Microaneurysm detection by multiple feature subset extraction and selection based on SVM-weights and Genetic Algorithm-Neural Network

Abstract

Diabetic Retinopathy(DR) is one of the prevalent eye disorder associated with diabetes. At the outset DR signs are Microaneurysms(MA) due to retinal vessel leakage. This article proposes an automated identification of the microaneurysm regions in retinal fundus images with an optimised sub-set of features. A set of pre-processing techniques are employed to enable image more effective for MAs recognition. Patches were made in order to identify microaneurysm from enhanced images. Texture features such as LBP and GLCM were extracted, subsets were generated based on SVM weights. The Genitic Algorithm-Neural Network(GA-NN) selection method is extended to find optimised subset of features. Using optimised subset features, independent classification is done using KNN, SVM and DT classifiers. By majority vote of each classificator, the final decision system projects the class mark as MA region or non-MA region. The suggested approach has been tested on e-ophtha database. The results demonstrate the potential to distinguish microaneurysms at the patch level.