The prevalence of diabetic retinopathy (DR) is increasing at an alarming rate worldwide. However, early detection and timely treatment of DR remains a challenge due to the shortage of ophthalmologists, who are responsible for carrying out fundus image examination for DR. To bridge the gap, ITRI developed an AI decision support technology for detecting DR features in fundus images, equipping non-ophthalmologists with an effective tool to provide early DR diagnosis for patients rapidly.
ITRI’s technology is the only in the world that uses AI to detect four major DR symptoms, including microaneurysms, hemorrhages, soft exudates, and hard exudates, to identify lesion locations, and to help doctors determine disease severity. The technology is developed for primary care physicians or endocrinologists to conduct early DR screening. Once early signs of DR (not merely microaneurysms) are detected, patients can be referred to the division of ophthalmology for further diagnosis and medical treatment. According to Grace Liu, an engineer of ITRI’s Computational Intelligence Technology Center, ITRI’s AI decision support system, in cooperation with non-ophthalmologists in DR screening, is expected to increase fundus examination rate by 20%.
This AI-based support system for DR applies deep convolutional neural network and ensemble learning strategies. Besides lesion localization, a lesion detection model is used to assist front-line medical doctors in severity level classification. The system provides a 5-level classification model for DR (i.e., normal, mild non-proliferative, moderate non-proliferative, severe non-proliferative, and proliferative). Moreover, it produces a binary classification model regarding the eye doctor referral decision. “We’re eager to contribute more detailed information in DR severity scale for Taiwan’s Diabetes Shared Care Network. Therefore, patients will be able to enjoy better healthcare solutions,” said Dr. Liu.
By combining human and artificial intelligence, this system has a positive impact on protecting human health in several ways. For instance, it reduces diagnosis time for DR with the help of AI. Moreover, the use of the system in primary clinics means that more patients can benefit from DR early screening. As a result, the increase of the early detection rate for potential patients may reduce healthcare and social costs. The system is also able to enhance the efficiency of diagnosing small lesions and eliminates the inconvenience of referrals to ophthalmologists.
From the industry point of view, the new system can be incorporated with and provide additional value to existing hardware. ITRI’s innovation has assisted manufacturers to upgrade their technological competence and helped create a high-end medical device market.
ITRI has already obtained more than 150,000 fundus images through cooperation with medical centers, with ophthalmologists’ expertise in interpreting the retinopathy stage of these images followed by massive input to the computer for AI training. ITRI’s AI technology on assistive diagnosis can also be applied in various other fields such as new drug development, colorectal cancer diagnosis, and telemedicine. It is hoped that the AI popularization from industries to daily life can allow the general public to enjoy AI-enabled medical care services.