensorimotor cortical activity.

Peripheral sensory stimulation has been used as a method to stimulate the sensorimotor cortex, with applications in neurorehabilitation. To improve delivery modality and usability, a new stimulation method has been developed in which imperceptible random-frequency vibration is applied to the wrist concurrently during hand activity. The objective of this study was to investigate effects of this new sensory stimulation on the sensorimotor cortex. Healthy adults were studied. In a transcranial magnetic stimulation (TMS) study, resting motor threshold, short-interval intracortical inhibition, and intracortical facilitation for the abductor pollicis brevis muscle were compared between vibration on vs. off, while subjects were at rest. In an electroencephalogram (EEG) study, alpha and beta power during rest and event-related desynchronization (ERD) for hand grip were compared between vibration on vs. off. Results showed that vibration decreased EEG power and decreased TMS short-interval intracortical inhibition (i.e., disinhibition) compared with no vibration at rest. Grip-related ERD was also greater during vibration, compared to no vibration. In conclusion, subthreshold random-frequency wrist vibration affected the release of intracortical inhibition and both resting and grip-related sensorimotor cortical activity. Such effects may have implications in rehabilitation.

Capturing Upper Limb Gross Motor Categories Using the Kinect® Sensor.

IMPORTANCE: Along with growth in telerehabilitation, a concurrent need has arisen for standardized methods of tele-evaluation. OBJECTIVE: To examine the feasibility of using the Kinect sensor in an objective, computerized clinical assessment of upper limb motor categories. DESIGN: We developed a computerized Mallet classification using the Kinect sensor. Accuracy of computer scoring was assessed on the basis of reference scores determined collaboratively by multiple evaluators from reviewing video recording of movements. In addition, using the reference score, we assessed the accuracy of the typical clinical procedure in which scores were determined immediately on the basis of visual observation. The accuracy of the computer scores was compared with that of the typical clinical procedure. SETTING: Research laboratory. PARTICIPANTS: Seven patients with stroke and 10 healthy adult participants. Healthy participants intentionally achieved predetermined scores. OUTCOMES AND MEASURES: Accuracy of the computer scores in comparison with accuracy of the typical clinical procedure (immediate visual assessment). RESULTS: The computerized assessment placed participants' upper limb movements in motor categories as accurately as did typical clinical procedures. CONCLUSIONS AND RELEVANCE: Computerized clinical assessment using the Kinect sensor promises to facilitate tele-evaluation and complement telehealth applications. WHAT THIS ARTICLE ADDS: Computerized clinical assessment can enable patients to conduct evaluations remotely in their homes without therapists present.

Effects of Gloves and Pulling Task on Achievable Downward Pull Forces on a Rung.

Objective We examined the impacts of pulling task (breakaway and pull-down tasks at different postures), glove use, and their interaction on achievable downward pull forces from a ladder rung. Background Posture, glove use, and the type of pulling task are known to affect the achievable forces. However, a gap in the literature exists regarding how these factors affect achievable downward pulling forces, which are relevant to recovery from a perturbation during ladder climbing. Methods Forty subjects completed four downward pulling tasks (breakaway force; pull force at maximum height, shoulder height, and a middle height), using three glove conditions with varying coefficient of friction (COF) levels (cotton glove, low COF; bare hand, moderate COF; and latex-coated glove, high COF) with their dominant and nondominant hand. The outcome variable was the maximum force normalized to body weight. Results The highest forces were observed for the highest hand postures (breakaway and maximum height). Increased COF led to higher forces and had a larger effect on breakaway force than the other tasks. The dominant hand was associated with higher forces than the nondominant hand. Male subjects generated greater forces than female subjects, particularly for higher hand positions. Conclusion This study suggests that a higher hand position on the ladder, while avoiding low-friction gloves, may be effective for improving recovery from ladder perturbations. Application This study may guide preferred climbing strategies (particularly those that lead to a higher hand position) for improving recovery from a perturbation during ladder climbing.

Delayed grip relaxation and altered modulation of intracortical inhibition with aging.

Grip relaxation is a voluntary action that requires an increase in short-interval intracortical inhibition (SICI) in healthy young adults, rather than a simple termination of excitatory drive. The way aging affects this voluntary inhibitory action and timing of grip relaxation is currently unknown. The objective of this study was to examine aging-related delays in grip relaxation and SICI modulation for the flexor digitorum superficialis muscle during grip relaxation. The main finding was that young adults increased SICI to relax their grips, whereas older adults did not increase SICI with a prolonged grip relaxation time (p < 0.05 for both SICI modulation and grip relaxation time). A secondary experiment showed that both young and older adults did not change H reflex excitability during grip relaxation. Our data suggest that grip relaxation is mediated by increased cortical inhibitory output in young adults, and aging-related impairment in increasing cortical inhibitory output may hamper timely cessation of muscle activity. Our data also suggest a lesser role of the spinal circuits in grip muscle relaxation. This knowledge may contribute to understanding of aging-related movement deterioration and development of interventions for improving modulation of SICI to improve muscle relaxation and movement coordination.

