Integrating explainable AI with clinical features to enhance ADHD diagnostic understanding

Area of Science:

• Neuroscience
• Artificial Intelligence
• Clinical Psychology

Background:

• Adult Attention Deficit Hyperactivity Disorder (ADHD) diagnosis is challenging due to subjective assessments.
• Machine learning (ML) offers potential but lacks transparency for clinical use.
• Limited research integrates broad clinical, substance-use, and quality-of-life data for ADHD prediction.

Purpose of the Study:

• To develop and evaluate a transparent ML framework for adult ADHD diagnosis.
• To integrate multimodal clinical data, including symptom scales, substance use, and quality of life.
• To compare explainable AI (XAI) insights with traditional statistical analyses.

Main Methods:

• Retrospective analysis of 786 adult assessments from a UK specialist mental health service.
• Utilized XGBoost classifier with SHapley Additive exPlanations (SHAP) for interpretability.
• Compared SHAP attributions with Pearson correlations and Welch's t-tests.

Main Results:

• Achieved 77% accuracy and 0.82 AUC-ROC for ADHD diagnosis.
• CAARS ADHD Raw scores and DIVA adulthood inattentiveness were primary predictors.
• SHAP revealed interactions, e.g., PHQ-9 amplifying ADHD symptom prediction; age and gender moderated effects.

Conclusions:

• Transparent ML with multimodal data improves adult ADHD diagnostic consistency.
• The SHAP-EDA approach provides interpretable, clinically relevant insights.
• Findings support a patient-centered, data-driven diagnostic framework.