Enhancing EEG-Based Schizophrenia Diagnosis with Explainable Multi-Branch Deep Learning.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Schizophrenia diagnosis relies on subjective clinical assessments, lacking objective biomarkers.
• Existing diagnostic methods face challenges in accuracy and early detection.

Purpose of the Study:

• To develop a robust and interpretable deep learning model for classifying schizophrenia using electroencephalography (EEG) data.
• To identify potential neural markers associated with schizophrenia through explainable AI techniques.

Main Methods:

• Proposed MBSzEEGNet, a multi-branch deep learning architecture designed to capture complex oscillatory and spatial-spectral features from resting-state EEG.
• Trained and validated the model on two independent schizophrenia EEG datasets.
• Employed saliency-based methods for model interpretability to pinpoint relevant EEG features.

Main Results:

• MBSzEEGNet demonstrated superior performance compared to existing deep learning models.
• Achieved high subject-wise classification accuracies: up to 85.71% on one dataset and 75.64% on another.
• Identified specific EEG frequency bands (delta, alpha) and brain regions (temporal, right parietal) as potential diagnostic biomarkers.

Conclusions:

• Explainable multi-branch deep learning models integrating EEG offer a promising avenue for objective schizophrenia diagnosis.
• The findings provide insights into schizophrenia-related neural mechanisms and support the development of novel biomarkers.