Optimization-Based Fundus Image Decomposition for Diagnosis Support of Diabetic Retinopathy

Abstract

Diabetes mellitus often leads to a serious eye disease called diabetic retinopathy, which is one major cause of blindness among adults. Since this blindness can be prevented if the diabetic retinopathy is detected at an early stage and appropriate medical treatment is provided, routine screening tests with fundus images are very important. However, as the number of diabetic patients increases, the routine screening tests are becoming big burdens for ophthalmologists. To reduce these burdens, in this paper, we propose a diagnosis support method by using convex optimization. The proposed method decomposes a green channel fundus image into a basic image composed of non-disease parts, a positive image including exudates, and a negative image including hemorrhages. Numerical experiments show the effectiveness of our method.