Interpretable Deep Learning for Alzheimer's Disease Through Genetic Data and Explainable Artificial Intelligence

Alzheimer's disease (AD) is a progressive neurodegenerative disorder causing cognitive decline and memory loss. With its significant impact on individuals' lives, AD is the most prevalent form of dementia, contributing to 60-80% of all dementia cases. At the same time, symptoms may not surface until years later, making early detection vital for effective intervention. Thus, this work presents an approach to early AD detection by integrating Genome-Wide Association Studies (GWAS) with deep learning models and Explainable Artificial Intelligence (XAI). First, different classical machine learning models are developed for AD, and a Convolutional Neural Network (CNN) model is trained using the AD GWAS dataset obtained from the AD neuroimaging initiative. We then employ transfer learning to train our CNN model as a base model over the ADNI dataset. In addition, XAI methods are used to interpret the transfer learning model decision. Acknowledging the well-known limitation that classical machine learning is not inherently a generic model. The results from this study will help determine the most critical genetic markers associated with AD and provide transparency in understanding the deep learning model decisions.