A quantitative and standardized robotic method for the evaluation of arm proprioception after stroke

The use of machine learning and deep learning techniques to assess proprioceptive impairments of the upper limb after stroke

BACKGROUND

Robots can generate rich kinematic datasets that have the potential to provide far more insight into impairments than standard clinical ordinal scales. Determining how to define the presence or absence of impairment in individuals using kinematic data, however, can be challenging. Machine learning techniques offer a potential solution to this problem. In the present manuscript we examine proprioception in stroke survivors using a robotic arm position matching task. Proprioception is impaired in 50-60% of stroke survivors and has been associated with poorer motor recovery and longer lengths of hospital stay. We present a simple cut-off score technique for individual kinematic parameters and an overall task score to determine impairment. We then compare the ability of different machine learning (ML) techniques and the above-mentioned task score to correctly classify individuals with or without stroke based on kinematic data.

METHODS

Participants performed an Arm Position Matching (APM) task in an exoskeleton robot. The task produced 12 kinematic parameters that quantify multiple attributes of position sense. We first quantified impairment in individual parameters and an overall task score by determining if participants with stroke fell outside of the 95% cut-off score of control (normative) values. Then, we applied five machine learning algorithms (i.e., Logistic Regression, Decision Tree, Random Forest, Random Forest with Hyperparameters Tuning, and Support Vector Machine), and a deep learning algorithm (i.e., Deep Neural Network) to classify individual participants as to whether or not they had a stroke based only on kinematic parameters using a tenfold cross-validation approach.

RESULTS

We recruited 429 participants with neuroimaging-confirmed stroke (< 35 days post-stroke) and 465 healthy controls. Depending on the APM parameter, we observed that 10.9-48.4% of stroke participants were impaired, while 44% were impaired based on their overall task score. The mean performance metrics of machine learning and deep learning models were: accuracy 82.4%, precision 85.6%, recall 76.5%, and F1 score 80.6%. All machine learning and deep learning models displayed similar classification accuracy; however, the Random Forest model had the highest numerical accuracy (83%). Our models showed higher sensitivity and specificity (AUC = 0.89) in classifying individual participants than the overall task score (AUC = 0.85) based on their performance in the APM task. We also found that variability was the most important feature in classifying performance in the APM task.

CONCLUSION

Our ML models displayed similar classification performance. ML models were able to integrate more kinematic information and relationships between variables into decision making and displayed better classification performance than the overall task score. ML may help to provide insight into individual kinematic features that have previously been overlooked with respect to clinical importance.

Effect of Short-Term Exposure to Supplemental Vibrotactile Kinesthetic Feedback on Goal-Directed Movements after Stroke: A Proof of Concept Case Series

Many survivors of stroke have persistent somatosensory deficits on the contralesional side of their body. Non-invasive supplemental feedback of limb movement could enhance the accuracy and efficiency of actions involving the upper extremity, potentially improving quality of life after stroke. In this proof-of-concept study, we evaluated the feasibility and the immediate effects of providing supplemental kinesthetic feedback to stroke survivors, performing goal-directed actions with the contralesional arm. Three survivors of stroke in the chronic stage of recovery participated in experimental sessions wherein they performed reaching and stabilization tasks with the contralesional arm under different combinations of visual and vibrotactile feedback, which was induced on the ipsilesional arm. Movement kinematics were encoded by a vibrotactile feedback interface in two ways: state feedback—an optimal combination of hand position and velocity; and error feedback—the difference between the actual hand position and its instantaneous target. In each session we evaluated the feedback encoding scheme’s immediate objective utility for improving motor performance as well as its perceived usefulness. All three participants improved their stabilization performance using at least one of the feedback encoding schemes within just one experimental session. Two of the participants also improved reaching performance with one or the other of the encoding schemes. Although the observed beneficial effects were modest in each participant, these preliminary findings show that supplemental vibrotactile kinesthetic feedback can be readily interpreted and exploited to improve reaching and object stabilizing actions performed with the contralesional arm after stroke. These short-term training results motivate a longer multisession training study using personalized vibrotactile feedback as a means to improve the accuracy and efficacy of contralesional arm actions after stroke.

