Use of ANN and Complexity Measures in Cognitive EEG Discrimination

Abstract

The purpose of this paper is to apply BP ANN to the discrimination of three kinds of subjects (clinical diagnosed 62 schizophrenic patients, 48 depressive patients and 26 normal controls) respectively in resting state with eyes closed and three cognitive tasks, with EEG complexity measures used as feature vectors. EEG activity is recorded from 16 scalp electrodes and recordings are digitized for off-line processing. Features vectors based on Lep-Ziv complexity and classification with ANN are implemented in Matlab6.5. The comparison between the results of classifying in four states is illustrated and discussed. The classification accuracies achieved are 60% and over. The results show that ANN is an effective approach for discrimination of these three kinds of objects both in baseline and some cognitive states.