Channel selection against electrode shift enables robust myoelectric control without retraining

Abstract

Myoelectric controlled interfaces driven by muscle activities have achieved good performance in ideal conditions and showed many potential medical-related and industrial applications. However, in practical applications, the performance could be drastically degraded due to the electrode (sensor) shift, which is inevitable in donning and doffing the system. In this study, we presented a novel channel selection method against electrode shift for robust pattern-recognition based myoelectric control. The proposed method was evaluated on twenty-four subjects, including twenty-two able-bodied subjects and two amputees, and compared with two traditional channel selection methods, i.e., uniform selection (UNI) and sequential feature selection (SFS). We demonstrated that the offline error rates of the proposed method were significantly lower than those of the other two methods (P<0.05), and its online performance in shift conditions was comparable to that in ideal conditions. These outcomes benefit the practical applications of robust myoelectric controlled interfaces.