Faculty Bibliography

Introduction: Quantifying upper extremity (UE) motor impairment after stroke is impractical, limiting our ability to tailor rehabilitation training in real time. The current gold-standard measure of impairment, the Fugl-Meyer Assessment (FMA), is time-consuming and requires a trained assessor. The FMA furthermore does not assess functional motions in real-world contexts, which is exactly where we aim our rehabilitation interventions. Here, we took initial steps to develop an approach to automatically quantify UE motor impairment during functional task performance.
Method(s): We studied 51 chronic stroke patients (28F:23M; 57.7 (21.3-84.3) years old; 28L:23R paretic; FMA 43.1 (8-65)).We recorded upper body motion with 9 inertial measurement units (IMUs) while patients performed the FMA and a functional task (moving an object on a horizontal 8-target array). We trained a long short-term memory (LSTM) deep learning model to estimate FMA scores from the recorded motion (training set n=40; test set n=11; 4 LSTM layers with between-layer batch normalization; IMU data windows of 4s with slide of 1s). LSTM-generated impairment scores were computed from FMA motions or from functional motions. To ascertain the accuracy of the approach, we calculated the root mean square error (RMSE) and the Spearman correlation coefficient (rho) between the LSTM scores and the FMA scores from a trained expert. We also examined whether the performance of particular classes of functional primitives (i.e. reach, transport, or reposition) would be sufficient to accurately estimate impairment.
Result(s): Using motions from the FMA performance, our approach estimated FMA scores within 1.1 points of a trained assessor. Using motions from the functional task performance, our approach estimated FMA scores within 1.6 points. Correlation values between the FMA scores and LSTM scores were rho = 0.98 for FMA motions and rho = 0.96 for functional motions. Among the three functional primitives, reaches were the most informative for estimating the impairment scores (RMSE: 1.9 points), followed by transports (RMSE: 2.1 points), and repositions (RMSE: 2.8 points).
Discussion(s): We present a new approach that uses sensor-based motion capture and deep learning to automatically estimate UE motor impairment. This approach has high accuracy and shows high concurrent validity with the FMA, even when it assesses unrelated functional motions. Thus, it may be possible to directly measure impairment from performance of real-world functional tasks, which the FMA does not offer. Estimating impairment during stroke rehabilitation would enable clinicians to tailor treatment strategy in real time.

Introduction: Activities of daily living (ADLs) are performed through a sequence of fundamental units of motion, called primitives. We previously observed that during ADLs, one upper extremity (UE) may engage two objects simultaneously, such as turning on a faucet while holding a toothbrush. These dual-object primitives (DOPs) may demand increased neural resources, as they likely entail the simultaneous execution of two motor plans. Skilled movement by the nondominant healthy UE or the paretic UE has also been found to require increased neural activity. We posited that performance of DOPs would exceed the neural resources available to the nondominant or paretic side, reducing their performance on these sides. We also predicted that the frequency of DOP performance by the paretic UE would relate to its degree of motor impairment.
Method(s): We studied 19 right-hand dominant healthy subjects (10M:9F; 62.0 +/- 13.6 years) and 43 premorbidly right-hand dominant stroke subjects (23M:20F; 24L:19R paretic; 57.5 +/- 14.5 years; 5.7 +/- 6.5 years post stroke). We evaluated subjects on the UE Fugl-Meyer Assessment (FMA) and videotaped their performance of a feeding and toothbrushing task. We analyzed the videos to extract the incidence and count of DOP performance by each UE. To control for dominance and paresis, we normalized DOP counts to the total number of primitives performed by the UE. We used two-tailed Fisher's Exact tests to compare the incidence of DOPs performed by each UE, and Spearman's correlation to examine the relationship between FMA score and DOP frequency.
Result(s): In healthy subjects, the incidence of DOPs was lower on the nondominant than dominant side (12/19 vs. 19/19; p<0.01). In stroke subjects, the incidence of DOPs was lower on the paretic than nonparetic side (19/43 vs. 43/43; p<0.01). The laterality of paresis did not affect whether that UE would perform DOPs (11/19 dominant paretic vs. 8/24 nondominant paretic; p=0.132). In stroke subjects, lower FMA scores were related to a lower frequency of DOP performance on their paretic UE (rho=0.368, p=0.015).
Discussion(s): Our results suggest that UE laterality and impairment may impact DOP performance in healthy and stroke subjects, respectively. DOPs were less commonly performed by the nondominant UE and the paretic UE, and worse impairment was associated with lower DOP performance. We speculate that engaging two objects simultaneously requires additional neural resources that are unavailable to the nondominant or injured motor network. It is conceivable that the return of DOP performance by the paretic UE may track with the availability of a recovered neural substrate.

