Utilizing Advanced Deep Learning for Accurate Multilabel Classification of Ocular Diseases

This study examined the potential of two cutting-edge computer programmes, ResNet50 and Inception V3, to assist in the diagnosis of various eye disorders. To create a wider variety of ocular images for training, they employed a unique technique. Four different eye conditions were depicted in these images: normal, diabetic, cataract- and glaucoma-afflicted. To ensure that there were adequate images of each kind, applied data preprocessing with GAN model and also using a technique known as transfer learning, these programmes were trained to identify eye disorders after initially being trained on a large collection of photographs (ImageNet). This study used two models on the dataset where it has divided into 80% train and 20% test, and then modifying the learning process as necessary. This model reached the accuracy of 89%, which held true for many output parameters such as recall and precision. Additionally, the deep learning techniques performed well on testing using different new data, proving their possibility for dependability in practical settings and then 90% accuracy rate gives positive impact that they can be helpful resources for ophthalmologists.