Feature Selection and Machine Learning for Online Upper Limb Spasticity Detection

Hemiparesis after stroke often accompanies muscle spasticity, affecting patients' normal rehabilitation training and motor function recovery. Currently, many studies use surface electromyography (sEMG) signals to assess the spasticity status of patients, but these methods are typically used for offline assessment of spasticity levels in patients. This paper proposes an online spasticity detection method for detecting sudden spasticity in patients' upper limbs, which can enhance the safety of robot-assisted rehabilitation training. Firstly, feature selection techniques are employed to analyze and select common sEMG features. Subsequently, different machine learning models are used to classify spasticity sEMG signals into two categories, and the optimal model is selected by comparing classification accuracy. Finally, the accuracy and real-time performance of the online spasticity detection method were validated using sEMG signals from six post-stroke spasticity patients.