Schizophrenia classification with single-trial MEG during language processing

Language disorder is a core symptom associated with schizophrenia. This study investigates schizophrenia classification based on brain activity during language processing. 6 healthy controls and 6 schizophrenia patients were instructed to read words and sentences silently while 248 channel magnetoencephalography (MEG) signals were recorded. For each trial, power spectral features were extracted in 8 frequency bands from all channels which form a spectral-spatial feature set. Top features ranked by F-score were fed into machine learning based classifiers for patient and control classification. Following cross validation procedure, 98.94% and 99.78% accuracies were achieved in classifying 470 word trials and 450 sentence trials, respectively. The high accuracy indicates abnormalities of brain activity during language processing in patient group and show that MEG patterns reflecting such abnormalities can be used to discriminate schizophrenia patients from healthy subjects. The proposed scheme may have potential application in schizophrenia diagnosis and classifying other mental diseases.