A Multi-Stage Deep Transfer Learning Method for Classification of Diabetic Retinopathy in Retinal images

Diabetic Retinopathy (DR) is one among the most dangerous complications of polygenic disease, and if it is untreated, it can result in permanent disability. Early detection is critical for clinical progress and it is one of the most difficult challenges. Unfortunately, determining the stage of DR is prominently difficult as well as this calls for skillful rendition from humans on the structural images of body. The significance in simplifying this process of detecting cannot be overstated. Convolutional neural networks have been effectually used in a various fields, as well as in the detection of
$D$
R. However, the increasing expense of huge labelled datasets and also variability among physicians, who are completely different block the effectiveness of these strategies. In this perspective, this research work has proposed a deep learning-based automated technique for DR stage classification and detection by utilizing sole photography of the fundus in human retina. Further, the proposed research work has suggested a multi-stage deep transfer learning technique that uses equivalent datasets for including labeling, which is completely different. The methodology suggested by this research work can be utilized as a screening method employed for early diagnosis of DR with a sensitivity and specificity of 0.99 as well as a quadratic weighted kappa grade at 0.925466 of blindness detection dataset.