MobileNetV3 based Classification Model for Diabetic Retinopathy

Diabetic retinopathy (DR) is a disease that affects the eyes and is caused by microvascular diabetic damage. Due to this disease, the blood vessels of the retina are compromised. Late detection of this disease increases the likelihood that the patient will lose his/her eyesight. With the rapid advancement of deep learning, it has been used in many fields, including analysis and classification of medical images. The aim of this study is to employ a recently developed convolutional neural network (CNN) classifier that can identify the retinopathy and develop a five class classification method which can categorize the retinopathy into five different stages: no DR, mild, moderate, severe and proliferative retinopathy. In this work, a MobileNetV3 training model is implemented using cross-validation training and testing with a set of 5590 images to identify the stages and patterns of diabetic retinopathy. MobileNetV3 model as a training substrate is used to transfer what has been learned to a pre-trained five class classification task which achieves the accuracy of 98.56%. In order to support the medical experts in assessing the influence of the retina at different stages, the corresponding thermal image is also provided.