Interpretable AI for Alzheimer's Diagnosis Using Cross-Sectional and Longitudinal Data

Alzheimer's disease (AD) is a progressive neurodegenerative disorder causing severe cognitive decline, particularly in older adults. Early diagnosis is crucial for effective treatment, yet traditional methods rely on costly neuroimaging techniques like MRI, limiting accessibility for lower-income groups. This paper proposes an affordable, interpretable machine learning framework for AD diagnosis using only cross-sectional and longitudinal clinical data. The model classifies patients into non-demented, mild, moderate, and severe dementia stages by leveraging temporal features such as MMSE score variations and brain volume changes over time. Explainable AI techniques enhance transparency, making the model clinically relevant and trustworthy. Experimental results show high classification accuracy, with LightGBM outperforming other models. By prioritizing clinical data over expensive imaging, this approach offers a scalable and costeffective diagnostic solution. The study underscores the role of interpretable AI in improving AD diagnosis and facilitating integration into real-world clinical settings.