Featured Discovery

Home > Featured Discovery > DeepDR: AI System Aids Accurate Detection of Diabetic Retinopathy

DeepDR: AI System Aids Accurate Detection of Diabetic Retinopathy

June 02, 2021      Author:

Recently, the team led by Prof. Jia Weiping of Department of Endocrinology and Metabolism, Sixth People's Hospital Affiliated to Shanghai Jiao Tong University, Shanghai Diabetes Research Institute and Shanghai Diabetes Clinical Medical Center, published a research article on the website of Nature Communications, titled "A deep learning system for detecting diabetes mellitus across the disease spectrum". Based on the world's largest fundus image database, the research team developed an innovative migration-enhanced multi-task learning framework and an auxiliary intelligent diagnosis system for diabetic retinopathy called "DeepDR". This system automatically diagnoses the whole course of diabetic retinopathy from the mild to the proliferative stage and provides real-time feedback on the quality of fundus images, as well as detection and segmentation of fundus diseases.

Compared with previous studies, DeepDR has achieved high sensitivity and specificity in the diagnosis of mild retinopathy. Besides, DeepDR provides not only grading but also visual cues to help users identify the existence and location of different types of lesions, imitating the thinking process of ophthalmologists as much as possible. This system will be an important option to reduce the diagnosis difficulty and workload of doctors at primary positions. With great potential in improving the accessibility and efficiency of fundus photography screening, it also blazes a new path for the management and targeted control of chronic diabetes and has been applied to clinical auxiliary diagnosis in many medical institutes and organizations.

This research is led by Prof. Jia Weiping and completed in cooperation with Prof. Sheng Bin from the Department of Computer Science and Engineering of SJTU and Prof. Zou Haidong, Executive Director of Shanghai Eye Disease Prevention and Control Center. The research is supported by "Three-year Action Plan of Shanghai Public Health System Construction", "International Cooperation and Exchange Project of National Natural Science Foundation of China", and the "BRI International Joint Laboratory Construction Project of Shanghai Science and Technology Commission".

The core research results have been granted 3 invention patents by China and 1 by the United States, which have been used in many hospitals across the country, exhibited at Shanghai Industry Fair and received special reports from CCTV International Channel and other media. DeepDR has been applied to the screening project of diabetic retinopathy in low- and middle-income countries of the International Diabetes Federation, covering more than 40 countries and regions, providing artificial intelligence solutions with Chinese characteristics for global management and prevention of diabetes.


Source: Sixth People's Hospital Affiliated to SJTU School of Medicine

Translated by Zhou Rong

Proofread by Xiao Yangning, Fu Yuhe



Retinal screening contributes to early detection of diabetic retinopathy and timely treatment. To facilitate the screening process, we develop a deep learning system, named DeepDR (Deep-learning Diabetic Retinopathy), that can detect early-to-late stages of diabetic retinopathy. DeepDR is trained for real-time image quality assessment, lesion detection and grading using 466,247 fundus images from 121,342 patients with diabetes. Evaluation is performed on a local dataset with 200,136 fundus images from 52,004 patients and three external datasets with a total of 209,322 images. The area under the receiver operating characteristic curves for detecting microaneurysms, cotton-wool spots, hard exudates and hemorrhages are 0.901, 0.941, 0.954 and 0.967, respectively. The grading of diabetic retinopathy as mild, moderate, severe and proliferative achieves area under the curves of 0.943, 0.955, 0.960 and 0.972, respectively. In external validations, the area under the curves for grading range from 0.916 to 0.970, which further supports the system is efficient for diabetic retinopathy grading.