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

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Schizophrenia (SCZ) diagnosis relies on clinical assessment, lacking objective biomarkers.
• Electroencephalogram (EEG) signals offer a non-invasive window into brain activity.
• Gamma (γ) wave activity in EEG has shown potential as a neural correlate of cognitive processes relevant to SCZ.

Purpose of the Study:

• To develop an objective and efficient diagnostic model for SCZ using EEG signals and deep learning.
• To leverage γ wave activity as a specific biomarker for improved diagnostic accuracy and model efficiency.
• To reduce model complexity and enhance training speed for practical clinical application.

Main Methods:

• An EEGNet deep learning architecture was employed, optimized for simplified feature engineering.
• Resting-state EEG recordings were utilized, with a focus on extracting γ band features.
• Leave-One-Subject-Out Cross-Validation (LOSOCV) was implemented for robust model evaluation, comparing SCZ patients and healthy controls (HC).

Main Results:

• The γ band feature model achieved high recognition accuracies: 98.19% for SCZ and 97.24% for HC.
• The model demonstrated significantly reduced training times, indicating enhanced computational efficiency.
• These results suggest a robust and efficient classification process suitable for large-scale datasets.

Conclusions:

• γ band features are effective biomarkers for EEG-based SCZ diagnostics.
• The proposed deep learning model provides high accuracy and improved training efficiency.
• This approach holds significant potential as an objective clinical diagnostic tool for SCZ.