Electroencephalography-based diagnosis of schizophrenia using machine learning.

Schizophrenia is a mental disorder with a high social burden. Identification of quantitative biomarkers has the potential to facilitate the diagnosis process. This study aims to explore a routine to gain such biomarkers using quantitative analysis of electroencephalography (EEG) data. Previous studies suggest that EEG data can be used to differentiate schizophrenia patients from healthy subjects. Various EEG features were used for such diagnostics using machine learning (ML) algorithms, but selecting the optimal EEG features and the classifiers is still insufficient. We propose an automatic selection of ML parameters using the Waikato Environment for Knowledge Analysis software. Using Waikato Environment for Knowledge Analysis's "Supervised Attribute Selection" tool, we identified attributes that allow the identification of schizophrenia patients with a high accuracy of 93%. The attributes identified were EEG signals enriched for alpha and gamma frequencies from specific brain areas (frontal right, central, parietal, and occipital). This proposed strategy can effectively identify schizophrenia patients with high accuracy. It could be used as an ML tool to support diagnosis and potentially provide insights into the underlying disease mechanism of schizophrenia.