Attempt for Learning Control of Paralyzed Upper Limb Motion by FES using Neural Network

Abstract

Automatic generation of stimulus parameters was clinically examined with machine learning control system using a neural network. The nonlinear relationship between hand posture and stimulus intensities were quantified by applying electrical stimulation to the supinator, wrist extensor and wrist flexor through percutaneous electrodes and measuring the supination and wrist extension angle in a hemiplegic subject. The measured relationship was modeled with a backpropagation neural network. The stimulus parameters generated by the trained network from the desired trajectory was applied to the subject. The result showed the feasibility to control the hand posture with the stimulus pattern generated automatically using a machine learning system.