Tactile-STAR: A Novel Tactile STimulator And Recorder System for Evaluating and Improving Tactile Perception

Many neurological diseases impair the motor and somatosensory systems. While several different technologies are used in clinical practice to assess and improve motor functions, somatosensation is evaluated subjectively with qualitative clinical scales. Treatment of somatosensory deficits has received limited attention. To bridge the gap between the assessment and training of motor vs. somatosensory abilities, we designed, developed, and tested a novel, low-cost, two-component (bimanual) mechatronic system targeting tactile somatosensation: the Tactile-STAR-a tactile stimulator and recorder. The stimulator is an actuated pantograph structure driven by two servomotors, with an end-effector covered by a rubber material that can apply two different types of skin stimulation: brush and stretch. The stimulator has a modular design, and can be used to test the tactile perception in different parts of the body such as the hand, arm, leg, big toe, etc. The recorder is a passive pantograph that can measure hand motion using two potentiometers. The recorder can serve multiple purposes: participants can move its handle to match the direction and amplitude of the tactile stimulator, or they can use it as a master manipulator to control the tactile stimulator as a slave. Our ultimate goal is to assess and affect tactile acuity and somatosensory deficits. To demonstrate the feasibility of our novel system, we tested the Tactile-STAR with 16 healthy individuals and with three stroke survivors using the skin-brush stimulation. We verified that the system enables the mapping of tactile perception on the hand in both populations. We also tested the extent to which 30 min of training in healthy individuals led to an improvement of tactile perception. The results provide a first demonstration of the ability of this new system to characterize tactile perception in healthy individuals, as well as a quantification of the magnitude and pattern of tactile impairment in a small cohort of stroke survivors. The finding that short-term training with Tactile-STAR can improve the acuity of tactile perception in healthy individuals suggests that Tactile-STAR may have utility as a therapeutic intervention for somatosensory deficits.

Development of a System Architecture for Evaluation and Training of Proprioceptive Deficits of the Upper Limb

Proprioception plays a fundamental role in maintaining posture and executing movement, and the quantitative evaluation of proprioceptive deficits in poststroke patients is important. But currently it is not widely performed due to the complexity of the evaluation tools required for a reliable assessment. The aims of this pilot study were to (a) develop a system architecture for upper limb evaluation and training of proximal and distal sense of position in the horizontal plane and (b) test the system in healthy and pathological subjects. Two robotic devices for evaluation and training of, respectively, wrist flexion/extension and shoulder-elbow manipulation were employed. The system we developed was applied in a group of 12 healthy subjects and 10 patients after stroke. It was able to quantitatively evaluate upper limb sense of position in the horizontal plane thanks to a set of quantitative parameters assessing position estimation errors, variability, and gain. In addition, it was able to distinguish healthy from pathological conditions. The system could thus be a reliable method to detect changes in the sense of position of patients with sensory deficits after stroke and could enable the implementation of novel training approaches for the recovery of normal proprioception.

Proprioceptive assessment in clinical settings: Evaluation of joint position sense in upper limb post-stroke using a robotic manipulator

Proprioception is a critical component for motor functions and directly affects motor learning after neurological injuries. Conventional methods for its assessment are generally ordinal in nature and hence lack sensitivity. Robotic devices designed to promote sensorimotor learning can potentially provide quantitative precise, accurate, and reliable assessments of sensory impairments. In this paper, we investigate the clinical applicability and validity of using a planar 2 degrees of freedom robot to quantitatively assess proprioceptive deficits in post-stroke participants. Nine stroke survivors and nine healthy subjects participated in the study. Participants’ hand was passively moved to the target position guided by the H-Man robot (Criterion movement) and were asked to indicate during a second passive movement towards the same target (Matching movement) when they felt that they matched the target position. The assessment was carried out on a planar surface for movements in the forward and oblique directions in the contralateral and ipsilateral sides of the tested arm. The matching performance was evaluated in terms of error magnitude (absolute and signed) and its variability. Stroke patients showed higher variability in the estimation of the target position compared to the healthy participants. Further, an effect of target was found, with lower absolute errors in the contralateral side. Pairwise comparison between individual stroke participant and control participants showed significant proprioceptive deficits in two patients. The proposed assessment of passive joint position sense was inherently simple and all participants, regardless of motor impairment level, could complete it in less than 10 minutes. Therefore, the method can potentially be carried out to detect changes in proprioceptive deficits in clinical settings.

A composite robotic-based measure of upper limb proprioception

Background

Proprioception is the sense of the position and movement of our limbs, and is vital for executing coordinated movements. Proprioceptive disorders are common following stroke, but clinical tests for measuring impairments in proprioception are simple ordinal scales that are unreliable and relatively crude. We developed and validated specific kinematic parameters to quantify proprioception and compared two common metrics, Euclidean and Mahalanobis distances, to combine these parameters into an overall summary score of proprioception.

