Ocular Disease Identification Using Fundus Images

Abstract

Fundus problems are the most typical causes of blindness in people globally. The ophthalmic disease is noteworthy since it has features that are irreversible and might result in long-term blindness. Early diagnosis of eye conditions may enable us to avoid visual impairment in clinical settings. The number of patients and opthalmologists however, is vastly out proportion. The majority of the
identification models in use exclusively concentrate on one particular ocular disease. Therefore, our goal is to create a model for automatically classifying many ocular diseases using fundus photos as input and reporting the disease’s name when it is present. There are currently a number of models available, but the benefit is that we are using multi-labeled picture datasets as opposed to binarylabeled data. The multi-labelled data covers disorders such as age-related macular degeneration (AMD), diabetes, glucoma, pathological myopia, and hypertension. The dataset for these classes is, however, rather uneven. The research advised tackling this multi-class classification challenge and using the same amount of photos to train the model unbiasedly in order to address this issue. Then, VGG-16 was used to train multi-class classifications. ResNet-50 model accuracy is 70%, compared to 75% for the VGG-16 model.