Analysis of FES-induced upper limb motion for machine learning control

Automatic generation of stimulus parameters is an attractive alternative to manual configuration in upper extremity FES due to the large number of electrodes and stimulus parameters that need to be adjusted. The nonlinear relationship between stimulus and motion, and the biomechanical interaction among muscles were analyzed to determine the feasibility of using machine learning technique. A stepped stimulation paradigm was applied to one electrode while a continuous bias stimulation was applied to a second electrode in an incomplete hemiplegic subject. 43 Stimulus combinations were tested. The resulting limb motions (7 joint angles) were measured using a three-dimensional magnetic tracking system. The biomechanical interaction, e.g. the supinator depression by the wrist extensor was quantitatively described. It was shown that the quantification of the stimulus-motion relationship and the muscle interaction is possible by using the proposed method