Enhancing Psychologists' Understanding through Explainable Deep Learning Framework for ADHD Diagnosis

AI can improve ADHD diagnosis accuracy: The proposed framework achieved a 99% F1 score for detecting ADHD and 94.2% accuracy for categorizing different types of ADHD, suggesting that AI-assisted diagnosis could help address the current challenges in reliably identifying this neurodevelopmental disorder.

Explainability is essential for clinical trust: The paper emphasizes that the framework provides "interpretable insights into the diagnostic process so that psychologists can better understand and trust the results," highlighting that AI tools in healthcare must go beyond accuracy to show why they reach their conclusions.

Feature importance analysis reveals diagnostic drivers: Using explainability techniques like SHAP and Permutation Feature Importance, the framework identifies which specific patient characteristics most influence ADHD diagnosis decisions, allowing psychologists to validate whether the AI is focusing on clinically relevant factors.

Standardized interpretation methods strengthen responsible AI: The paper uses systematic approaches for feature selection, model comparison, and interpretation to "ensure the interpretability of the proposed framework," demonstrating that responsible AI requires transparent, reproducible methodologies—not just high accuracy numbers.

Human expertise remains central: By designing the system to enhance psychologists' understanding rather than replace their judgment, the framework treats AI as a decision-support tool that augments clinical expertise, maintaining human oversight in sensitive diagnostic decisions.