A spasticity assessment method for voluntary movement using data fusion and machine learning

Abstract The assessment of spasticity under voluntary movement is helpful for the therapist to comprehensively assess the patient's dyskinesia. However, current researches focus on spasticity evaluation based on passive motion. We propose a new method for evaluating spasticity under active motion. Our method is based on the following three steps: (i) Empirical Mode Decomposition (EMD) is used to reduce involuntary movement noise in patients' active movement; (ii) Extract voluntary movement segments of each muscle for feature extract and fusion; (iii) Use machine learning methods to evaluate the degree of spasm in patients. To investigates the feasibility of the method proposed in this paper, An experiment of elbow flexion and extension against gravity is designed, and the electromyographic signal of brachioradialis (BR), biceps brachialis (BB), triceps brachialis (TB) and elbow motion data of 13 subjects were collected. We compared the classification effect of filter method, window length and classifier type. Moreover, we analyze the improvement of classification effect by data fusion. The results showed that the random forest with a window length of 256 ms had the best effect (F1-score = 0.952). Compared with the electromyographic signal (F1-score = 0.756) or motion signal only (F1-score = 0.7053), the method presented in this paper had better classification accuracy. Result demonstrated the feasibility of our method. This study can assist doctors to evaluate patients' spasmodic state under active movement, and has the application potential of wearable devices.