Integrating Genetic Information for Alzheimer's Diagnosis through MRI Interpretation

Abstract

Early detection of Alzheimer’s disease (AD) is crucial, yet predicting AD in the mild cognitive impairment stage remains challenging. Integrating biological data from genomics and neuroimaging can provide valuable insights into early detection and treatment. Although recent deep learning studies have shown promise in AD prediction tasks, they often lack the interpretation of multimodal data interactions. Therefore, there is a need for further research on deep learning methods that can effectively integrate and interpret multimodal biological data for AD diagnosis and prediction. This study proposes a novel approach for identifying regions where interactions occur in sMRI (structural MRI) and genetic information and for detecting discriminative features in AD progression. Through the use of an attention mechanism and contrastive loss, it effectively models the inter-relationships between these modalities, leading to a more accurate understanding of AD. Our proposed method achieved remarkable performance, with an accuracy of 92%. Additionally, through model interpretation, we were able to identify genetic and brain feature associations in AD progression. This integrating study provides an effective and interpretable approach for AD diagnosis and prediction.