A Meta-Learning-Based Ensemble Model for Explainable Alzheimer's Disease Diagnosis

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Explainable AI (XAI) models for Alzheimer's disease (AD) diagnosis lack transparency, limiting clinical use.
• Full-scale MRI in current XAI models introduces excessive features, creating a 'black-box' problem.

Purpose of the Study:

• To develop an explainable ensemble-based diagnostic framework for Alzheimer's disease.
• To enhance the transparency and clinical relevance of AI models in AD diagnosis.

Main Methods:

• An ensemble model integrating Random Forest, SVM, XGBoost, and Gradient Boosting was trained on clinical data and mid-slice axial MRI.
• The framework exclusively utilized mid-slice MRI, focusing on lateral ventricles to improve explainability.
• Meta-logistic regression was employed for the final diagnostic decision.

Main Results:

• Achieved high diagnostic accuracy: 99% on OASIS and 97.61% on ADNI with clinical data; 99.38% on OASIS and 98.62% on ADNI with mid-slice MRI.
• Combined modality approach yielded 99% accuracy.
• The AI consistently linked predictions to dilated lateral ventricles, a verifiable clinical biomarker for AD.

Conclusions:

• The proposed framework offers a step towards transparent AI-driven diagnostics for Alzheimer's disease.
• This research bridges the gap between diagnostic accuracy and explainability in XAI for AD.
• The use of mid-slice MRI enhances model transparency and clinical relevance.