Cataract Detection Using Deep Learning in Teleophthalmology Systems

Blindness and visual impairment are urgent issues that reduce the quality of human resources and productivity while also costing much money. This study developed a classification system to simplify classifying cataracts using the Convolutional Neural Network (CNN) Deep Residual Network (ResNet) model. Roboflow datasets were used and categorized as Mature, immature, and normal eye cataract images. After that, the preprocessing data included image cropping, resizing, and image augmenting before being trained using the ResNet CNN model. Several experiments were carried out in this study to obtain optimal results, including a comparison of the ResNet50 and ResNet101 architectural models. The ResNet50 model produced the best results, with an average accuracy of 92.71%, sensitivity of 93.25%, and specificity of 96.16%.