Seon Diagnostics Ltd

Explainable AI as the future of ARIAS and the development of a retinal abnormality detection medical device

Diabetic Retinopathy (DR) is a common potentially serious complication of diabetes which can lead to irreversible damage to sight and even blindness if not detected early and treated. The only proven way to ensure that DR is detected early and treated is through routine, annual eye examinations. The NHS has established a world-leading diabetic eye screening programme that offers annual eye examinations to all people aged 12 and over. Comprehensive diabetic eye screening programmes are very effective in minimising the risk of irreversible sight loss. However, whilst cost-effective, these programmes are expensive to run and, crucially, there is a persistent shortage of qualified staff to undertake the examinations and assess the pictures taken in the examination.

Over the last decade several Automated Diabetic Retinopathy Image Assessment Softwares (ARIASs) have been developed as tools to aid, or replace, the current approach to DR screening that is based on grading by trained professionals (human diabetic retinopathy graders). However, while all the currently available tools claim to detect, with variable levels of confidence, moderate or serious disease that needs to be seen by a senior doctor, none of them can provide the patient and their doctors with an explanation of how the ARIAS arrived at its result. These technologies are often referred to as ‘black boxes’ as the results they produce cannot be interrogated.

This project aims to address the black box issue by developing a system based on sophisticated computing technology that closely follows the approach of a human diabetic retinopathy grader. Our proposed solution will mimic the steps a human grader takes in assessing the disease present in a patient’s eye - firstly by identifying the kinds of disease present, the amount and its location and then, secondly, by translating this to a disease grade based on local grading procedures. In this way, the patient’s doctor will be presented with a grading result and the basis on how that result was determined by the ARIAS. Underpinned by extensive testing, we believe this approach will give patients and their doctors the confidence to integrate automated DR detection into the national diabetic eye screening programme. Additionally, we believe that millions of people around the world living with diabetes and at risk of sight loss can be helped by the ARIAS as a low-cost way to make routine diabetic eye screening available to them.

In a UK aging population with a rapidly growing incidence of type II diabetes, earlier detection and intervention will result in better outcomes for patients and positively impact on the cost to people and society of the burden of diabetes and other nondiabetics related abnormalities. The benefits of the ARIAS for patients living with diabetes in the UK would be expected to translate across all health systems around the world, especially those (the vast majority) where national diabetic eye screening programmes do not exist. Seon Diagnostics believes that its core technology has the potential to be deployed in the detection and early diagnosis of other pathologies in the eye based on image data acquired using other imaging modalities (for example, Optical Coherence Tomography). It is also feasible that this core technology may be able to detect signs in the eye that are predictive of other diseases and the metadata associated with the INSIGHT dataset holds the potential to expand these possibilities markedly.

18/11/2022

Bespoke Diabetic Retinopathy with grades

Data released under Data Licensing Agreement to a secure Trusted Research Environment provided by the University Hospitals Birmingham as the NHS Data Controller. Data Use Application 005

Not yet published