Modifying Kinect placement to improve upper limb joint angle measurement accuracy.

STUDY DESIGN: Repeated measures. INTRODUCTION: The Kinect (Microsoft, Redmond, WA) is widely used for telerehabilitation applications including rehabilitation games and assessment. PURPOSE OF THE STUDY: To determine effects of the Kinect location relative to a person on measurement accuracy of upper limb joint angles. METHODS: Kinect error was computed as difference in the upper limb joint range of motion (ROM) during target reaching motion, from the Kinect vs 3D Investigator Motion Capture System (NDI, Waterloo, Ontario, Canada), and compared across 9 Kinect locations. RESULTS: The ROM error was the least when the Kinect was elevated 45° in front of the subject, tilted toward the subject. This error was 54% less than the conventional location in front of a person without elevation and tilting. The ROM error was the largest when the Kinect was located 60° contralateral to the moving arm, at the shoulder height, facing the subject. The ROM error was the least for the shoulder elevation and largest for the wrist angle. DISCUSSION: Accuracy of the Kinect sensor for detecting upper limb joint ROM depends on its location relative to a person. CONCLUSION: This information facilitates implementation of Kinect-based upper limb rehabilitation applications with adequate accuracy. LEVEL OF EVIDENCE: 3b.

Altered phalanx force direction during power grip following stroke.

Many stroke survivors with severe impairment can grasp only with a power grip. Yet, little knowledge is available on altered power grip after stroke, other than reduced power grip strength. This study characterized stroke survivors' static power grip during 100 and 50 % maximum grip. Each phalanx force angular deviation from the normal direction and its contribution to total normal force was compared for 11 stroke survivors and 11 age-matched controls. Muscle activities and skin coefficient of friction were additionally compared for another 20 stroke and 13 age-matched control subjects. The main finding was that stroke survivors gripped with a 34 % greater phalanx force angular deviation of 19° ± 2° compared to controls of 14° ± 1° (p < .05). Stroke survivors' phalanx force angular deviation was closer to the 23° threshold of slippage between the phalanx and grip surface, which may explain increased likelihood of object dropping in stroke survivors. In addition, this altered phalanx force direction decreases normal grip force by tilting the force vector, indicating a partial role of phalanx force angular deviation in reduced grip strength post-stroke. Greater phalanx force angular deviation may biomechanically result from more severe underactivation of stroke survivors' first dorsal interosseous and extensor digitorum communis muscles compared to their flexor digitorum superficialis or somatosensory deficit. While stroke survivors' maximum power grip strength was approximately half of the controls, the distribution of their remaining strength over the fingers and phalanges did not differ, indicating evenly distributed grip force reduction over the entire hand.

Cortical Sensorimotor Integration as a Neural Origin of Impaired Grip Force Direction Control following Stroke.

BACKGROUND: Stroke is a major cause of disability worldwide. Upper limb impairment is prevalent after stroke. One of the post-stroke manifestations is impaired grip force directional control contributing to diminished abilities to grip and manipulate objects necessary for activities of daily living. The objective of this study was to investigate the neural origin of the impaired grip force direction control following stroke. Due to the importance of online adjustment of motor output based on sensory feedback, it was hypothesized that grip force direction control would be associated with cortical sensorimotor integration in stroke survivors. METHODS: Ten chronic stroke survivors participated in this study. Cortical sensorimotor integration was quantified by short latency afferent inhibition (SAI), which represents the responsiveness of the primary motor cortex to somatosensory input. Grip force direction control was assessed during paretic grip. RESULTS: Grip force direction control was significantly associated with SAI. This relationship was independent of sensory impairment level. CONCLUSIONS: Cortical sensorimotor integration may play a significant role in the grip force direction control important for gripping and manipulating objects with the affected hand following stroke. This knowledge may be used to inform personalized rehabilitation treatment. For example, for patients with impaired grip force direction control, behavioral therapy focusing on feedback motor control, augmented by use of brain stimulation to reinforce cortical sensorimotor integration such as paired associative stimulation, may be applied.

Remote vibrotactile noise improves light touch sensation in stroke survivors' fingertips via stochastic resonance.

