Automated Feature-Based Detection, Grading, and Staging of Sight-Threatening Eye Conditions In Colour Fundus Photographs
Approved Research ID: 74127
Approval date: November 30th 2021
The aim of this project is to develop an automated system for the detection of sight-threatening eye conditions. This system will allow a machine to interpret eye images the same way a clinician would. The type of image this system will read is called Fundus photographs, which are images of the back of the eye that are used to assess eye health. Fundus photographs are used at multiple level of cares, from optometrists to specialised secondary care settings.
Crucially, this system will not simply provide a decision such as "disease detected" or "no disease detected", but will instead provide a granular report that can be directly related to existing diagnosis procedures. This way, the system can be used to supplement the decisions made by the healthcare professional, rather than overriding them.
Globally, 253 million people are blind or visually impaired. This disproportionately impacts patients in low- to middle-income countries (LMIC).
Artificial Intelligence (AI), which allows computers to be trained to carry out specific tasks, has been hailed as one of the solutions to alleviate the effects of this public health emergency. However, it has yet to demonstrate its full potential, especially in low- to middle-income countries (LMIC) where it is likely to be the most impactful. This is due to economic factors, lack of evidence based on patient data originating from these regions, and products that focus on overriding the user's decisions rather than supplementing them.
We want to take advantage of the UK Biobank data to build a system that will be truly useful when used on-the-ground. Based on previous experience trialling AI approaches in multiple clinical settings, we identified an approach that is likely to be successful in doing so. Instead of relying on computer algorithms that are not informative for the healthcare professional users (so-called "black box" system), we opted to develop a system that would provide granular information in a way that is directly relatable to diagnosis procedures the user is familiar with.
The project is estimated to last up to three years.
Public Health Impact:
We expect that this project is likely to have a significant impact on the diagnosis and treatment of those conditions in LMIC. Imaging date annotations to be shared back will likely spur more research in this area.