Selection of abnormal neural oscillation patterns associated with sentence-level language disorder in Schizophrenia

Abstract

Language disorder is one of the core symptoms in schizophrenia. We propose a new framework based on machine intelligence techniques to investigate abnormal neural oscillations related to this impairment. Schizophrenia patients and healthy control subjects were instructed to discriminate semantically and syntactically correct sentences from syntactically correct but semantically incorrect sentences presented visually, and 248-channel MEG signals were recorded with a whole head machine during the task performance. Oscillation patterns were extracted from the MEG recordings in 8 frequency sub-bands throughout sentence processing, which form a large feature set. A two-step feature selection algorithm combining F-score filtering and Support Vector Machine recursive feature elimination (SVM-RFE) was designed to pick out a small subset of features which could discriminate patients and controls with high accuracy. We achieved a 90.48% prediction accuracy based on the selected top features, following the leave-one-out cross validation procedure. These top features provide interpretable spectral, spatial, and temporal information about the electrophysiological basis of sentence processing abnormality in schizophrenia which may help understand the underlying mechanism of this disease.