Deep Learning in Image Classification using VGG-19 and Residual Networks for Cataract Detection

Abstract

Cataracts are often touted as the number one cause of blindness in Indonesia. In fact, referring to data from the World Health Organization (WHO), cataracts account for about 48% of blindness cases in the world and are number one in Indonesia. The research that has been done on cataracts is classified through various objects such as blood vessels, optic disc, the object used is the optical disk in the retinal fundus camera image. The purpose of this study is to produce an automatic cataract early detection application program by classifying cataracts into two categories, normal cataracts, and cataracts. Early examination of cataract patients for people who have less economic capacity such as most of the population in developing countries is considered very helpful. Classification is needed to assist doctors in deciding when to operate on cataract patients. Processing of 1088 patient retinal fundus image data consisting of 500 normal retinal images and 594 cataract images. Furthermore, the classification process is carried out using VGG-19, ResNet-50 and ResNet-101 which is processed with Jupyter Notebook. From the results of training and testing, the average accuracy of VGG19 is 91.06%, ResNet-50 93,50% and ResNet-101 is 93,50% in all retinal classes.