Discriminant EEG analysis for differential diagnosis of schizophrenia. Methodological aspects

Abstract

A search for accurate linear discriminant function (LDF) allowing the diagnosis of schizophrenia and estimation of treatment effectiveness according to EEG is an urgent problem. To develop a methodology for discriminant EEG analysis for minimizing the overlearning effect, selection of optimal LDF model and evaluation of its generalizing ability. Two hundred and twenty patients with schizophrenia and 1400 people without psychiatric diseases, who were comparable in basic characteristics, were enrolled. EEG was recorded using 16 leads, 10-20 system and aural reference electrodes. EEG was processed by spectral and coherent analysis. After linear discriminant analysis, LDF was obtained to differentiate people with schizophrenia from healthy subjects and a formula was selected from LDF that included 8 predictors (spectral and coherent parameters of standard EEG ranges theta, alpha and beta) with 90% sensitivity and 80% specificity, significance level for Wilks' lambda p<3.9E-28 and the Mahalanobis distance between training set centroids 4,6. A method for obtaining optimal LDF models and selection the best one with further LDF generalizing ability assessment is suggested.