BACKGROUND AND PURPOSE: Stroke rehabilitation does not often integrate both sensory and motor recovery. While subthreshold noise was shown to enhance sensory signal detection at the site of noise application, having a noise-generating device at the fingertip to enhance fingertip sensation and potentially enhance dexterity for stroke survivors is impractical, since the device would interfere with object manipulation. This study determined if remote application of subthreshold vibrotactile noise (away from the fingertips) improves fingertip tactile sensation with potential to enhance dexterity for stroke survivors. METHODS: Index finger and thumb pad sensation was measured for ten stroke survivors with fingertip sensory deficit using the Semmes-Weinstein Monofilament and Two-Point Discrimination Tests. Sensation scores were measured with noise applied at one of three intensities (40%, 60%, 80% of the sensory threshold) to one of four locations of the paretic upper extremity (dorsal hand proximal to the index finger knuckle, dorsal hand proximal to the thumb knuckle, dorsal wrist, volar wrist) in a random order, as well as without noise at beginning (Pre) and end (Post) of the testing session. RESULTS: Vibrotactile noise of all intensities and locations instantaneously and significantly improved Monofilament scores of the index fingertip and thumb tip (p < .01). No significant effect of the noise was seen for the Two-Point Discrimination Test scores. CONCLUSIONS: Remote application of subthreshold (imperceptible) vibrotactile noise at the wrist and dorsal hand instantaneously improved stroke survivors' light touch sensation, independent of noise location and intensity. Vibrotactile noise at the wrist and dorsal hand may have enhanced the fingertips' light touch sensation via stochastic resonance and interneuronal connections. While long-term benefits of noise in stroke patients warrants further investigation, this result demonstrates potential that a wearable device applying vibrotactile noise at the wrist could enhance sensation and grip ability without interfering with object manipulation in everyday tasks.

Involuntary contralateral upper extremity muscle activation pattern during unilateral pinch grip following stroke.

STUDY DESIGN: Repeated measures design. INTRODUCTION: Mirror activity refers to emergence of motion not only in the intended, but also in the contralateral limb. PURPOSE: To characterize post-stroke mirror activities across multiple muscles during unilateral pinch. METHODS: Chronic stroke survivors performed unilateral pinch grip using the paretic and nonparetic hand, while four muscles' EMGs were recorded for both hands. RESULTS: During the paretic hand grip, the relaxed nonparetic hand showed mirror activity that was more pronounced for the intrinsic (FDI and thenar eminence) than extrinsic muscles (FDS and EDC). During the nonparetic hand grip, mirror activity in the paretic hand was suppressed for the intrinsic than extrinsic muscles. CONCLUSION: Chronic stroke survivors' relaxed hand did not mirror the task hand's muscle activation pattern, but displayed altered muscle activation patterns depending on muscles and sides, possibly mediated by disturbed interhemispheric inhibition and relative reliance on corticospinal tracts among muscles. LEVEL OF EVIDENCE: N/A.

Development and evaluation of an efficient training program to facilitate the adoption of a novel neurorehabilitation device.

Many rehabilitation devices are not adopted by therapists in practice. One major barrier is therapists' limited time and resources to get training. The objective of this study was to develop/evaluate an efficient training program for a novel rehabilitation device. The program was developed based on structured interviews with seven therapists for training preference and composed of asynchronous and in-person trainings following efficient teaching methods. The training program was evaluated for six occupational therapy doctoral students and six licensed therapists in neurorehabilitation practice. Training effectiveness was evaluated in a simulated treatment session in which 3 trainees shifted their roles among therapist applying the device, client, and peer assessor. In results, 11 of the 12 trainees passed the assessment of using the device in simulated treatment sessions. One trainee did not pass because s/he did not plug in the device to charge at the end. The in-person training fit within 1-h lunch break. All trainees perceived that they could effectively use the device in their practice and both asynchronous and in-person training easily fit into their schedule. This project serves as an example for development of an efficient and effective training program for a novel rehabilitation device to facilitate clinical adoption.

Effect of Self-Directed Home Therapy Adherence Combined with TheraBracelet on Poststroke Hand Recovery: A Pilot Study.

Hand impairment is a common consequence of stroke, resulting in long-term disability and reduced quality of life. Recovery may be augmented through self-directed therapy activities at home, complemented by the use of rehabilitation devices such as peripheral sensory stimulation. The objective of this study was to determine the effect of adherence to self-directed therapy and the use of TheraBracelet (subsensory random-frequency vibratory stimulation) on hand function for stroke survivors. In a double-blind, randomized controlled pilot trial, 12 chronic stroke survivors were assigned to a treatment or control group (n = 6/group). All participants were instructed to perform 200 repetitions of therapeutic hand tasks 5 days/week while wearing a wrist-worn device 8 hours/day for 4 weeks. The treatment group received TheraBracelet vibration from the device, while the control group received no vibration. Home task repetition adherence and device wear logs, as well as hand function assessment (Stroke Impact Scale Hand domain), were obtained weekly. Repetition adherence was comparable between groups but varied among participants. Participants wore the device to a greater extent than adhering to completing repetitions. A linear mixed model analysis showed a significant interaction between repetition and group (p = 0.01), with greater adherence resulting in greater hand function change for the treatment group (r = 0.94; R 2 = 0.88), but not for the control group. Secondary analysis revealed that repetition adherence was greater for those with lower motor capacity and greater self-efficacy at baseline. This pilot study suggests that adherence to self-directed therapy at home combined with subsensory stimulation may affect recovery outcomes in stroke survivors. This trial is registered with NCT04026399.