Methods

We used the KINARM robotic exoskeleton to assess proprioception of the upper limb in subjects with stroke (N = 285. Mean days post-stroke = 12 ± 15). Two aspects of proprioception (position sense and kinesthetic sense) were tested using two mirror-matching tasks without vision. The tasks produced 12 parameters to quantify position sense and eight to quantify kinesthesia. The Euclidean and Mahalanobis distances of the z-scores for these parameters were computed each for position sense, kinesthetic sense, and overall proprioceptive function (average score of position and kinesthetic sense).

Results

A high proportion of stroke subjects were impaired on position matching (57%), kinesthetic matching (65%), and overall proprioception (62%). Robotic tasks were significantly correlated with clinical measures of upper extremity proprioception, motor impairment, and overall functional independence. Composite scores derived from the Euclidean distance and Mahalanobis distance showed strong content validity as they were highly correlated (r = 0.97–0.99).

Conclusions

We have outlined a composite measure of upper extremity proprioception to provide a single continuous outcome measure of proprioceptive function for use in clinical trials of rehabilitation. Multiple aspects of proprioception including sense of position, direction, speed, and amplitude of movement were incorporated into this measure. Despite similarities in the scores obtained with these two distance metrics, the Mahalanobis distance was preferred.

Inter-rater reliability of kinesthetic measurements with the KINARM robotic exoskeleton

Background

Kinesthesia (sense of limb movement) has been extremely difficult to measure objectively, especially in individuals who have survived a stroke. The development of valid and reliable measurements for proprioception is important to developing a better understanding of proprioceptive impairments after stroke and their impact on the ability to perform daily activities. We recently developed a robotic task to evaluate kinesthetic deficits after stroke and found that the majority (~60%) of stroke survivors exhibit significant deficits in kinesthesia within the first 10 days post-stroke. Here we aim to determine the inter-rater reliability of this robotic kinesthetic matching task.

Methods

Twenty-five neurologically intact control subjects and 15 individuals with first-time stroke were evaluated on a robotic kinesthetic matching task (KIN). Subjects sat in a robotic exoskeleton with their arms supported against gravity. In the KIN task, the robot moved the subjects’ stroke-affected arm at a preset speed, direction and distance. As soon as subjects felt the robot begin to move their affected arm, they matched the robot movement with the unaffected arm. Subjects were tested in two sessions on the KIN task: initial session and then a second session (within an average of 18.2 ± 13.8 h of the initial session for stroke subjects), which were supervised by different technicians. The task was performed both with and without the use of vision in both sessions. We evaluated intra-class correlations of spatial and temporal parameters derived from the KIN task to determine the reliability of the robotic task.

Results

We evaluated 8 spatial and temporal parameters that quantify kinesthetic behavior. We found that the parameters exhibited moderate to high intra-class correlations between the initial and retest conditions (Range, r-value = [0.53–0.97]).

Conclusions

The robotic KIN task exhibited good inter-rater reliability. This validates the KIN task as a reliable, objective method for quantifying kinesthesia after stroke.

Wrist proprioceptive acuity: A comprehensive robot-aided assessment

Proprioception is the sense of the body awareness. Proprioceptive deficits represent frequent consequences of several neurological conditions like stroke, Parkinson's disease and others. The assessment of such somatosensory function is crucial, although the available clinical tests are not sensitive enough. The human wrist is a crucial joint for many activities of daily living and to address the lack of its characterization in terms of proprioceptive acuity the authors in previous studies proposed a novel method that combined the use of a 3-DoF robot and a threshold haunting paradigm. Further experiments were performed to characterize the proprioceptive acuity of the dominant wrist for adduction, extension, pronation and supination by using a 2-alternative-forced-choice test. The acuity thresholds obtained from six subjects (mean values ± standard deviation of 1.65±0.39 for extension, 1.13±0.34 for adduction, 1.90±0.58 for pronation and 1.70±0.30 for supination) were finally combined with the ones harvested in the previous studies for flexion and abduction in order to build the first comprehensive database of human wrist proprioceptive acuity.

Robot-aided assessment of wrist proprioception

INTRODUCTION

Impaired proprioception severely affects the control of gross and fine motor function. However, clinical assessment of proprioceptive deficits and its impact on motor function has been difficult to elucidate. Recent advances in haptic robotic interfaces designed for sensorimotor rehabilitation enabled the use of such devices for the assessment of proprioceptive function.

