Microaneurysms Extraction with Vessel Neighborhood Separation, SVM, and Connected Component Extraction

Diabetic retinopathy is an important branch of ophthalmology. Non - proliferative diabetic retinopathy is used to detect Microaneurysms in the early stage. Microaneurysms are verified through fundus images, where in the fine red-dots near the blood vessels confirm this defect. Conventional methods and their weak resolution seldom can identify to such accuracies. In this work, we present a procedure to identify Microaneurysms with higher accuracy. The retinal vessels are extracted, from collected fundus image, using a Gabor wavelet which delivers high accuracy output. For accurate analysis the image it is sub divided into two regions, neighborhood and non-vessel neighborhood for expediting support vector machine (SVM) analysis. Further the SVM engine is trained for positive and negative samples of identified region fundus images. Then by sliding window technique, the entire test image is analyzed limiting analysis by SVM engine for near vessel region. This improves overall performance of the analysis and permits time available for a deeper/ sensitivity analysis of near vessel areas. The logic and the code has been tested on sample images and the results have been satisfactory.