Harnessing the Strength of ResNet50 to Improve the Ocular Disease Recognition

Abstract

Among the most prevalent eye conditions, cataract is the leading cause of blindness, impairing vision. A cataract is a condition that means clouding of the lens of the eye. Cataract-related blindness can be mainly prevented with early detection and prompt treatment. Artificial intelligence systems that grade cataracts based on fundus pictures are a practical way to help clinicians detect cataracts more accurately. For early detection of cataracts, Convolutional neural networks, also referred to as CNNs, have been reported to have a great deal of promise in several different domains, including the identification of many eye illnesses. In this research, a deep CNN model based on ResNet50 architecture has been proposed to classify the images into cataract-infected and normal classes. For fulfilling this task Ocular Disease Intelligent Recognition dataset has been chosen. This dataset contains real-time patient reports of both eyes. The model has shown a very good accuracy of 95.63% and 90.37% of validation accuracy while using SGD optimizer. The loss was 0.64 which is nominal and this model has shown very promising results in classifying the images. This model has a very innovative approach in the medical field so it can be used as a tool in the biomedical or healthcare field.