Published Paper Details:

Diabetic Retinopathy, Glaucoma, Cataract, and Age-related Macular Degeneration (AMD) are retinal disorders that cause blindness and irreversible cases of vision impairment in the world. The early detection of a disease is very important to avoid the final loss of the vision permanently, but the traditional methods of screening are time-consuming and require the skill of the specialist. A retinal fundus image analysis framework in the form of an automated multi-disease eye classification framework that employs deep learning is introduced. The system is developed in Python 3.10 and relies on the EfficientNetB0 Convolutional Neural Network architecture to find the optimal balance between the computational performance and the ability to represent the features. ODIR-5K correlates with eight ocular disease types which are used as model training. Image processing involves resizing of images, data normalization, and data augmentation, which involves rotation, zooming, and horizontal flipping of images to improve generalization. Class imbalance is dealt with the help of customized approach to class weighting. The Adam optimizer is used to train the model to ensure that 30 epochs are completed with a Binary Cross-Entropy loss. The performance evaluation shows a high level of accuracy, precision, recall and F1-score in various classes of diseases. Flask-based web app allows the real-time upload of images, prediction of the disease with the confidence probability, and the creation of PDF diagnosis reports that can be downloaded. Results of experiments confirm efficiency and high level of practical implementation in terms of large scale ophthalmic screening and computer aided clinical decision support systems.