Explainable artificial intelligence for early Alzheimer's diagnosis using enhanced grey relational features and...

Area of Science:

• Neuroscience and Computational Biology
• Artificial Intelligence in Healthcare

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder with increasing global prevalence.
• Current diagnostic methods lack accessibility for early detection, posing a significant challenge.
• Balancing predictive performance and interpretability in machine learning for AD diagnosis remains a concern.

Purpose of the Study:

• To develop an interpretable and sustainable machine learning architecture for early Alzheimer's disease diagnosis.
• To leverage multimodal, structured clinical and behavioral data for enhanced diagnostic accuracy.
• To improve the correlation between features and diagnosis using a novel Grey Relational Grade index.

Main Methods:

• Extensive feature engineering, including composite features like MMSE age ratio and cognitive decline score.
• Addressing class imbalance with the Synthetic Minority Oversampling Technique.
• Implementing a strengthened Grey Relational Grade index and comparing seven mainstream classifiers, including Deep Neural Networks and CatBoost-based Stacking Ensembles.
• Utilizing Shapley Additive Explanations for model interpretability.

Main Results:

• Deep Neural Networks achieved the highest performance (Accuracy: 98.01%, AUC: 99.43%), followed by a CatBoost-based Stacking Ensemble (Accuracy: 97.91%, AUC: 98.10%).
• The enhanced Grey Relational Grade index significantly improved feature-diagnosis correlation from 0.725 to 0.891.
• Shapley Additive Explanations identified modifiable predictors such as family history, smoking, and early cognitive symptoms.

Conclusions:

• Combining enhanced Grey Relational Grade metrics with advanced machine learning and deep learning models offers an accurate and interpretable framework for early AD risk assessment.
• This approach can facilitate the implementation of effective, behavior-centric prevention strategies for aging populations.
• The study highlights the potential of interpretable AI in addressing the challenges of early Alzheimer's disease diagnosis and management.