Development of AI Techniques to Predict Eye Cancer Using Big Longitudinal Data

Choroidal naevi are common pigmented fundus lesions (8% of the population) with a very low risk of conversion into choroidal melanoma.

The increasing use of wide-angle fundus examinations by High Street Opticians and the fear of litigation if a melanoma is missed, has led to a rising number of referrals of pigmented fundus lesions to Ophthalmology and Ocular Oncology Units.

This places a significant amount of pressure on Ophthalmology Departments in the country regarding new patient and follow-up appointments.

In order to address this clinical need, this project brings together an interdisciplinary team with distinct, complementary expertise in AI, medical image computing, and clinical medicine.

The overarching aim is to develop and evaluate a novel, fully-automatic AI-powered diagnostic tool to support the accurate diagnosis and monitoring of choroidal naevi and to predict the risk of ocular melanoma.

This research will harness (1) a unique database of ocular tumours, the largest of its kind worldwide, with a vast enriched dataset of images and clinical data of over 3000 patients, and (2) novel AI prognostic models that have been developed in Liverpool funded by EPSRC and NIHR.

Our specific objectives are: (1) to complete curation of a dataset essential for the development of AI tools, (2) to adopt a software development approach to develop and rigorously validate new AI tools by following TRIPOD and PROBAST principles and (3) to deliver the project by proactive managements, public engagement and risk mitigation and develop a robust commercialisation strategy to market.

We expect by completion of the project to have developed AI tools ready for future prospective clinical study and commercialisation.

The project objective is well aligned with the NIHR AI funding strategy to advance translation of AI enabled healthcare technologies into new diagnostic tools to increase patient benefit and contribute economically to UK PLC.