Detection of Diabetic Retinopathy (DR) Severity from Fundus Photographs: An Ensemble Approach Using Weighted Average

Diabetic retinopathy is a common diabetic disease that affects the retina and can result to blindness if not treated initially. Deep learning (DL)-based models are proposed to detect the blood abnormalities in the retinal tissue due to diabetes mellitus obtained from fundus camera. The drawback with these models is the lack of performance. To address this, we propose to automate the process of detection of severity of diabetic retinopathy (DR) using ensembles of pretrained models, thus exploring the power of transfer learning in the field of automated diagnosis. Deep learning models perform well when the model is trained on a large amount of data. In this regard, we also put forth data augmentation and preprocessing techniques to generate the synthetic images and to improve image quality. Extensive experimental results on publicly available database illustrate that the proposed ensemble model achieves fair accuracy when compared to existing models. Thus, the proposed model shows good scope for deployment in real-time diagnosis.