Explainable AI framework for improved Thalassemia mental health classification and feature selection.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Psychosomatic Medicine

Background:

• Mental health issues in Thalassemia patients are frequently underestimated, impacting their quality of life.
• Conventional statistical and machine learning methods struggle with the complex, nonlinear interactions between psychosocial and clinical factors in Thalassemia.
• Accurate prediction and interpretation of mental health status in Thalassemia patients remain challenging.

Purpose of the Study:

• To develop and validate a novel feature selection framework, Adaptive Multi-Stage Ensemble with Dynamic Feature Interaction (AMSE-DFI), for identifying mental health predictors in Thalassemia.
• To enhance the accuracy and interpretability of mental health assessments in Thalassemia patients.
• To provide a practical tool for early detection and personalized management of mental health challenges in Thalassemia care.

Main Methods:

• Developed AMSE-DFI, integrating mutual information, ensemble learning, and graph attention mechanisms for dynamic feature interaction.
• Utilized SF-36 health survey data from 356 Bangladeshi Thalassemia patients.
• Employed Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance and Local Interpretable Model-Agnostic Explanations (LIME) for model interpretability.

Main Results:

• AMSE-DFI identified key predictors of mental health in Thalassemia patients, including total SF score, role emotional, and physical health summary.
• The proposed framework demonstrated superior predictive reliability and generalization compared to conventional methods.
• LIME provided clear, interpretable insights into feature impact on individual patient outcomes.

Conclusions:

• AMSE-DFI offers a robust and interpretable approach for identifying mental health challenges in Thalassemia patients.
• The framework facilitates a deeper understanding of the interplay between clinical and psychosocial factors.
• This AI-driven tool supports clinicians in the early detection and personalized management of mental health in Thalassemia care.