Up to two-thirds of stroke survivors experience persistent sensorimotor impairments. Recovery relies on the integrity of spared brain areas to compensate for damaged tissue. Deep grey matter structures play a critical role in the control and regulation of sensorimotor circuits. The goal of this work is to identify associations between volumes of spared subcortical nuclei and sensorimotor behaviour at different timepoints after stroke. We pooled high-resolution T₁-weighted MRI brain scans and behavioural data in 828 individuals with unilateral stroke from 28 cohorts worldwide. Cross-sectional analyses using linear mixed-effects models related post-stroke sensorimotor behaviour to non-lesioned subcortical volumes (Bonferroni-corrected, P < 0.004). We tested subacute (≤90 days) and chronic (≥180 days) stroke subgroups separately, with exploratory analyses in early stroke (≤21 days) and across all time. Sub-analyses in chronic stroke were also performed based on class of sensorimotor deficits (impairment, activity limitations) and side of lesioned hemisphere. Worse sensorimotor behaviour was associated with a smaller ipsilesional thalamic volume in both early (n = 179; d = 0.68) and subacute (n = 274, d = 0.46) stroke. In chronic stroke (n = 404), worse sensorimotor behaviour was associated with smaller ipsilesional putamen (d = 0.52) and nucleus accumbens (d = 0.39) volumes, and a larger ipsilesional lateral ventricle (d = -0.42). Worse chronic sensorimotor impairment specifically (measured by the Fugl-Meyer Assessment; n = 256) was associated with smaller ipsilesional putamen (d = 0.72) and larger lateral ventricle (d = -0.41) volumes, while several measures of activity limitations (n = 116) showed no significant relationships. In the full cohort across all time (n = 828), sensorimotor behaviour was associated with the volumes of the ipsilesional nucleus accumbens (d = 0.23), putamen (d = 0.33), thalamus (d = 0.33) and lateral ventricle (d = -0.23). We demonstrate significant relationships between post-stroke sensorimotor behaviour and reduced volumes of deep grey matter structures that were spared by stroke, which differ by time and class of sensorimotor measure. These findings provide additional insight into how different cortico-thalamo-striatal circuits support post-stroke sensorimotor outcomes.

KEY POINTS/CONCLUSIONS:The corticoreticulospinal tract (CReST) is a descending motor pathway that reorganizes after corticospinal tract (CST) injury in animals. In humans, the pattern of CReST innervation to upper limb muscles has not been carefully examined in healthy individuals or individuals with CST injury. In the present study, we assessed CReST projections to an arm and hand muscle on the same side of the body in healthy and chronic stoke subjects using transcranial magnetic stimulation. We show that CReST connection strength to the muscles differs between healthy and stroke subjects, with stronger connections to the hand than arm in healthy subjects, and stronger connections to the arm than hand in stroke subjects. These results help us better understand CReST innervation patterns in the upper limb, and may point to its role in normal motor function and motor recovery in humans. ABSTRACT/UNASSIGNED:The corticoreticulospinal tract (CReST) is a major descending motor pathway in many animals, but little is known about its innervation patterns in proximal and distal upper extremity muscles in humans. The contralesional CReST furthermore reorganizes after corticospinal tract (CST) injury in animals, but it is less clear whether CReST innervation changes after stroke in humans. We thus examined CReST functional connectivity, connection strength, and modulation in an arm and hand muscle of healthy (n = 15) and chronic stroke (n = 16) subjects. We delivered transcranial magnetic stimulation to the contralesional hemisphere (assigned in healthy subjects) to elicit ipsilateral motor evoked potentials (iMEPs) from the paretic biceps (BIC) and first dorsal interosseous (FDI) muscle. We operationalized CReST functional connectivity as iMEP presence/absence, CReST projection strength as iMEP size and CReST modulation as change in iMEP size by head rotation. We found comparable CReST functional connectivity to the BICs and FDIs in both subject groups. However, the pattern of CReST connection strength to the muscles diverged between groups, with stronger connections to FDIs than BICs in healthy subjects and stronger connections to BICs than FDIs in stroke subjects. Head rotation modulated only FDI iMEPs of healthy subjects. Our findings indicate that the healthy CReST does not have a proximal innervation bias, and its strong FDI connections may have functional relevance to finger individuation. The reversed CReST innervation pattern in stroke subjects confirms its reorganization after CST injury, and its strong BIC connections may indicate upregulation for particular upper extremity muscles or their functional actions.

