Cataract Disease Detection in Human Eye Image Using ResNet-50 Method

Abstract

Cataract is a leading cause of blindness that requires quick and accurate diagnosis to prevent further deterioration in vision quality. However, conventional examination methods often require a long time and specialized expertise, making them difficult to access widely. Along with technological developments, digital image processing offers a solution to detect cataracts more efficiently. This research aims to develop an image processing-based cataract identification system using a deep learning approach through the ResNet-50 architecture for pattern recognition in eye images. The research process includes image matrix transformation and file compression to improve data processing efficiency. Eye image datasets are used as training and testing data in the classification process using first-order parameters and 100 epochs. Test results showed the system was able to identify cataracts with an accuracy of 95.7% and the best computation time of 1.888 seconds, using 400 training data and 381 validation data. The resulting software simulation can be a digital image-based cataract early diagnosis tool, which is expected to support medical personnel in providing faster treatment and expanding access to eye health services.