RETINAL VESSEL DETECTION USING RESIDUAL Y-NET

Abstract

In this Corona Pandemic, Diabetic patients are affected a lot. Unfortunately, due to the consumption of steroids, people are affected by mucormycosis, which is a kind of fungal infection, making this situation worse. Patients get swollen red eyes, and diabetic patients are more vulnerable to it, as they suffer from Diabetic Retinopathy. It has become essential to determine the damage caused to the eye to save patients from vision loss. Only doctors can identify how the condition of the eye by physical examination. But, this is a tricky and time-consuming job. With the help of fundus photography and deep learning algorithms, the detection and classification process will speed up. There are many existing image detection algorithms, but they do not have efficient feature retention and lightweight architecture model. This paper proposes Residual Y-net architecture that works excellently on a balanced medium-size. With the help of segmented features it acquires reliable features which help in classification. It is a very lightweight architecture inspired by U-net, Deep Residual U-net, and Y-net. The addition of residual units in the network has significantly improved the accuracy rate. It is observed that a balanced dataset gives a much accurate performance than an unbalanced dataset. The proposed model's test accuracies on medium-size unbalanced and balanced datasets are 90.39% and 93.60%, respectively.