Restoration of movement using functional electrical stimulation and Bayes' theorem

Abstract

Various computational approaches have been applied to predict aspects of animal behavior from the recorded activity of populations of neurons. Here we invert this process to predict the requisite neuromuscular activity associated with specified motor behaviors. A probabilistic method based on Bayes' theorem was used to predict the patterns of muscular activity needed to produce various types of desired finger movements. The profiles of predicted activity were then used to drive frequency-modulated muscle stimulators to evoke multijoint finger movements. Comparison of movements generated by electrical stimulation with desired movements yielded root mean squared errors between ∼18 and 26%. This reasonable correspondence between desired and evoked movements suggests that this approach might serve as a useful strategy to control neuroprosthetic systems that aim to restore movement to paralyzed individuals.