Analysis and Automatic Detection of Microaneurysms in Diabetic Retinopathy using transfer learning

Abstract

For the quick identification of severe disorders like diabetic retinopathy (DR), the recognition of retinal microaneurysms (MA) is absolutely necessary. As a result of diabetics, where it creates lesions on the retina of the eyes, DR is a degenerative condition which affects the eyes. The challenging task is the detection of lesions in retinopathy in retinal images. To properly maintain the patient's eyesight, early identification of diabetic retinopathy is essential. The biggest problem with DR detection is that physical detection requires an ophthalmologist to examine retinal fundus pictures of the eye, which takes a lot of time, money, and effort. Various machine learning techniques have been used in literature reviews to identify DR. Classification and feature extraction are the two main steps involved. Blood vessels segmentation and Identification of lesions are two approaches that are involved in the detection of DR. Based on metrics including sensitivity, specificity and accuracy this study evaluates the experimental outcomes of several machine learning algorithms. The proposed method is combination of Convolutional Neural Network along with a machine learning algorithm (Random Forest). To compare the findings, the best of 4 approach has been analyzed with 2 different datasets E-Ophtha and DIARETDB1. The model with InceptionV3 which has soft-max classification without transfer learning acquires the best accuracy, sensitivity and specificity.