Analysis of CNN Models for Eye Disease Classification Using Retinal Images

The main goal of our study is to create an automated system that combines retinal pictures and convolutional neural networks (CNNs) to classify eye illnesses. We investigate distinct CNN designs, such as VGG19, ResNet50, EfficientNetB0, MobileNetV2, InceptionV3, and DenseNet121, by utilizing the TensorFlow and Keras packages. Transfer learning allows us to fine-tune pretrained models into four categories of retinal pictures classification: normal condition, glaucoma, diabetic retinopathy, and cataract. With a dataset of labeled retinal pictures, the models are assessed and trained, and notable accuracy is attained. With a remarkable accuracy of 90.95%, EfficientNetB0 emerges as the best-performing model. The objective of our initiative is to lessen the strain on healthcare systems caused by manual screening and diagnosis while also facilitating early detection and diagnosis of eye illnesses.