Detection of Retinopathy of Prematurity Stages Utilizing Deep Neural Networks

Abstract

Retinopathy of prematurity is the leading cause of blindness in children around the world. This paper exhibited ten deep convolutional neural networks (DCNNs) models to detect ROP stages in fundus images using deep neural networks. A dataset of 3720 fundus images was collected from the private clinic Al-Amal eye centre, which consisted of 3 classes of ROP stages. A training dataset and a test dataset were created from the images. VGG16, ResNet50, ResNet101, ResNet152, SqueezNet1_0, SqueezNet1_1, DenseNet121, DenseNet169, AlexNet169, and Inception_v3 were trained to make differential diagnoses and then tested. The classification accuracies for the highest three DCNN (ResNet152, DenseNet169, Inception_v3) were 73.95, 77.14, and 99.50%, respectively.To conclude, after training with an extensive dataset, the Inception v3 DCNN model presented large potential in facilitating the diagnosis of ROP stages utilizing fundus images.