A Real-Time Wrist Action Reconstruction System Design Based on BP Neural Network Model to Predict Multiple FES Parameters

Abstract

Diseases such as stroke and spinal cord injury can cause limb movement disorders, significantly impacting daily life. To restore wrist mobility in patients and achieve precise synchronization of the affected wrist with healthy wrist movement, this article designs a real-time wrist motion reconstruction system using a backpropagation (BP) neural network model to predict functional electrical stimulation (FES) parameters. The system uses a multichannel surface electromyography (sEMG) signals acquisition device to collect real-time sEMG signals from the wrist extensor and flexor muscles of the healthy forearm. The upper computer uses a sliding window strategy to capture sEMG motion frames in real time and extract five types of time-domain features. Among these, the waveform length (WL) feature is used to identify wrist flexion and wrist extension. The five extracted time-domain features are inputted in real time into a BP neural network model to predict the parameters of FES, including amplitude, pulsewidth, and frequency. The dual-channel FES device generates real-time FES pulses on the corresponding muscles of the paralyzed upper limb, achieving motion reconstruction. The experimental results show that the average correlation coefficient of the wrist joint angle curve is 0.9, with an average system delay time of 332 ms, and the average root mean square error (RMSE) of the angular velocity curve is 0.252. These results demonstrate that the system exhibits high performance in terms of real-time capability and accuracy, indicating its potential application value in wrist motion recovery.