Schizophrenia Classification using Working Memory MEG ERD/ERS Patterns

Abstract

In this paper we investigate the use of event related desynchronization (ERD) and synchronization (ERS) patterns extracted from magnetoencephalogram (MEG) in a working memory task to discriminate between controls and patients with schizophrenia. In the experimental paradigm, sequential letters appearing on a screen are memorized by subjects. In one of two conditions the letters constituted a word. The ERD and ERS patterns are extracted in the theta, alpha, beta and gamma bands from 248 electrode locations covering the whole head. We noticed that most of the ERD patterns are localized on the left frontotemporal area in both word and nonword conditions in the late memorization stage. The beta band showed the most significant difference in this cortical area between controls and schizophrenia patients. By using a decision tree, 94.7% and 87.5% classification accuracy was obtained for controls and patients individually in both word and nonword conditions. Furthermore, we report that on the left frontotemporal lobe, the discrimination within the beta band between patients and controls in the word condition was higher than in the nonword condition. The higher discrimination within the word condition can be linked to the abnormalities in language processing in schizophrenia patients. Our results show that the ERD/ERS patterns extracted from MEG can be successfully used in patient-control discrimination with appropriate adjustment of spatial, spectral, temporal and functional process parameters