Enhancing schizophrenia diagnosis efficiency with EEGNet: a simplified recognition model based on γ band features

Abstract

Objective: This study aims to develop an objective and efficient diagnostic model for schizophrenia (SCZ) by integrating electroencephalogram (EEG) signals with deep learning techniques. Building on previous research, γ wave activity is selected as a potential biomarker to achieve high recognition accuracy while significantly reducing model complexity and enhancing training efficiency.

Methods: We implemented an EEGNet architecture optimized for simplified feature engineering, targeting γ band features extracted from resting-state EEG recordings. The model was trained and evaluated using Leave-One-Subject-Out Cross-Validation (LOSOCV) to ensure robustness in distinguishing SCZ patients from healthy controls (HC).

Results: The γ band feature model achieved average recognition accuracies of 98.19 % for the SCZ group and 97.24 % for the HC group. Additionally, the model significantly reduced training time, indicating an efficient classification process that is more conducive to training on large datasets.

Conclusion: The findings highlight the effectiveness of γ band features for EEG-based SCZ diagnostics, with the proposed model offering both high accuracy and improved training efficiency. This study underscores the potential clinical utility of γ band-focused EEG analysis as an objective diagnostic tool for SCZ.

Keywords: EEGNet; Resting-state EEG; Schizophrenia diagnosis; γ band features.