A Survey on AI and Deep Learning Approaches for Medical Image–Genomics Fusion in Disease Risk Prediction

Progress in medical imaging and genomics has provided new methods for early and accurate prediction of disease.Most diagnostic models examine these modalities separately, which restricts the ability to study the interaction of anatomical and genetic determinants of disease susceptibility.In this study, we present a multimodal deep learning framework that combines whole-body computed tomography (CT) scans and genomic data to predict the risk rate of two specific diseases such as Type II Diabetes and Coronary Artery Disease (CAD).The framework uses nnU-Net for automated organ segmentation to extract spatial 3-Dimensional (3D) volumetric features, which are then fused with genomic representations in fully-connected fusion layers to allow the model to learn joint patterns between different modalities.In addition, to promote interpretability and clinical trustworthiness, Explainable AI (XAI) methods like Grad-CAM and feature attribution are implemented to visualize anatomical and genomic predictors.Altogether, this work provides a multimodal approach for enhanced diagnostic accuracy and transparency, which is also a step towards precise disease management.