The proposed research aims to design advanced AI-driven methodologies for the accurate detection, segmentation, and classification of retinal arteries and veins using multimodal retinal imaging, with the ultimate goal of supporting cardiovascular disease (CVD) risk prediction, early detection, and monitoring. This project will explore the question: How can deep learning and data fusion techniques be leveraged to improve precision in retinal vessel analysis for cardiovascular health assessment? The main objectives are: (1) to develop robust, multifunctional deep learning algorithms based on encoder-decoder architectures for retinal vessel segmentation and artery-vein classification; (2) to implement AI models capable of identifying bifurcations and crossover points; (3) to create data fusion techniques that integrate information from multiple retinal imaging modalities-colour fundus, OCT, and retinal oximetry-into a unified framework; and (4) to extract, measure, and analyze critical vascular parameters relevant to systemic cardiovascular health. By using publicly available datasets like INSPIRE-AVR, DRIVE, STARE, and Messidor, along with expert-annotated clinical images from medical collaborators, the research will ensure model reliability and real-world applicability. Scientifically, this study is grounded in the known correlation between retinal microvascular changes and CVD, leveraging the retina as a non-invasive window into systemic health. Developing automated, accurate AI tools for retinal vascular analysis addresses the current gap in reliable quantitative assessment methods, potentially transforming early-stage CVD diagnostics. The outcomes of this project directly support national and global priorities in health and biomedical research by advancing technology-enabled disease prevention strategies.
