Estimating upper-extremity function from kinematics in stroke patients following goal-oriented computer-based training

 Why are we estimating? The idea is to create EXACT PROTOCOLS delivering EXACT RESULTS based upon an objective damage diagnosis. Survivors who can do this are much higher functioning that me, what will bring back my upper limb?

Estimating upper-extremity function from kinematics in stroke patients following goal-oriented computer-based training

• Belén Rubio Ballester,
• Fabrizio Antenucci,
• Martina Maier,
• Anthony C. C. Coolen &
• Paul F. M. J. Verschure 

Journal of NeuroEngineering and Rehabilitation volume 18, Article number: 186 (2021) Cite this article

• 10 Accesses

• Metrics details

Abstract

Introduction

After a stroke, a wide range of deficits can occur with varying onset latencies. As a result, assessing impairment and recovery are enormous challenges in neurorehabilitation. Although several clinical scales are generally accepted, they are time-consuming, show high inter-rater variability, have low ecological validity, and are vulnerable to biases introduced by compensatory movements and action modifications. Alternative methods need to be developed for efficient and objective assessment. In this study, we explore the potential of computer-based body tracking systems and classification tools to estimate the motor impairment of the more affected arm in stroke patients.

Methods

We present a method for estimating clinical scores from movement parameters that are extracted from kinematic data recorded during unsupervised computer-based rehabilitation sessions. We identify a number of kinematic descriptors that characterise the patients’ hemiparesis (e.g., movement smoothness, work area), we implement a double-noise model and perform a multivariate regression using clinical data from 98 stroke patients who completed a total of 191 sessions with RGS.

Results

Our results reveal a new digital biomarker of arm function, the Total Goal-Directed Movement (TGDM), which relates to the patients work area during the execution of goal-oriented reaching movements. The model’s performance to estimate FM-UE scores reaches an accuracy of R2

: 0.38 with an error (σ: 12.8). Next, we evaluate its reliability ( for test-retest), longitudinal external validity (95% true positive rate), sensitivity, and generalisation to other tasks that involve planar reaching movements (R2: 0.39). The model achieves comparable accuracy also for the Chedoke Arm and Hand Activity Inventory (R2: 0.40) and Barthel Index (R2

: 0.35).

Conclusions

Our results highlight the clinical value of kinematic data collected during unsupervised goal-oriented motor training with the RGS combined with data science techniques, and provide new insight into factors underlying recovery and its biomarkers.

Introduction

Stroke is the second major cause of death and disability worldwide, with about 15 million new cases every year [1]. One-third of these cases lead to persistent cognitive and motor disabilities [2]. About 80% of stroke survivors present weakness and partial loss of voluntary control in the upper-extremities [3], or hemiparesis, which is often associated with other sensorimotor alterations, such as hypertonia or tremor.

Although hemiparesis is a highly prevalent symptom and severely limits the independence of affected patients, its causes and recovery dynamics are not fully understood [4]. Recent literature converges on the idea that recovery is mainly due to a combination of residual corticospinal tract capacity and an upregulation of the reticulospinal tract [5, 6]. Further, recovery seems to follow a temporal structure where most of the improvement occurs during the first months post-stroke [7, 8]. So far the assessment of the hemiparesis phenotype and its progression, however, are based on assessment methods with known limitations (e.g. Fugl-Meyer Assessment [9, 10], Action Research Arm Test [11, 12]) and there is a need for more sensitive, objective, and reliable alternatives that are also compatible with contemporary digital health technologies.

A recent systematic review [13] of a total of 225 studies (N = 6197) using 151 different kinematic metrics found that kinematic assessments of upper limb sensorimotor function are poorly standardised and rarely measure clinimetrics in an unbiased manner. Specifically, using descriptors of accuracy, efficacy, efficiency, movement planning, precision, spatial posture, speed, temporal posture, and range of movement together with clinimetric properties of these descriptors (i.e., reliability, measurement error, convergent validity, and external validity), the authors showed that the studies analysed exclusively focused on finding correlations between measures of impairment, and only two of the studies reported correlations in change. Overall, there is very limited information regarding test-retest reliability and the external validity of the change of kinematic outcome measures of reaching performance [14]. Exceptionally, Murphy et al. [15] explored external validity of the change in a number of kinematic descriptors and found a significant covariation of the Action Research Arm Test (ARAT) scores with movement time (R2= 0.36), smoothness (R2 = 0.31), and trunk displacement (R2= 0.35). Although the results are promising, this study involves a limited number of subjects (N = 24) from a highly homogeneous sample (i.e., acute patients only). Further, the ARAT clinical scale presents poor robustness to compensation and is especially vulnerable to the use of explicit strategies to improve performance. Majeed et al. [16] explored the application of models based on LASSO regression to predict changes in motor ability (FM-UE) and motor function (Wolf Motor Function Test, WMFT). These models proposed that recovery in both scales can be approximated by the patient’s age, the patient’s motor control during the execution of fast movements, and other demographic and clinical features, altogether accounting for 65% and 86% of the variability for the FM-UE and WMFT scales respectively. Although these models reached exceptional accuracy, their utility is limited because they make use of kinematic data obtained during the execution of very specific pointing movements supervised by clinicians and/or researchers, and are based on generic unbounded linear models, with the consequence that their estimated values could be largely outside the meaningful range of the scale.

We propose a new approach towards using kinematic data obtained in unsupervised rehabilitation sessions to estimate the level of impairment and functional recovery. Data is obtained from patients engaging with goal-oriented embodied individualised training with the Rehabilitation Gaming System (RGS) [17, 18]. The RGS combines the paradigm of action execution with that of observation of the corresponding movement in Virtual Reality (VR), this goal is achieved by having the patients perform tasks from a first-person perspective, where the movement of their limbs are captured by a camera or a depth sensor (i.e. Microsoft Kinect) and mapped to an analogous virtual representation on a computer screen. RGS includes individualisation mechanisms to adjust the difficulty of the task to the capabilities of the patient, contextual restrictions, and explicit and implicit feedback.

We first explore the potential of hand movements collected during unsupervised RGS sessions to characterise hemiparesis in stroke patients. Secondly, we build and analyse the performance (i.e., test-retest reliability, validity, sensitivity, and generalisation) [13] of a model for estimating impairment and recovery scores captured by standardised clinical scales.