Explainable artificial intelligence-driven visual task-specific electroencephalogram analysis for attention deficit...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Attention Deficit Hyperactivity Disorder (ADHD) is a common childhood neurological disorder impacting cognitive and sensory functions.
• Existing machine learning (ML) models for ADHD diagnosis from electroencephalography (EEG) data often lack interpretability.
• There is a need for explainable AI (AI) methods to understand diagnostic predictions in ADHD.

Purpose of the Study:

• To develop an explainable AI-driven approach for ADHD diagnosis using EEG data.
• To enhance the interpretability of ML model predictions for ADHD.
• To identify specific EEG patterns associated with ADHD in children.

Main Methods:

• Computed entropy, mutual information, and transfer entropy from EEG data using the 10-20 system.
• Aggregated electrode-level features into lobe-wise representations for ML models.
• Applied and selected the optimal ML model (Support Vector Machine) using explainability techniques (LIME, SHAP, PDP).

Main Results:

• Support Vector Machine achieved 92% accuracy in classifying ADHD from EEG data.
• Explainability methods revealed higher entropy and transfer entropy in the frontal lobe indicate uncertainty.
• Reduced participation of the occipital lobe suggests visual perception dysfunction in children with ADHD.

Conclusions:

• The proposed explainable AI approach effectively diagnoses ADHD with high accuracy using EEG data.
• The findings highlight the importance of frontal and occipital lobe activity patterns in ADHD.
• This method is suitable for clinical screening and provides insights into ADHD neurophysiology.