Deep Learning Framework in Fundus Image Analysis for Diabetic Retinopathy

Abstract

Adults over the age of forty now commonly suffer from diabetes. Diabetes that has been present for a long time can lead to complications such as diabetic retinopathy (DR), which impairs eyesight. To properly preserve the individual s vision, diabetic retinopathy must be identified early. The biggest problem with DR detection is that manual diagnosis requires a lot of effort, money, and time and requires an ophthalmol- ogist to examine retinal fundus images of the patient s eyes. This research provides an autonomous diagnostic system based on a hybrid deep and machine learning technique, to overcome this issue. Based on the LuNet network, the suggestions involves the seg- mentation of lesions in retinal pictures. Next, global and local features are extracted using a Refined Attention Pyramid Network (RAPNet). The Aquila Optimizer (AO) technique is used to choose the unique features from the extracted feature set in order to improve the classifier s performance. Lastly, the input image is categorized according to its severity using the LightGBM model. The effectiveness of the suggested frame- work has been examined in a number of studies using three publicly accessible datasets (MESSIDOR, APTOS, and IDRiD). Furthermore, the intricacy of the tumor character- istics has made it difficult to classify the DR intensity level. A screening procedure needs an effective identification method to classify the subtle diseases of the retina. In order to diagnose eye conditions and allow optometrist to treat patients right away, deep neural networks are crucial. Using photos from the global IDRiD, MESSIDOR, and KAGGLE datasets, this study offers an efficient Hybrid Optimized Deep Learning Network (HODLNet) model for classifying the extent of DR. The preprocessed retinal images that were acquired from the subpar fundus images are the first to undergo the segmentation process. After that, an improved ResUNet model is utilized to segment the circulatory veins and the optic disc. Following this research will eventually be expanded u