Dr. Ruogu Fang’s SMILE Lab, in collaboration with Shanghai Jiao Tong University and Shanghai No. Six Hospital in China, won two first place awards at the Diabetic Retinopathy Segmentation and Grading Challenge (IDRiD) organized by the premier International Symposium on Biomedical Imaging (ISBI) 2018 held by Institute of Electrical and Electronic Engineers (IEEE). BME Ph.D. student, Peng Liu, represented the team to participate in the on-site final competition and ranked first in contests of optic disc detection and fovea detection, and third in optic disc segmentation.

Diabetic retinopathy (DR) is the leading cause of blindness globally among working-age population, and is a severe complication of diabetes with high incidence and prevalence. Regular screening and timely treatment can substantially reduce the risk of vision loss and is critical for the prognosis of diabetic patients. The clinical importance of early diagnosis and treatment of DR has made it a research hotspot in the integration of Artificial Intelligence (AI) and medicine.

With the increasing research interest in AI in retinal image analysis, Diabetic Retinopathy Segmentation and Grading Challenge (IDRiD) held by IEEE ISBI attracted a great number of universities and institutions worldwide. The challenge has three sub-challenges: lesion detection, DR grading, and optic disc & fovea detection & segmentation. University of Florida formed the team DeepDR with Shanghai Jiao Tong University and Shanghai No. Six Hospital and participated in sub-challenges of “Optic Disc & Fovea Detection & Segmentation”.

Fang’s team, DeepDR, used high-resolution original images as input into deep learning networks which needed a large amount of network parameters and had high memory demand. However simple compression will lose information. To overcome this challenge, the team proposed a novel multi-task iterative optimization scheme based on coarse-to-find principle, to simultaneous detect the centers of both optic disc and fovea.

In both contests the team outperformed other competitors with large margins. DeepDR only used the dataset provided by the challenge organizer to guarantee the efficacy of their algorithms, and with more data, the computer assisted diagnosis system can achieve higher performance.

The IEEE ISBI is a premier scientific conference dedicated to mathematical, algorithmic, and computational aspects of biological and biomedical imaging, across all scales of observation. It fosters knowledge transfer among different imaging communities and contributes to an integrative approach to biomedical imaging.