PURPOSE/UNASSIGNED:To address the gap in the literature and clarify the expanding role of wearable sensor data in stroke rehabilitation, we summarized the methods for upper extremity (UE) sensor-based assessment and sensor-based treatment. MATERIALS AND METHODS/UNASSIGNED:The guideline outlined by the preferred reporting items for systematic reviews and meta-analysis extension for scoping reviews was used to complete this scoping review. Information pertaining to participant demographics, sensory information, data collection, data processing, data analysis, and study results were extracted from the studies for analysis and synthesis. RESULTS/UNASSIGNED:We included 43 articles in the final review. We organized the results into assessment and treatment categories. The included articles used wearable sensors to identify UE functional motion, categorize motor impairment/activity limitation, and quantify real-world use. Wearable sensors were also used to augment UE training by triggering sensory cues or providing instructional feedback about the affected UE. CONCLUSIONS/UNASSIGNED:Sensors have the potential to greatly expand assessment and treatment beyond traditional clinic-based approaches. This capability could support the quantification of rehabilitation dose, the nuanced assessment of impairment and activity limitation, the characterization of daily UE use patterns in real-world settings, and augment UE training adherence for home-based rehabilitation.IMPLICATIONS FOR REHABILITATIONSensor data have been used to assess UE functional motion, motor impairment/activity limitation, and real-world use.Sensor-assisted treatment approaches are emerging, and may be a promising tool to augment UE adherence in home-based rehabilitation.Wearable sensors may extend our ability to objectively assess UE motion beyond supervised clinical settings, and into home and community settings.

A large number of stroke survivors suffer from a significant decrease in upper extremity (UE) function, requiring rehabilitation therapy to boost recovery of UE motion. Assessing the efficacy of treatment strategies is a challenging problem in this context, and is typically accomplished by observing the performance of patients during their execution of daily activities. A more detailed assessment of UE impairment can be undertaken with a clinical bedside test, the UE Fugl-Meyer Assessment, but it fails to examine compensatory movements of functioning body segments that are used to bypass impairment. In this work, we use a graph learning method to build a visualization tool tailored to support the analysis of stroke patients. Called NE-Motion, or Network Environment for Motion Capture Data Analysis, the proposed analytic tool handles a set of time series captured by motion sensors worn by patients so as to enable visual analytic resources to identify abnormalities in movement patterns. Developed in close collaboration with domain experts, NE-Motion is capable of uncovering important phenomena, such as compensation while revealing differences between stroke patients and healthy individuals. The effectiveness of NE-Motion is shown in two case studies designed to analyze particular patients and to compare groups of subjects.

BACKGROUND AND PURPOSE/OBJECTIVE:The brain stem is a complex configuration of small nuclei and pathways for motor, sensory, and autonomic control that are essential for life, yet internal brain stem anatomy is difficult to characterize in living subjects. We hypothesized that the 3D fast gray matter acquisition T1 inversion recovery sequence, which uses a short inversion time to suppress signal from white matter, could improve contrast resolution of brain stem pathways and nuclei with 3T MR imaging. MATERIALS AND METHODS/METHODS:-space to reduce motion; total scan time = 58 minutes). One subject returned for an additional 5-average study that was combined with a previous session to create a highest quality atlas for anatomic assignments. A 1-mm isotropic resolution, 12-minute version, proved successful in a patient with a prior infarct. RESULTS:The fast gray matter acquisition T1 inversion recovery sequence generated excellent contrast resolution of small brain stem pathways in all 3 planes for all 10 subjects. Several nuclei could be resolved directly by image contrast alone or indirectly located due to bordering visualized structures (eg, locus coeruleus and pedunculopontine nucleus). CONCLUSIONS:The fast gray matter acquisition T1 inversion recovery sequence has the potential to provide imaging correlates to clinical conditions that affect the brain stem, improve neurosurgical navigation, validate diffusion tractography of the brain stem, and generate a 3D atlas for automatic parcellation of specific brain stem structures.

