Detection and Segmentation of Retinopathy Diseases using EAD-Net with Fundus Images

Abstract

Diabetic retinopathy is one of the major concerns affecting most of the population. It causes the injury in the blood vessels of the retina which is very sensitive to light and is located at the back of the eye, causes this. In the early stages, it may did not cause any symptoms or it will be only minor vision problems. When blood vessels are damaged, they can leak, causing dark spots to appear in our vision. The Diabetic Retinopathy (DR) can be recognized by presence of Hard Exudate, Soft Exudates, Microaneurysms and Haemorrhages. The most important aspect is accurate detection of diseases at an early stage. Manually annotating these scratches is a significant task in clinical survey. This method is completely based on the convolutional neural network and further that can be classified into three modules attention module, encoder module and decoder module. The fundus images were normalised and augmented before sent to the EAD-Net for pixel-wise label forecast and for Self-operating feature extraction. After pre-processing, the image is sent into the EAD Net for training which is followed by testing of an image and finally the segmentation of the image will be done. optimizer is used here is Adam and categorical CE as the loss function. This EAD-Net is the novel method for diagnosing different stages of DR. It produces fitting results with an accuracy of 95 percentage when segmenting 4 different lesions. These active segmentations have significant clinical implications in the monitoring and in the diagnosis of DR.