Effect of Shoulder Movement on Assessing Upper Limb Performance of Stroke Patient

Abstract

Upper limb assessment using the rehabilitation device is entirely depends on the movements performed by the stroke patients. However, stroke patients use available motor strategies to reach the target position to compensate for their upper limb weakness. This can lead to inaccurate assessment data for the predictive analysis. The primary goal of this study is to examine the effect of shoulder movement in upper limb assessment for predicting the Motor Assessment Scale (MAS) score using Partial Least Squares (PLS), Artificial Neural Network (ANN), and hybrid (i.e. PLS-ANN) predictive models. However, the feature selection method on the model's input predictor influences the predictive model's performance. To attain the greatest prediction performance, an appropriate feature selection method should be investigated. The results reveal that PLS-ANN with all kinematic variables (KVs) as the input predictors has a better prediction accuracy after the implementation of shoulder movement compared to other predictive models. Furthermore, the results proving that by considering shoulder movement to generate the KVs based on the actual distance of reaching movement may enhance the prediction accuracy in predicting MAS score of stroke patients.