Transfer learning approach for diabetic retinopathy detection using residual network

Abstract

The condition known as diabetic retinopathy is a severe and common complication of diabetes. It affects the retina, which is a light-sensitive organ inside the eye, and it can lead to blindness or loss of vision. It is therefore important to improve the diagnosis and classification of this disorder. In deep learning, transfer learning is a process that aims to improve the performance of a given task by taking advantage of the knowledge that has been acquired from another. The main idea of this approach is to speed up the learning process by applying the obtained knowledge to a new task. In this paper, with the help of migration learning, a pre-trained deep learning model known as InceptionV3 was used to classify fundus images the Diabetic Retinopathy 2015 Data Colored Resized database in five categories according to the severity of the lesions. It was able to achieve a 92.314% accuracy on a test set.