ETD-XAI: An explainable two-stage EEG-biochemical framework for pediatric tic disorders

Area of Science:

• Neuroscience
• Biomedical Engineering
• Pediatric Neurology

Background:

• Current pediatric tic disorder assessment relies on behavioral observation and rating scales, lacking objective physiological biomarkers.
• Existing EEG and vitamin D-calcium-phosphorus indicators offer potential but suffer from limited subject-level interpretation.
• There is a need for an interpretable, physiology-informed framework for objective pediatric tic disorder assessment.

Purpose of the Study:

• To develop an explainable, two-stage assessment framework (ETD-XAI) for pediatric tic disorder.
• To integrate subject-level EEG neurophysiological features with vitamin D-calcium-phosphorus metabolic indicators and clinical variables.
• To provide interpretable, physiology-informed assessment for subject-level analysis.

Main Methods:

• Proposed the explainable two-stage assessment framework with XAI (ETD-XAI).
• Integrated subject-level EEG features, vitamin D-calcium-phosphorus indicators, and clinical variables.
• Utilized training-fold-restricted explainability for feature contribution, modality weighting, and cross-modal interaction modeling.

Main Results:

• ETD-XAI demonstrated feasible performance in cross-validation and independent validation (AUROC 0.90-0.92 for screening, 0.87-0.90 for severity grading).
• SHAP analysis identified β/γ-band EEG activity, complexity features, and serum 25-hydroxyvitamin D as key contributors.
• Feature contributions showed severity-related gradients and age-stratified consistency.

Conclusions:

• ETD-XAI shows feasibility for interpretable multimodal modeling in pediatric tic disorder assessment.
• The framework provides subject-level, physiology-informed insights.
• Further validation in larger cohorts and clinical evaluation are necessary before clinical application.