A Deep Learning Approach for Diabetic Retinopathy detection using Transfer Learning

Diabetic Retinopathy is a primary complication of diabetes which more often than not, affects both eyes and anyone with type-1 or type-2 diabetes can develop it. A Diabetic patient should undergo eye tests periodically as the pace of development of this condition is slow. A Dataset of Fundus Photographs of retina is considered. Thus, there is a notable value in automatically categorizing the Fundus Photographs. Therefore, to get a consolidated and objective medical diagnosis, this paper proposes a transfer learning based approach for Diabetic Retinopathy categorization. The resizing of the dataset is performed, which converts the varied images into 224x224 format. The images are augmented using AUGMIX and pooled using GeM. Then, we have used pretrained models, namely SEResNeXt32x4d and EfficientNetb3. The pretraining of the aforementioned neural networks has been done on the ImageNet dataset. Then, the Diabetic Retinopathy images are migrated to these models. Based on the dataset already available, the output is ultimately split up into 5 levels according to the seriousness of the degree of DR. The experimental results show that the training accuracy of this method can reach as high as 0.91. Hence, the retina images of the Diabetic and Healthy patients can be easily classified using our proposed methodology, consequently reducing the number of reviews by medical professionals.