Recognition of Sensory-based Motion Information using Virtual Reality based Rehabilitation Exercises for COVID19 Recovered Patients

Abstract

Exercise therapy is observed as one of the major treatments for rehabilitation of patients, those attacked by any injury, mental disorder, and even Covid-19. Virtual reality in healthcare rehabilitation demands accurate classification of gestures or movements. This study aims to recognize the prescribed rehabilitation exercises of computer-assisted sensory systems using machine learning (ML) and deep learning (DL) approaches for automated assessment of the quality of physical rehabilitation exercises for COVID19 recovered patients. The Kinect and Vicon sensory systems were used in this study for motion capture of 10 movements of the rehabilitation program. Patients perform these 10 movements in their physical therapy and rehabilitation program. The ML models such as K-Nearest Neighbour (KNN), Decision Tree (DT), and DL based Multi-Layer Perceptron (MLP) were used to classify Kinect and Vicon. The binary Classification performance is measured with standard methods and metrics.  This study utilized 10 movements of the 10 subjects of two sensory systems and mean accuracy in training and testing part of KNN, DT, and MLP was 88.3%, 89.4%, and 78.95% respectively. DT misclassified 102 training and 104 testing values, MLP misclassified 333 training and 152 testing values, and KNN misclassified 153 training and 94 testing values. Overall mean accuracy of DT was greater than the other two models. The DT method showed better performance than other methods with best-learned features for any rehabilitation exercise. The DT revealed that the most important rehabilitation exercises were deep squats and sit-to-stands to recognize the Kinect and Vicon sensory system information. This study achieved an overall 89.4% accuracy rate to recognize the prescribed rehabilitation exercises of computer-assisted sensory systems. Therefore, our findings are helpful in clinical gait analysis, medical applications and related computer-assisted sensory systems for the recognition process.