PURPOSE

This study evaluated the feasibility of a wrist robot system to determine proprioceptive discrimination thresholds for two different DoFs of the wrist. Specifically, we sought to accomplish three aims: first, to establish data validity; second, to show that the system is sensitive to detect small differences in acuity; third, to establish test-retest reliability over repeated testing.

METHODOLOGY

Eleven healthy adult subjects experienced two passive wrist movements and had to verbally indicate which movement had the larger amplitude. Based on a subject's response data, a psychometric function was fitted and the wrist acuity threshold was established at the 75% correct response level. A subset of five subjects repeated the experimentation three times (T1, T2, and T3) to determine the test-retest reliability.

RESULTS

Mean threshold for wrist flexion was 2.15°± 0.43° and 1.52°± 0.36° for abduction. Encoder resolutions were 0.0075°(flexion-extension) and 0.0032°(abduction-adduction). Motor resolutions were 0.2°(flexion-extension) and 0.3°(abduction-adduction). Reliability coefficients were r T2-T1 = 0.986 and r T3-T2 = 0.971.

CONCLUSION

We currently lack established norm data on the proprioceptive acuity of the wrist to establish direct validity. However, the magnitude of our reported thresholds is physiological, plausible, and well in line with available threshold data obtained at the elbow joint. Moreover, system has high resolution and is sensitive enough to detect small differences in acuity. Finally, the system produces reliable data over repeated testing.

A robotic object hitting task to quantify sensorimotor impairments in participants with stroke

Background

Existing clinical scores of upper limb function often use observer-based ordinal scales that are subjective and commonly have floor and ceiling effects. The purpose of the present study was to develop an upper limb motor task to assess objectively the ability of participants to select and engage motor actions with both hands.

Methods

A bilateral robotic system was used to quantify upper limb sensorimotor function of participants with stroke. Participants performed an object hit task that required them to hit virtual balls moving towards them in the workspace with virtual paddles attached to each hand. Task difficulty was initially low, but increased with time by increasing the speed and number of balls in the workspace. Data were collected from 262 control participants and 154 participants with recent stroke.

Results

Control participants hit ~60 to 90% of the 300 balls with relatively symmetric performance for the two arms. Participants with recent stroke performed the task with most participants hitting fewer balls than 95% of healthy controls (67% of right-affected and 87% of left-affected strokes). Additionally, nearly all participants (97%) identified with visuospatial neglect hit fewer balls than healthy controls. More detailed analyses demonstrated that most participants with stroke displayed asymmetric performance between their affected and non-affected limbs with regards to number of balls hit, workspace area covered by the limb and hand speed. Inter-rater reliability of task parameters was high with half of the correlations above 0.90. Significant correlations were observed between many of the task parameters and the Functional Independence Measure and/or the Behavioural Inattention Test.

Conclusions

As this object hit task requires just over two minutes to complete, it provides an objective and easy approach to quantify upper limb motor function and visuospatial skills following stroke.

The independence of deficits in position sense and visually guided reaching following stroke

Background

Several studies have found correlations between proprioception and visuomotor function during stroke recovery, however two more recent studies have found no correlation. Unfortunately, most of the studies to date have been conducted with clinical assessments of sensation that are observer-based and have poor reliability. We have recently developed new tests to assess position sense and motor function using robotic technology. The present study was conducted to reassess the relationship between position sense and upper limb movement following stroke.

Methods

We assessed position sense and motor performance of 100 inpatient stroke rehabilitation subjects and 231 non-disabled controls. All subjects completed quantitative assessments of position sense (arm-position matching task) and motor performance (visually-guided reaching task) using the KINARM robotic device. Subjects also completed clinical assessments including handedness, vision, Purdue Pegboard, Chedoke-McMaster Stroke Assessment-Impairment Inventory and Functional Independence Measure (FIM). Neuroimaging documented lesion localization. Fisher’s exact probability tests were used to determine the relationship between performances on the arm-position matching and visually-guided reaching task. Pearson’s correlations were conducted to determine the relationship between robotically measured parameters and clinical assessments.

Results

Performance by individual subjects on the matching and reaching tasks was statistically independent (Fisher’s test, P<0.01). However, performance on the matching and reaching tasks both exhibited relationships with abilities in daily activities as measured by the FIM. Performance on the reaching task also displayed strong relationships with other clinical measures of motor impairment.

Conclusions

Our data support the concept that position sense deficits are functionally relevant and point to the importance of assessing proprioceptive and motor impairments independently when planning treatment strategies.