Automatic Diagnosis of Schizophrenia in EEG Signals using CNN Models

Abstract

This study uses convolutional neural networks (CNNs) to analyze electroencephalogram (EEG) signals to tackle the crucial problem of early schizophrenia diagnosis. In the study, two different CNN models with different architectural characteristics are compared and contrasted, with a focus on measures related to accuracy, loss, selectivity, and specificity. A meticulously selected dataset is obtained and preprocessed, containing EEG signals from people with and without schizophrenia. The chosen CNN architectures differ in terms of activation functions, filter sizes, and layer depth. The performance of the models is carefully assessed using metrics appropriate to the particulars of the problem, such as accuracy, loss, selectivity (true positive rate), and specificity (true negative rate), after they have undergone rigorous training on the preprocessed dataset. The goal of the comparison analysis is to offer a thorough grasp of how various architectural decisions affect the model’s capacity to correctly identify patterns linked to schizophrenia in EEG data. In addition to examining optimization and fine-tuning techniques to improve the model’s performance, the research also focuses on ethical issues to guarantee the responsible use of private health information. In the end, this work aims to further the development of deep learning-based EEG signal analysis diagnostic tools for schizophrenia, with a focus on the selection of assessment metrics that are important benchmarks for judging model performance. Keywords: schizophrenia, mental disorders, machine learning, biomedical signals, EEG signals, CNN