Taiwan-based nonprofit develops AI software for diagnosing diabetic eye disease

The Industrial Technology Research Institute, a non-profit R&D organization in Taiwan, has unveiled its AI-assisted system for detecting diabetic eye disease.


Point-of-care AI-DR can instantly diagnose diabetic retinopathy and diabetic ocular edema in five to 10 seconds. It can locate from fundus images four main lesions – microaneurysms, hemorrhages, soft exudates and hard exudates – and detect two anatomical landmarks, the optic disc and the macular area. It can additionally identify 14 other common fundus abnormalities, including retinal-related diseases, blood vessel changes, and optic nerve-related diseases.

AI-DR also provides five levels of diabetic retinopathy severity and produces binary classifications for diabetic retinopathy and diabetic macular edema.

The AI ​​model was trained using 150,000 fundus images. In one study, AI-DR was found to have 98% sensitivity and 96% specificity in diagnosing diabetic retinopathy.

According to ITRI, AI-DR can support all fundus cameras available in the world and be used to create other solutions such as advanced AI systems, autonomous web applications and services based on the private and public cloud. The organization recently provided a use the license for AI-DR to Acer for its VeriSee DR diagnostic software.


Last year, about 537 million adults aged 20 to 79 had diabetes. The International Diabetes Foundation expects their number to reach 643 million by 2030 and 783 million by 2045.

Currently, more than half of people with diabetes worldwide will develop DR and about one in 15 people will develop DME, according to the US National Eye Institute.

ITRI ​​said that with diagnostic solutions like AI-DR, the risk of blindness caused by diabetes complications can be reduced. He also pointed out that in the management of diabetes today, it is crucial to track changes in fundus symptoms as this can be better monitored than blood sugar levels.

Additionally, AI-DR’s intuitive design allows GPs and other non-ophthalmologists to perform rapid diabetic retinopathy screening as trained ophthalmologists, which may lead to increased screening rates. , lower treatment costs and better patient outcomes by enabling early detection.


Other existing diabetes eye-screening software products in Asia-Pacific are also powered by AI, including the ORACLE platform of the New Zealand company Toku Eyes, Singapore Eye Damage Analyzer by the Singapore Eye Research Institute, and VUNO Med-Fundus AI by South Korean medical AI company VUNO.

Meanwhile, a recent Google-supported study in Thailand demonstrated the accurate detection of diabetic retinopathy using a new deep learning system.

Additionally, Samsung is turn older Galaxy smartphones into AI-powered fundus cameras to screen for diabetes-related eye disease as part of its retraining program to improve access to eye health screening in underserved communities around the world.