Development and Validation of Assessment Tools Using Robotic and Virtual Reality Technologies in Stroke Rehabilitation

Abstract

Introduction: Stroke is a leading cause of disabilities worldwide. It is important to recover the functional independence of patients after stroke to decrease socioeconomic burdens. Rehabilitation is necessary as an integral part of stroke care in modern medicine. Among many components in stroke rehabilitation, assessment is important for planning current function-based rehabilitation and monitoring the recovery. Recently, technologies such as robotics and virtual realities are actively applied to stroke rehabilitation. These two technologies could be used to objectively assess rehabilitation interventions, increase the reliability of preexisting assessment tools, monitor patients function remotely, and assess their functional independence more quantitatively. The objective of this study was to determine the usefulness of robotics and virtual realities-based technologies for assessments in stroke rehabilitation. Methods: Four experiments were performed in this study. In the first experiment, robotic elbow device was used to investigate the differences between isotonic and isokinetic elbow extension exercises. Nine stroke patients performed three sets of isotonic elbow extensions at 30% of their maximal voluntary isometric torque followed by three sets of maximal isokinetic elbow extensions. The mean angular velocity and total amount of work were standardized for each matched set in two strengthening modes. All exercises were performed using 1-DoF planner robot to regulate the exact resistive torque and speed. Surface electromyographic activities of eight muscles in the hemiplegic shoulder and elbow were recorded. Normalized root mean square (RMS) values and co-contraction index (CCI) were used for the analysis. Robotic elbow device was used to increase the reliability of modified Tardieu Scale (MTS) in the second experiment. Two independent raters measured the catch angle three times per assessor. The isokinetic robotic device was then used to measure the catch angle at velocity of 200 degree/s. Inter- and intra-rater reliabilities were calculated both manually and with robotic assessments. Box and Block Test (BBT), a conventional assessment tool, is simple and easy to apply. Its usefulness in stroke rehabilitation has been demonstrated in previous studies. In the third study, we developed a virtual BBT (VBBT) and investigated its validity for stroke patients. Using a conventional depth-sensing camera, we developed an assessment system for hand, finger, and grasping. The BBT used for grasping ability test in hospitals was virtualized with the same setting. VBBT was validated in patients with mild hemiplegia after stroke. In the fourth experiment, we developed a Fugl-Meyer Assessment (FMA) tool using Kinect (Microsoft, USA) and validated it for hemiplegic stroke patients. A total of 41 patients with hemiplegic stroke were enrolled. Thirteen of 33 items were selected for upper extremity motor FMA. One occupational therapist assessed motor FMA while upper extremity motion was recorded with Kinect. FMA score was calculated using principal component analysis and artificial neural network learning from the saved motion data. The degree of jerky motion was transformed to jerky scores. Prediction accuracy of each of the 13 items was determined. Correlations between real FMA scores and scores using Kinect were analyzed. Results: The isokinetic mode was shown to activate agonists of elbow extension more efficiently than the isotonic mode (normalized RMS for pooled triceps: 96.0±17.0 (2nd) and 87.8±14.4 (3rd) in isokinetic vs. 80.9±11.0 (2nd) and 81.6±12.4 (3rd) in isotonic contraction, F[1,8] = 11.168

P = 0.010) without increasing the co-contraction of muscle pairs, implicating spasticity or synergy. The test-retest and inter-rater reliabilities for measuring catch angle (R1) using isokinetic robotic devices were extremely excellent. The number of blocks moved in the BBT showed strong correlation with the VBBT in the non-hemiplegic side (Pearsons r =0.904, P=0.001) and the hemiplegic side (Pearsons r =0.788, P=0.012). With regard to prediction of FMA using Kinect, the prediction accuracies ranged from 65% to 87% in each item and more than 70% for 9 items. For the summed score of the 13 items, real FMA scores were highly correlated with scores obtained using Kinect (Pearsons r=0.873, P<0.0001). Total upper extremity scores (66 in full score) were also highly correlated with scores using Kinect (26 in full score) (Pearsons r=0.799, P<0.0001). Log transformed jerky scores in the hemiplegic side (1.81 ± 0.76) were significantly higher than those in the non-hemiplegic side (1.21 ± 0.43). Log transformed jerky scores showed significant negative correlations with Brunnstrom stage (3 to 6

Spearman correlation coefficient= -0.387, P =0.046). Conclusions: Robotic device is useful for comparing different modes of rehabilitation in more standardized manner to increase the reliability of preexisting assessment tool. Virtual reality technology with depth-sensing camera can be used as a useful and inexpensive tele-assessment tool for measuring post-stroke motor function in a home-based setting to provide quantitative measures of motion for stroke patients. However, assessment tools used in our study have some limitations. Further efforts are needed to increase the fidelity of rehabilitation care by combining different current technologies or developing novel technologies.