Making every microaneurysm count: A hybrid approach to monitor progression of diabetic retinopathy

Previous studies have shown that the progression of diabetic retinopathy (DR) can be monitored by counting the number of microaneurysms (MAs). In this work, we propose an approach to maximize the sensitivity of MA detection by using a combination of two different techniques, namely local thresholding and Canny edge detection. The proposed approach was evaluated with publically available dataset DIARETDB1 where ten fundus images containing a total of 315 MAs were randomly selected. Out of these, the proposed approach was able to detect 295 MAs, achieving a sensitivity of 93.65%. Limiting factors were identified to be noise filtering, presence of artifacts and dark-shaded retinal vessels.