3D ADHD-Net and DeepTrace: Decoding ADHD from EEG with neurophysiological insights.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Informatics

Background:

• Electroencephalography (EEG)-based Attention Deficit Hyperactivity Disorder (ADHD) diagnosis faces challenges with data leakage and lack of interpretability.
• Existing models often fail to preserve crucial spatial information from EEG data.

Purpose of the Study:

• To develop a robust and interpretable deep learning framework for EEG-based ADHD diagnosis.
• To address data leakage and enhance the clinical trustworthiness of ADHD assessment systems.

Main Methods:

• Developed 3D ADHD-Net, a topology-aware spatiotemporal model preserving scalp geometry.
• Introduced DeepTrace, an explainability framework for tracing diagnostic information to input electrodes.
• Employed strict subject-independent 5-fold cross-validation to prevent data leakage.

Main Results:

• Achieved 84.23% mean accuracy and 91.20% ROC-AUC on a pediatric EEG dataset.
• Significantly outperformed baseline models under rigorous evaluation.
• DeepTrace identified a fronto-central hypoactivation signature, validated through neurophysiological experiments.

Conclusions:

• Diagnostic performance and interpretability are achievable simultaneously in EEG-based ADHD models.
• Physiologically faithful attribution can be integrated into deep learning for clinical ADHD assessment support.
• The proposed framework offers a pathway toward clinically trustworthy EEG-based diagnostic tools.