Professor Tunde Peto, Queens University Belfast
Professor Anat Loewenstein , Tel Aviv University / Sourasky Medical Centre
Age-related vision loss has a major impact on quality of life, with age-related macular degeneration (AMD) being one of the most common causes of age-related vision loss. In late-stage AMD, new blood vessels grow at the back of the eye, leaking fluid that disrupts the intricate structure of the retina causing vision loss. Treatments are based on drugs that prevent new vessel growth and lead to improvements in vision. Treated eyes must be monitored monthly by imaging the retina and need to be re-treated if there is evidence of disease progression. Monitoring progression in an increasingly large patient population is a major logistical challenge for healthcare providers, leading to delayed appointments and the risk of patients losing vision. Reducing monitoring burden through automated analysis of retinal images may be key to the sustainability of retina clinics. Despite continuing treatment, a high proportion of patients start to lose vision in the long term (>4 years). Understanding the drivers behind the gradual loss in treatment effectiveness over time is a key challenge.
In this project, Professor Peto and Professor Loewenstein aim to investigate the performance of new automated software algorithms (the Notal OCT Analyser–NOA) to monitor AMD and detect progression towards implementation in routine patient’s management. In addition, they will investigate the potential for machine learning to identify factors and biomarkers associated with long-term AMD treatment response. The researchers will analyse linked electronic medical records and 3D retinal images collected over ten years in two large retina clinics, one in the UK and the other in Israel. They will apply novel image analysis algorithms developed by Notal Vision in conjunction with the Israeli clinical partner to quantify potential biomarkers of AMD progression, validated against clinical judgement of retinal specialists and real-world treatment decisions. Machine learning will be used to determine the utility of automatically extracted image features as biomarkers for predicting treatment responses and progression. Models will be trained and validated using data from the UK and performance tested using data from Israel.
This project will provide an evaluation of a software tool (the NOA) designed to automatically analyse and interpret retinal images taken at AMD monitoring appointments. This has the potential to increase sustainability of retina clinics under increasing demand by implementing automated, high quality decision support and freeing clinicians to deal with the most serious cases. The project will show the value of large-scale linkage of electronic medical records and imaging data and assess whether modern statistical and machine learning techniques can yield additional insight into disease mechanisms underlying AMD. This knowledge could form the basis for stratified treatment or provide new directions for research into treatments.
Research databases will be constructed in parallel by the UK and Israeli research teams, as will deployment of the NOA software at clinic scale. Manual grading of subsets of retinal images for detailed evaluation of the NOA will be led by the UK partner. Statistical analysis for the NOA evaluation will be led by the Israeli team and the machine learning component will be led by the UK team.