https://doi.org/10.1140/epjs/s11734-025-01507-0
Regular Article
Schizophrenia diagnosis using latent components of event-related potentials and machine learning approach
1
Institute of Problems in Mechanical Engineering, Russian Academy of Sciences, St. Petersburg, Russia
2
Institute of the Human Brain, Russian Academy of Sciences, St. Petersburg, Russia
3
Brain and Trauma Foundation, Switzerland, Chur, Switzerland
Received:
2
November
2024
Accepted:
6
February
2025
Published online:
21
February
2025
Schizophrenia is a severe mental illness that can cause lifelong impairment, so early and accurate diagnosis is of particular importance. One of the possibilities to improve the reliability of schizophrenia diagnostics may be the use of machine learning to classify electroencephalogram (EEG) data from patients of healthy subjects. In our recent work, applying Support Vector Machine (SVM) model to event-related potentials (ERP) data of schizophrenia patients and healthy control yielded sensitivity and specificity rates of 91% and 90.8%, respectively. Due to the volume conductivity of the brain, ERPs recorded from the scalp surface are a mixture of signals from many sources. Mathematical algorithms called Blind Source Separation (BSS) methods can allow to reconstruct the original signals from EEG data with acceptable accuracy even in the presence of significant temporal overlap of these signals and noise. Using this method allows observing the phenomena that are hidden by other signals in the conventional ERPs and provides additional information about the functioning of the cortex. Therefore, the aim of the present study was to test if the applying of the BSS method to ERPs can improve data classification accuracy in diagnosing schizophrenia. In this study, machine learning models were applied to latent components of ERPs recorded from patients with schizophrenia and healthy control while performing the visual cued Go/NoGo test. Data sample consisted of 132 healthy subjects and 68 patients in the age range 18–50 years old. ERPs were transformed into latent components using the BSS method. The latent component signals were compared between groups of subjects to identify intervals with significant differences, which were then used to train the model. A range of features was derived from these latent components across defined time intervals to serve as inputs for machine learning models. The SVM model achieved sensitivity and specificity rates of 96.7% and 97.7%, respectively improving upon past results for the same dataset obtained using ERP signals. A relatively high percentage of correct classifications allowed us to consider the applied methodology as promising for the development of a tool for diagnosing schizophrenia. Time intervals used for model training correspond not only to the well-known early and late ERP waves but also to fluctuations, which currently have no neurophysiological interpretation. The results of the study provide a basis for further studies of neural networks impaired in schizophrenia.
Copyright comment Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
© The Author(s), under exclusive licence to EDP Sciences, Springer-Verlag GmbH Germany, part of Springer Nature 2025
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
