Control of hand prosthesis using fusion of information from bio-signals and from prosthesis sensors

The paper deals with an enhanced approach of recognising intentions of a patient to move a hand prosthesis when manipulating and grasping items in a way that is skillful. The method follows a 2-level multi-classifier system (MCS) with heterogeneous classified bases with a relationship to EMG and MMG signals and a mechanism that combines the use of a probabilistic competence functions of base classifiers and dynamic ensemble selection scheme. Additionally, two original concepts of the use of feedback in signal deriving from the prosthesis sensors to improve the classification accuracy are presented. In the first method, the feedback signal is dealt as a data source about a correct class of hand movement and competence functions of base classifiers are dynamically tuned according to this information. In the second approach, classification procedure is organized into multistage process based on a decision tree scheme and consequently, feedback signal indicating an interior node of a tree allows us to narrow down the set of classes. The performance of MCS with both methods of using feedback signal were experimentally tested on real datasets concerning the recognition of six types of grasping movements. The development of the systems accomplished high classification accuracy showing the value of multiple classifier systems with multimodal biosignals and signal of feedback from the prosthetic sensors for the control of bioprosthetic hand.