Research Themes
AI-Enabled Retinal Vasculometry
The retina provides a unique window into the vascular system, allowing early assessment of both ocular and systemic health, including conditions such as heart attack, stroke, diabetes, and glaucoma. Within our team, we have developed QUARTZ, a fully automated AI-enabled retinal vasculometry system designed to process large-scale retinal image datasets to extract measures of vessel size and shape. Using these measurements, our epidemiologists investigate associations between vascular features and disease pre-cursors, risk factors, and disease outcomes. These insights inform their development of disease risk prediction models.
End-to-End AI using retinal images and auxiliary data
We have established a Trusted Research Environment with access to over four million images with linked clinical and demographic data from a high-risk population. Leveraging this resource, we are developing a retinal foundation model trained on retinal images and auxiliary data to detect and predict a spectrum of systemic and ocular diseases, including cardiovascular conditions. To optimise performance, we are investigating novel pretext training tasks that advance learning in this domain.
Evaluation of Commercial AI Diabetic Retinopathy Detection Systems
We developed a framework to evaluate automated retinal image analysis systems for diabetic retinopathy within large-scale screening programs in a way that is transparent, equitable, and clinically meaningful. Using over 1.2 million images in a Trusted Research Environment within the North East London Diabetic Eye Screening Programme, the evaluation involved diverse ethnic groups and a spectrum of disease severity, ensuring robust assessment across populations. The findings highlighted key technical and operational considerations and demonstrate how independent, large-scale evaluations can build trust and guide standards for safe deployment of AI in clinical screening.