Application of Transfer Learning Approach for Diabetic Retinopathy Classification

Abstract

Diabetes is a disorder of the metabolism caused by high glucose levels in the body. Diabetes causes eye deficiency, also known as Diabetic Retinopathy (DR), which causes significant vision loss over time. Diabetes patients’ vision can be saved if DR is detected and diagnosed early. Microaneurysms, haemorrhages, and exudates are prior signs of DR that emerge on the surface of retina. Nevertheless, diagnosing DR is a challenging problem that necessitates the services of an experienced ophthalmologist. Using an automated classifier, an artificial intelligence based deep learning can assist the ophthalmologist in providing an expert advice related to the assessment of the DR. A large volume of data is required to effectively train the model for the classification of DR, that is a major constraint in the DR area. Transfer learning is a methodwhich could assist in combating image limitation. The central idea behind transfer learning approach is that, this framework was already trained on large set of images which could be fine-tuned to fit for the required set of data. This paper applied transfer learning based VGG16 and InceptionV3 model for the DR classification on a public benchmark IDRiD dataset (Indian Diabetic Retinopathy Image Dataset). These models are used to address the problem and to maximize the results.