An intelligent electromyogram signal characterization method based on neuro-fuzzy model

Abstract

Neuro-muscular and musculoskeletal disorders and injuries highly affect the life style and the motion abilities of an individual. The primary purpose of this work is to develop a systematic method for detection of the level of muscle power declining in musculoskeletal and Neuro-muscular disorders. To this aim, the EMG signals of five skeletal muscles as biceps, deltoid, triceps, tibialis anterior and quadriceps muscles are recorded in three states of isometric contraction (ISO), maximum voluntary contraction (MVC) and dynamic contraction from 22 normal subjects aged between 20 and 30 half of them are male. Totally, 14 combinatory extracted features are analyzed to find which of them or a combinatory set of them are discriminative and selective for muscle force quantification and classification. The neuro-fuzzy system is trained with 70 percent of the recorded EMG cut off windows and then it is employed for classification and modeling purposes. For each muscle the most effective extracted features are found for males and females separately by a reference classifier. In the experiments, after the optimum set of combinatory features is found by a reference classifier, the neuro-fuzzy classifier is validated in comparison to some other well-known classifiers in classification of the recorded EMG signals with the three states of contractions corresponding to the extracted features. Then, different structures of the neuro-fuzzy classifier are also comparatively analyzed to find the optimum structure of the classifier used.