Generating evidence to support the role of AI in diabetic eye screening: considerations from the UK National Screening Committee.
Macdonald T., Zhelev Z., Liu X., Hyde C., Fajtl J., Egan C., Tufail A., Rudnicka AR., Shinkins B., Given-Wilson R., Dunbar JK., Halligan S., Scanlon P., Mackie A., Taylor-Philips S., Denniston AK.
Screening for diabetic retinopathy has been shown to reduce the risk of sight loss in people with diabetes, because of early detection and treatment of sight-threatening disease. There is long-standing interest in the possibility of automating parts of this process through artificial intelligence, commonly known as automated retinal imaging analysis software (ARIAS). A number of such products are now on the market. In the UK, Scotland has used a rules-based autograder since 2011, but the diabetic eye screening programmes in the rest of the UK rely solely on human graders. With more sophisticated machine learning-based ARIAS now available and greater challenges in terms of human grader capacity, in 2019 the UK's National Screening Committee (NSC) was asked to consider the modification of diabetic eye screening in England with ARIAS. Following up on a review of ARIAS research highlighting the strengths and limitations of existing evidence, the NSC here sets out their considerations for evaluating evidence to support the introduction of ARIAS into the diabetic eye screening programme.