Machine learning analysis of neuromuscular signals in the human arm

Myoelectric signals are nervous impulses that activate skeletal muscles. Deciphering the messages encoded in myoelectric signals could open the door to the next generation of prosthetics and UI devices, creating seamless human machine interfaces that do not rely on physical controls. Unfortunately, myoelectric signal interpretation remains a challenging problem. The current state of the art is to use machine learning algorithms to classify patterns of myoelectric signals into gestures that can be interpreted by prosthetics or UI systems. A major complication of myoelectric signal analysis is that the signals are sent in bursts along multiple nerve endings, these bursts are further amplified in muscle tissue, creating a convoluted signal mixture. Independent component analysis (ICA) is a technique useful for deconvoluting mixed signals. In this work, we apply ICA to myoelectric signal analysis. We show the utility of ICA when applied to high resolution myoelectric data by significantly improving accuracy of gesture classification with the use of ICA.