Utility of a radial basis function classifier in the detection of red lesions in retinal images

Abstract

Diabetic Retinopathy (DR) has become an important cause of visual impairment among people of working age in industrialized countries. The aim of this study is to automatically detect some of the DR clinical signs: red lesions (RLs), like hemorrhages (HEs) and microaneurysms (MAs). As they appear as early markers of DR, their detection could be an important contribution to the diagnosis and follow-up of the disease. For this task, we extracted a set of features from image regions and selected the subset which best discriminated between RLs and the retinal background. A radial basis function (RBF) classifier was subsequently used to obtain the final segmentation of RLs. Our database contained 115 images with variable color, brightness, and quality. 50 of them were used to train the RBF classifier. The remaining 65 images were used to test the performance of the method. Using a lesion based criterion, we reached a mean sensitivity of 86.0% and a mean positive predictive value of 52.0%. With an image-based criterion, we achieved a 100% mean sensitivity, 56.0% mean specificity and 83.1% mean accuracy. © 2009 Springer-Verlag.