Enhanced Method for Classification of Diabetic Retinopathy using Capsule Networks

Nowadays, one of the most common causes of blindness in diabetics is Diabetic Retinopathy. A favorable prognosis depends on the problem's early and prompt detection. The goals of automated systems for diabetic retinopathy are to increase productivity, expedite the screening procedure, and guarantee that people with diabetes receive accurate and timely evaluations of their eye health. The researchers were motivated to use deep learning capsule networks for diagnosing Diabetic Retinopathy due to their notable advantages over conventional machine learning techniques in a range of applications. This paper develops a designed capsule network for Diabetic Retinopathy Classification. Convolutional layers in a deep learning architecture would be used to extract low-level features like edges and textures from fundus images. The retina's more intricate spatial relationships and structures would be captured by the primary capsule layer. Based on the features that were extracted, the class capsule layer and SoftMax layer would then collaborate to assign probabilities to various classes of diabetic retinopathy, resulting in a classification or grading for the severity of the ailment. The proposed network's effectiveness is confirmed in terms of four performance metrics by taking the Messidor-2 dataset into account. The developed capsule network obtained prominent results of Accuracy of 95.41, Precision of 98.14, Recall of 92.23, F<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> Score of 94.91 when compared to the existing models