Automatic Microaneurysm Detection in Retinal Fundus Images by Density Gradient Vector Concentration

Abstract

Early detection of diabetic retinopathy (DR) is important because it may cause visual loss. In this paper, we propose an automatic detection technique of microaneurysms (MAs) which are early signs of DR from retinal fundus images by density gradient vector concentration. After image preprocessing, the density gradient vector concentration was calculated in green-channel component and MA candidates were detected by adaptive thresholding. Forty eight features based on texture analysis, pixel value, and shape were calculated from each candidate and the candidates were classified into MAs or false positives by thresholding and a support vector machine. The average sensitivity at selected false positive rates by the proposed method was 0.395 in the evaluation with ROC (Retinopathy Online Challenge) database. The proposed method may be useful in automatic MA detection.