Effect of Using TheraBracelet on Grasping versus Reaching in Poststroke Rehabilitation.

A peripheral sensory stimulation named TheraBracelet has recently been shown to have a potential to improve gross manual dexterity following stroke. Upper limb function requires both reach and grasp. It is unknown whether TheraBracelet affects one more than other. The objective of this study was to determine whether TheraBracelet improves reaching versus grasping. In a pilot randomized controlled trial, persons with stroke received TheraBracelet (treatment) or no stimulation (control) during task practice therapy (n = 6/group). Effects of TheraBracelet on reaching versus grasping were determined using breakdown of movement times in the Box and Block Test video recordings. Improvements in movement times for the treatment compared with control group were more pronounced for grasping than for reaching at both post and follow-up time points. TheraBracelet may be beneficial for persons with grasping deficits. This knowledge can guide clinicians for targeted use of TheraBracelet, resulting in effective implementation of the new treatment.

Different aspects of hand grip performance associated with structural connectivity of distinct sensorimotor networks in chronic stroke.

Knowledge regarding the neural origins of distinct upper extremity impairments may guide the choice of interventions to target neural structures responsible for specific impairments. This cross-sectional pilot study investigated whether different brain networks explain distinct aspects of hand grip performance in stroke survivors. In 22 chronic stroke survivors, hand grip performance was characterized as grip strength, reaction, relaxation times, and control of grip force magnitude and direction. In addition, their brain structural connectomes were constructed from diffusion tensor MRI. Prominent networks were identified based on a two-step factor analysis using the number of streamlines among brain regions relevant to sensorimotor function. We used regression models to estimate the predictive value of sensorimotor network connectivity for hand grip performance measures while controlling for stroke lesion volumes. Each hand grip performance measure correlated with the connectivity of distinct brain sensorimotor networks. These results suggest that different brain networks may be responsible for different aspects of hand grip performance, which leads to varying clinical presentations of upper extremity impairment following stroke. Understanding the brain network correlates for different hand grip performances may facilitate the development of personalized rehabilitation interventions to directly target the responsible brain network for specific impairments in individual patients, thus improving outcomes.

Data-driven biomarkers outperform theory-based biomarkers in predicting stroke motor outcomes.

Chronic motor impairments are a leading cause of disability after stroke. Previous studies have predicted motor outcomes based on the degree of damage to predefined structures in the motor system, such as the corticospinal tract. However, such theory-based approaches may not take full advantage of the information contained in clinical imaging data. The present study uses data-driven approaches to predict chronic motor outcomes after stroke and compares the accuracy of these predictions to previously-identified theory-based biomarkers. Using a cross-validation framework, regression models were trained using lesion masks and motor outcomes data from 789 stroke patients (293 female/496 male) from the ENIGMA Stroke Recovery Working Group (age 64.9±18.0 years; time since stroke 12.2±0.2 months; normalised motor score 0.7±0.5 (range [0,1]). The out-of-sample prediction accuracy of two theory-based biomarkers was assessed: lesion load of the corticospinal tract, and lesion load of multiple descending motor tracts. These theory-based prediction accuracies were compared to the prediction accuracy from three data-driven biomarkers: lesion load of lesion-behaviour maps, lesion load of structural networks associated with lesion-behaviour maps, and measures of regional structural disconnection. In general, data-driven biomarkers had better prediction accuracy - as measured by higher explained variance in chronic motor outcomes - than theory-based biomarkers. Data-driven models of regional structural disconnection performed the best of all models tested (R2 = 0.210, p < 0.001), performing significantly better than predictions using the theory-based biomarkers of lesion load of the corticospinal tract (R2 = 0.132, p< 0.001) and of multiple descending motor tracts (R2 = 0.180, p < 0.001). They also performed slightly, but significantly, better than other data-driven biomarkers including lesion load of lesion-behaviour maps (R2 =0.200, p < 0.001) and lesion load of structural networks associated with lesion-behaviour maps (R2 =0.167, p < 0.001). Ensemble models - combining basic demographic variables like age, sex, and time since stroke - improved prediction accuracy for theory-based and data-driven biomarkers. Finally, combining both theory-based and data-driven biomarkers with demographic variables improved predictions, and the best ensemble model achieved R2 = 0.241, p < 0.001. Overall, these results demonstrate that models that predict chronic motor outcomes using data-driven features, particularly when lesion data is represented in terms of structural disconnection, perform better than models that predict chronic motor outcomes using theory-based features from the motor system. However, combining both theory-based and data-driven models provides the best predictions.