Enhancing schizophrenia diagnosis through EEG frequency waves and information-based neural connectivity feature fusion

• Published in: Biomedical signal processing and control Vol. 106; p. 107717
• Main Author: Goshvarpour, Ateke
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.08.2025
• Subjects: Brain connectivity assessment; Correntropy; Cross-information potential; EEG frequency waves; Information-based feature fusion; Schizophrenia detection; EEG frequency waves; Cross-information potential; Correntropy; Brain connectivity assessment; Information-based feature fusion; Schizophrenia detection
• ISSN: 1746-8094
• DOI: 10.1016/j.bspc.2025.107717

•Introduces a novel brain connectivity matrix for improved schizophrenia detection.•Addresses limitations of traditional methods using a single connectivity measure.•Employs information-based techniques to enhance classification efficiency.•Demonstrates improved accuracy in schizophrenia detection using machine learning. The early diagnosis of schizophrenia is crucial, and the identification of this disorder through the analysis of electroencephalography (EEG) signals and the application of classification methods is essential. This study presents a novel approach for the detection of schizophrenia through the analysis of EEG frequency waves and a fused EEG connectivity measure. Our methodology employs an information-based feature fusion technique that integrates the Correntropy and Cross-Information Potential algorithms to evaluate the similarity and information sharing between EEG signals. The RepOD dataset was utilized for the evaluation, from which biomarkers relevant to the diagnosis of schizophrenia were extracted. A variety of machine learning algorithms, including Support Vector Machine, Naïve Bayes (NB), AdaBoost (AB), Decision Tree (DT), and k-Nearest Neighbors (KNN), were employed to assess the efficacy of our approach. The NB model exhibited an accuracy rate of 100% across all brainwave classes, with the exception of the alpha wave during leave-one-subject-out (LOSO) cross-validation (CV). Similarly, both the DT and AB algorithms achieved 100% accuracy for sensorimotor rhythm and theta waves when utilizing fused information. The SVM demonstrated variable performance across different frequency bands, attaining perfect accuracy (100%) for sensorimotor rhythm, alpha, and theta waves when employing fused information and LOSO CV. The KNN algorithm exhibited considerable variability contingent upon the value of K, with the theta wave showing performance levels that differed from those of other frequencies during LOSO CV. Overall, LOSO CV provided superior performance compared to K-fold CV. The methodology presented in this study provides a comprehensive approach to the detection of schizophrenia, utilizing detailed EEG analysis, connectivity assessment, and feature extraction techniques. While there are potential limitations, including issues related to dataset representativeness and computational complexity, this research lays a foundational framework for future advancements in diagnostic practices within the field of neuroscience.