Early detection of diabetic retinopathy

Software based image analysis software has now become good enough to be considered a cost effective alternative to manual diabetic retinopathy screening systems, according to Adnan Tufail MD, FRCOphth, professor of ophthalmology, Moorfields Eye Hospital, London, UK. “As a result partly of diabetic retinopathy screening in the UK, diabetic retinopathy is no longer the leading cause blindness in the working age population, for the first time in 50 years. We’re all aware of the treatment for diabetic retinopathy and diabetic oedema, but to get the best out of these therapies we need to treat our patients in a timely manner. I order to do that we need to see our diabetic patients regularly to pick up the disease,” he said. Automatic screening systems have been implemented in small programs in recent years and there are publications to support their use However, none of these systems has been tested independently in large real life screening programs to determine if they are safe or cost effective, he noted Dr Tufail and colleagues evaluated the screening utility of several diabetic retinopathy image assessment software (ARIAS) systems available in Europe in 20,258 consecutive images. The systems included iGradingM (Megalytics Group), Retmarker (Retmarker), and EyeArt (Eyenuk). Using manual grading as the reference standard, their study used the ARIAS systems to review 20,258 consecutive images, identify the presence of disease in patients with diabetes, and determine if the retinopathy was referable. Each system included a test set to optimize image importation. The Retmarker system showed good sensitivity in proliferative disease, identifying 98% of cases, but proved less sensitive to lower levels of retinopathy. Overall the system was as sensitive as human graders, filtering out 50% of patients without Some patients with mild disease were misclassified, but this would not effect the referral pathway as mild or no disease cases would still be seen in one year, he said. The Eyeart system had even greater sensitivity, with a nearly 100% detection rate for proliferative disease and 95% for any level of retinopathy. However, this system showed much lower specificity, filtering out only 20% of patients. The iGradingM system could not adequately detect the set of study images and classified them as upgradeable. “Both Retmarker and EyeArt achieved acceptable sensitivity for referable retinopathy when compared with human graders. They had sufficient specificity to make them cost-effective alternatives to manual grading as triage for the presence of diabetic retinopathy. This study suggests that ARIAS have the potential to reduce costs in developed-world health care economies and to aid delivery of diabetic retinopathy screening in developing or remote health care settings,” he concluded.