Knowledge based decomposition of myoelectric signals

The authors present research relating to the use of computers for the intelligent decomposition of myoelectric signals. Digital filtering algorithms are used to reduce the noise in the signal. A normalisation and compression of the filtered signal is then performed to reduce the time of the analysis. The individual motor unit action potentials (MUAPs) in the myoelectric signal are identified using a pattern recognition method. The method uses features to describe the MUAPs in the myoelectric signal. Diagonal factor analysis is used to form uncorrelated factors from these features. The factors are then used in an adaptive clustering technique that groups together MUAPs from the same motor unit. A method is proposed whereby the superimposed waveforms formed by a summation of overlapping MUAPs are decomposed using a knowledge based expert system. Initially, a template matching procedure is used to identify the possible MUAPs that make up the superimposed waveform. The motor unit firing times are then analysed by the expert system to decide which combination of MUAPs is most probable. Rules are used in the decision making of the system and are driven by an interpreter that incorporates uncertain reasoning based on fuzzy set theory