Estimation of Hand Posture from Stimulus Position in FES: An Attempt using Machine Learning

This study is to construct a system for patients to easily search for the appropriate stimulation positions of functional electrical stimulation (FES) to obtain the desired posture for daily rehabilitation. We applied 125 patterns of electrical stimulation using integrated power-net multi-point electrodes that we developed. The posture discrimination of each five fingers was performed using hand tracking. 20 different machine learning algorithms were investigated with the accuracy of for robust hand estimation based on the relationship with the induced hand posture and the stimulation patterns. The experimental results illustrated the degree of difficulty of estimating hand posture using the stimulus center based on the evaluation results using AUC as the evaluation index of machine learning, and that the consideration of direction vector of electrical stimulation increased the accuracy.