METHODICAL INVESTIGATIONS FOR AUTOMATIC R ECOGNITION OF TOPOGRAPHIC PATTERNS OF MYOELECTRIC ACTIVITY

For automatic classification of defined patterns of myoelectric activi ty, a processing method of the topographical (multi-channel) recorded surface EMG was developed. This strategy is based on the combination o f the dynamic EMG-mapping, achieved by Hilbert-transformation, and aut omatic classification by artificial Neural Networks. Neural Network cl assifiers were trained with interval-related spectral parameters (FFT- power spectrum) and the classification was carried out with a higher t ime resolution (equivalent dynamic spectral parameters via Hilbert-tra nsformation) than in the training phase. It can be shown that the onse t and the end of the topographical activation pattern can be detected efficiently. This strategy was tested on healthy volunteers and patien ts during chewing activity based on EMG-recordings from M. masseter.