Ai-Driven Multi-Disease Detection In Retinal Imaging

Abstract

Ophthalmology must detect retinal disorders as early as possible to prevent irreversible visual impairment (including blindness). This thesis offers a deep learning method to automate the detection of 45 retinal diseases using AI. We developed a strong CNN model which can classify retinal fundus images into several diseases using transfer learning with EfficientNetB0. Transfer learning techniques were used on a set of advanced neural network architectures for carrying out the research focussing onEfficientNetB0, VGG16, ResNet50 and DenseNet121 models. The models got training on a diverse set of data and were evaluated on important performance evaluation measures such as the accuracy of the test, loss percentage, precision, recall, and the F1 score. The EfficientNetB0 and DenseNet121 models achieved the best AUC score and also had the best test loss score. The results are designed to help ophthalmologists examine and diagnose diseases accurately and on time if possible, especially in resource-poor countries. They also help develop smart diagnostic systems for medical imaging