Classification of Schizophrenia Patients by EEG Microstate Features Based on Polarity Sensitivity of EEG Signals

Schizophrenia is a chronic mental disorder whose symptoms are disabling and challenging for those who suffer. The concept of EEG microstates helps analyze the spontaneous EEG activity by segmenting it into quasi-stable topographic patterns. The topographical patterns are represented by an optimal number of microstate topographic clusters. The purpose of this study is to investigate the classification of schizophrenia patients and healthy controls using various methods to determine the optimal number of microstate topographic clusters based on the polarity sensitivity of EEG signals. We find the optimal number of clusters according to two new aspects: polarity-sensitive analysis containing Calinski-Harabasz (CH) and Davies-Bouldin (DB) indices, and polarity-insensitive analysis involving the Kneedle algorithm and crossvalidation (CV) index. Additionally, microstate parameters and conventional EEG features are used for classification separately. The results show high reliability for both methods, with high accuracies. For CH and DB, the results are the same, with the best accuracy being ${8 5. 8 0 \%}$ using the Random Forest classifier. For the Kneedle algorithm and CV index, the best accuracies are ${9 1. 6 8 \%}$ and ${7 9. 8 4 \%}$ both using Random Forest, respectively. On the other hand, the accuracy of classification using conventional EEG features is ${9 4. 4 2 \%}$ with MLP.