This paper addresses the development of a computer-aided diagnosis (CAD) system for early detection of diabetic retinopathy (DR), a sight threatening disease, using digital fundus photography (DFP). More specifically, the proposed CAD system is intended for detection of microaneurysms (MA) which are the earliest indicators of DR; CAD systems for MA detection involve two stages: coarse segmentation for candidate MA detection and fine segmentation for false positive elimination. The system addresses the common challenges in candidate MA detection, which includes detection of subtle MAs and MAs close to each other and those close to blood vessels which leads to low sensitivity. The system employs four major steps. The first step involves preprocessing of the fundus images, which comprises of shade correction, denoising and intensity normalization. The second step aims at the segmentation of candidate MAs using bottom hat transform, thresholding and hit or miss transformation. The use of modified morphological contrast enhancement and multiple structuring elements (SEs) in the hit or miss transform has improved the detection rate of MAs. The proposed method has been validated using a set of 20 fundus images from the DIARETDB1 database. The Free Response Operating Characteristics (FROC) curve demonstrates that many MAs that are otherwise missed out are detected by the proposed CAD system.
V. Dharani and Dr. Lavanya R., “Improved Microaneurysm Detection in Fundus Images for Diagnosis of Diabetic Retinopathy”, in Advances in Signal Processing and Intelligent Recognition Systems, Cham, 2018.