Introduction: Rehabilitation training after stroke commonly focuses on practicing activities of daily living (ADLs), comprised of functional movements and, more fundamentally, functional primitives. Animal models have demonstrated extensive motor recovery if many functional movements are trained early after stroke. In humans, the optimal rehabilitation dose to maximize recovery is not known, in part because a tool to precisely but pragmatically measure rehabilitation does not currently exist. We are building a measurement tool that can objectively decompose ADLs into their constituent primitives. We report here developments in the first important step of building this tool-the automatic identification of functional primitives that constitute various ADLs.
Method(s): 32 stroke subjects (gender: 18F/14M; paretic side: 14R/18L; age: 56.2 +/- 13.54 years; time since stroke: 6.7 +/- 7.57 years; mean FuglMeyer score: 44.21 +/- 14.26) performed a battery of 9 ADLs in an inpatient gym. Participants wore 9 inertial measurement units (IMUs) on their cervical spine, thoracic spine, pelvis, and bilateral hands, forearms, and arms. The IMU system generated linear accelerations, orientations, quaternions, and joint angles at 100 Hz. Human coders used synchronously recorded video to segment each activity into its constituent primitives: reach, transport, stabilize, reposition, and idle. This segmentation step also assigned primitive labels to the IMU data. Using labeled IMU data, we trained a sequence-to-sequence convolutional neural network (CNN) in 21 subjects and tested it in 11 subjects. Subjects were chosen randomly and were balanced for paretic side. The model had 14 convolutional layers with batch normalization between each layer to reduce the covariate shift. Data windows of 1 s (with a slide of 0.25 s) were fed into the CNN. Using a softmax activation function, the final layer of the CNN generated the probability of the data sample being each primitive. The winning probability was chosen as the label name. To measure the classification accuracy (positive predictive value, PPV) of the approach, we compared the CNNgenerated label against the human-generated label for all data windows.
Result(s): Our approach had an average classification accuracy of 64% for identifying the five primitives. Its lowest accuracy was in identifying reaches (PPV 37%), which were commonly confused with transports. It was moderately accurate in identifying repositions (PPV 46%), which were also confused with transports. The approach performed well in identifying idles (PPV 67%), stabilizations (PPV 62%), and transports (PPV 60%).
Discussion(s): We present a novel approach for classifying functional primitives embedded in ADLs, an important step toward dose quantitation in rehabilitation. Though classification performance was modest, the approach performs well above chance (PPV 20%), affirming its plausibility for use in stroke patients. Future work will test other deep network architectures and data augmentation techniques to improve classification performance

Recent advances in wearable sensor technology and machine learning (ML) have allowed for the seamless and objective study of human motion in clinical applications, including Parkinson's disease, and stroke. Using ML to identify salient patterns in sensor data has the potential for widespread application in neurological disorders, so understanding how to develop this approach for one's area of inquiry is vital. We previously proposed an approach that combined wearable inertial measurement units (IMUs) and ML to classify motions made by stroke patients. However, our approach had computational and practical limitations. We address these limitations here in the form of a primer, presenting how to optimize a sensor-ML approach for clinical implementation. First, we demonstrate how to identify the ML algorithm that maximizes classification performance and pragmatic implementation. Second, we demonstrate how to identify the motion capture approach that maximizes classification performance but reduces cost. We used previously collected motion data from chronic stroke patients wearing off-the-shelf IMUs during a rehabilitation-like activity. To identify the optimal ML algorithm, we compared the classification performance, computational complexity, and tuning requirements of four off-the-shelf algorithms. To identify the optimal motion capture approach, we compared the classification performance of various sensor configurations (number and location on the body) and sensor type (IMUs vs. accelerometers). Of the algorithms tested, linear discriminant analysis had the highest classification performance, low computational complexity, and modest tuning requirements. Of the sensor configurations tested, seven sensors on the paretic arm and trunk led to the highest classification performance, and IMUs outperformed accelerometers. Overall, we present a refined sensor-ML approach that maximizes both classification performance and pragmatic implementation. In addition, with this primer, we showcase important considerations for appraising off-the-shelf algorithms and sensors for quantitative motion assessment.