Related Experiment Video
Updated: Mar 22, 2026
12:00
Investigating the Effects of Antipsychotics and Schizotypy on the N400 Using Event-Related Potentials and Semantic Categorization
Published on: November 19, 2014
13.4K
Exploring brain lobe-specific insights in an explainable framework for EEG-based schizophrenia detection
Md Milon Hossain1, Md Nurul Ahad Tawhid1
1Institute of Information Technology, University of Dhaka, Dhaka, Bangladesh.
Plos One
|March 20, 2026
Summary
This study introduces a novel framework using electroencephalography (EEG) mel-spectrograms and convolutional neural networks (CNNs) for accurate schizophrenia detection. The frontal lobe shows the most promise for diagnosis, enhanced by explainable AI techniques.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Medical Diagnostics
Background:
- Schizophrenia (ScZ) poses a significant global health challenge, necessitating improved diagnostic tools.
- Current Electroencephalography (EEG) methods for ScZ detection lack specific brain lobe biomarkers and explainability.
- Integrating Explainable Artificial Intelligence (XAI) is crucial for building trust in AI-driven medical diagnoses.
Purpose of the Study:
- To develop and validate a novel framework for Schizophrenia (ScZ) detection using EEG mel-spectrograms and Convolutional Neural Networks (CNNs).
- To investigate the diagnostic significance of different brain lobes in ScZ using EEG data.
- To enhance the interpretability of the diagnostic model through the application of XAI techniques.
Main Methods:
- EEG signals were transformed into mel-spectrogram images using Short-Time Fourier Transform (STFT).
- A CNN model was employed for classification between ScZ patients and healthy controls (HC).
- The brain was segmented into five regions to assess lobe-specific diagnostic performance, and XAI methods (LIME, SHAP, Grad-CAM) were applied for explainability.
Main Results:
- The proposed framework achieved high accuracy in ScZ detection (99.82% on repOD, 98.31% on Kaggle dataset).
- The frontal lobe demonstrated the highest diagnostic accuracy (97.02%, 88.03%), followed by the temporal lobe, while the occipital lobe showed lower accuracy.
- XAI techniques successfully identified key factors contributing to the classification, enhancing model transparency.
Conclusions:
- EEG-based mel-spectrogram analysis with CNNs offers a highly accurate and explainable approach for ScZ detection.
- Brain lobe analysis, particularly focusing on the frontal lobe, can significantly improve diagnostic specificity.
- The integration of XAI is vital for clinical adoption and trust in AI-powered neurological diagnostic systems.
