Taiwan has successfully developed Point-of-Care (PoC) to provide rapid diagnosis of diabetic retinopathy (DR) and diabetic macular oedema (DME). The PoC machine is a winner in the software/services category of the 2022 R&D 100 Awards. Taiwan’s most prominent high-tech applied research institutions claimed that the AI-assisted system is the first in the world and can be integrated into handheld and desktop fundus cameras commonly used in hospitals and clinics to detect DR and DME symptoms.
Using the PoC system, DR and DME can be diagnosed in as little as 5 to 10 seconds. The machine then assesses image quality instantly before diagnosis to avoid AI misinterpretation. Based on fundus images, it marks lesions and assigns severity levels to them. Furthermore, the Point-of-Care AI-DR can detect 14 other common ocular fundus abnormalities such as retinal diseases, blood vessel changes, and optic nerve diseases.
Due to its ease of use, general practitioners can conduct early screening for DR and DME complications. It can also determine whether a patient should see an ophthalmologist. Screening through primary care physicians raises screening rates, lowers treatment costs, and allow patients early detection outcome.
According to the International Diabetes Foundation, 537 million adults aged 20 to 79 worldwide had diabetes in 2021. The Foundation expects the number to rise to 643 million by 2030 and 783 million by 2045. According to the National Eye Institute, more than half of people with diabetes will develop DR, and one in every 15 will develop DME.
Point-of-Care AI-DR improves diabetes health monitoring and management by identifying lesions and severity stages of DR and DME. The PoC also lowers the risks of vision loss or blindness caused by diabetic eye complications. In addition, tracking changes in fundus symptoms over time allows physicians to gain a thorough understanding of a patient’s diabetes development status. The function is critical for diabetes management because the AI PoC can monitor changes in fundus symptoms. This is more reliable than the usual tracking of blood sugar levels which can change right before patient examination.
“Point-of-Care AI-DR is a collaboration result of human and artificial intelligence. The machine works based on ophthalmologists’ expertise and AI analysis. In addition, it uses complementary medical AI models to perform individual diagnostic tasks such as classifying and detecting symptoms to improve overall interpretation efficiency,” said Dr Pang-An Ting, Industrial Technology Research Institute (ITRI) General Director of Information and Communications Research Laboratories.
To develop Point-of-Care AI-DR, ITRI enlisted 50 ophthalmologists to collect and label 150,000 fundus images as training data for robust AI models. As a result, Point-of-Care AI-DR outperforms existing products in terms of sensitivity (>98%) and specificity (>96%) in DR diagnosis. Moreover, the outcome allows non-ophthalmologists to perform rapid point-of-care DR screening in the same way that ophthalmologists do. It can support all existing fundus cameras worldwide and be used to build various solutions, such as edge AI systems, standalone web applications, and private/public cloud-based services.
The detection provided is in multiple areas such as locating lesions (microaneurysms, haemorrhages, soft exudates, and hard exudates), detecting anatomical landmarks of the optic disc and macular area, classifying the severity levels of diabetic retinopathy (from none to severe) and producing binary classifications to assist general practitioners in providing ophthalmologists with data-informed decisions.
Licensing is available for Point-of-Care AI-DR.