A. Shaeidi Abstract

Microaneurysms (MAs) identification is a critical step in diabetic retinopathy screening.
Diabetic retinopathy is a major cause of blindness and microaneurysms are the first clinically
observable manifestations of diabetic retinopathy. This paper describes a new digital imageprocessing
algorithm for microaneurysm diagnosis in digital retinal images for diabetic
retinopathy patients. The recorded 90000 pixels have been extracted from 135 color retinal
images which have formed a learning data set of Microaneurysms and non-Microaneurysms.
This our learning data set sample includes 50000 Microaneurysm pixels and 40000 non-
Microaneurysm pixels. After a color normalisation and contrast enhancement preprocessing
step, the color retinal image is segmented using Pixel-Level Microaneurysms Recognition and a
Dynamic Thresholding technique. Then, the segmented regions were classified into two disjoint
classes, Microaneurysms and non-Microaneurysms.
Having high precision, this approach demonstrates that computer assisted diagnosis
removes many problems in the analysis with manual detection. The proposed approach was
realized with full implementation of programming languages. Here, 135 retinal images were
analyzed among which 70 images showed abnormalities (i.e. include DR), while in 65 cases the
conditions were normal. Finally, in step test using classifier neural network, the system
achieved (in percent) 98.5 sensitivity, 96.9 specificity features and 97.7 accuracy.