DVT-Net: A Multimodal Deep Vascular topology Network for Disease Prediction

Disease complications can alter vascular network morphology and disrupt tissue functioning. Diabetic retinopathy (DR), for example, is a complication of types 1 and 2 diabetes mellitus and the leading cause of vision loss in working-age adults worldwide. Interestingly, DR is not an isolated case; many systemic and ocular diseases are associated with altered retinal vessel morphology, suggesting their detection may have a diagnostic value. Currently, microvascular diseases are assessed by visual inspection of retinal images, but this can be challenging when diseases exhibit silent or subtle symptoms or patients cannot attend in-person appointments. Thus, comprehensive screening is not widely performed, facing challenges related to availability of human assessors. To facilitate a streamlined screening process for DR and other microvascular diseases, we developed DVT-Net' (Deep Vascular Topology Network): a novel multimodal pipeline that fuses topological summaries of segmented retinal vascular images with deep learning (DL) features for the purpose of enhancing disease detection interpretability and accuracy from early to late stages.