Stratification of the Lesions in Color Fundus Images of Diabetic Retinopathy Patients Using Deep Learning Models and Machine Learning Classifiers

Abstract

Diabetic Retinopathy (DR) is a frequently occurring eye disease that is diagnosed in people suffering from diabetes for a long period of time. Patients who live with diabetes for a long duration of time exhibit mild to severe symptoms of DR. This pathos flourishes gradually and leads to complete blindness over time. The diagnosis of DR is a time-consuming and error-prone process for ophthalmologists due to the pictorial complexities of the images. Therefore, a method based on Machine Learning (ML) and Deep Learning (DL) is proposed to classify the retinal fundus images of the patients into classes based on the severity level of the disease. We employ a CNN architecture armed with the power of deep learning and pre-trained with Transfer Learning to accomplish the task. Outsmarting the already existing approaches, the proposed model functions via extracting a feature vector from the test set of the images which on feeding to classifier models can classify the new images with high accuracy. In our work, we apply various CNN models to extract the features from several diabetic fundus images. The extracted features are provided as the input to various classifiers which as a result classify several lesions accurately. The results show that using deep learning along with transfer learning can accurately classify the fundus images into the right category of lesions.