Automatic Detection of Microaneurysms from Digital Fundus Images using LBP Features

Abstract

The human eye is most often affected by Diabetes. It deteriorates the functioning of the retina. Retinal disorders such as Glaucoma, Macular Edema and Diabetic Retinopathy are frequently found among diabetic people. Microaneurysms are the earliest symptoms of Diabetic Retinopathy. Detecting these symptoms in early stages and its treatment can avoid vision loss among patients. The algorithm proposed in this paper is based on morphological operations eliminating blood vessels and the optic disc, followed by the detection of Microaneurysms. LBP features are extracted to identify the Microaneurysms and various classifiers are tested to facilitate tasks such as classification, detection and recognition. Various classifiers are tested on public fundus databases with the proposed method. However, experimental results have shown that decision tree classifier yields better accuracy than other classifiers on most public fundus databases. For the STARE database, the Decision tree classifier yielded values for accuracy 1.0, recall 1.0, and precision 1.0. Similarly, accuracy 0.93, recall 0.86, and precision 0.95 were computed for the e-Ophtha database and accuracy 0.97, recall 1.0, and precision 0.97 for the DIARETDB1 database.