A Novel Physiologically-Inspired Method for Myoelectric Prosthesis Control Using Pattern Classification

The contemporary myoelectric prostheses are advanced mechatronic systems, but human-machine interfacing for robust control of these devices is still an open challenge. We present a novel method for the recognition of user intention based on pattern classification which is inspired by the natural coordination of multiple muscles during hand and wrist motions. The coordinated muscle activation produces a characteristic distribution of the amplitude features of the electromyography signals, and the novel method establishes the class boundaries to capture this natural distribution. The method has been tested in healthy subjects operating a prosthesis during a challenging functional task (bottle grasping, turning and releasing). The novel approach outperformed the commonly used benchmark (linear discriminant analysis), while using shorter training and fewer features. Further developments can, therefore, lead to a method that is suitable for practical implementation and allows robust and efficient control.