Enhancement Of Finger State Progress Model for Markerless Virtual Fine Motor Stroke Rehabilitation

Abstract

The use of machine learning as a tool for analyzing and pattern extraction from the results is widely applied in various medical applications in stroke rehabilitation. It will help the therapist to make a consistent and precise evaluation for a viable recommendation for an optimal future exercise to improve the patient’s progress. The objective of this study is to produce a prediction model to analyze patient finger rehabilitation progress by comparing four regression classifiers' efficiency. In this study, we proposed an Enhancement of the Finger State Progress (E-FSP) model to produce prediction results of progress and performance which also consists of a markerless VR application using markerless motion sensors and can capture kinematic data through Time-based Simplified Denavit Heartenberg (TSDH) model and measure the results of rehabilitation exercises through the integration of Finger State Progress (FSP) model. 30 patients have undergone rehabilitation sessions using VR applications in the Kuala Nerus Rehabilitation and Hemodialysis Health Organization. The study shows the result of an optimum evaluation is the RandomForest classifier which has the lowest Mean Absolute Error (MAE) value of 8.26 and Root Mean Square Error (RMSE) value of 12.38. In conclusion, The VR application and machine learning can produce a very promising combination of attractive visual and viable prediction analysis for virtual fine motor stroke rehabilitation.