Classification of Lesions in Retinal Fundus Images for Diabetic Retinopathy Using Transfer Learning

Abstract

Diabetic Retinopathy (DR) is an eye disorder that affects the small blood vessels in the retina and is featured by the presence of different types of lesions in the affected area. If an early prognostication of DR is not done, then it may lead to loss of vision. Some of the diagnosing tools for detection of DR are Indirect Ophthalmoscope, Slit Lamp Examination, Color Photograph, and Optical Coherence Tomography (OCT). The DR dataset contains 5 types of lesions ranging from mild to severe. These lesions are hard to distinguish from each other. The manual diagnosis of DR involves the proper classification of these lesions into their appropriate classes that is quite tedious and error-prone process marred with low level of accuracies. Researchers therefore rely on automatic detection and prognosis of DR based on Machine Learning (ML) methods. In this work, we propose a Deep Learning (DL) framework to classify these lesions with high level of accuracy. To accomplish the task, we train a DL model Visual Geometry Group (VGG19) on Indian Diabetic Retinopathy Image Dataset (IDRID) dataset to extract the features from the color fundus eye photography provided in the dataset. The extracted features are then fed into different classifiers such as Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbors (KNN) etc. to classify the lesions properly. We thus show, that using DL along with Transfer Learning (TL) can classify the affected areas such as Microaneurysms (MA), Soft Exudates (SE), Hard Exudates (EX), and Hemorrhage (HE) of a DR eye with high accuracy.