Harnessing Deep Learning Models in Fundus Imaging for Early Diagnosis in Eye Disorders

Throughout the globe, eye conditions such as diabetic retinopathy, glaucoma, cataract, and age-related macular degeneration are the major causes of blindness and visual impairment. Early detection is crucial to prevent permanent loss of vision. Fundus imaging is important for early detection of these disorders because it produces detailed images of the retina. The project aims to develop an automated system that can detect various eye disorders from these fundus images using deep learning. To build the model, a diverse collection of labeled fundus images was sourced from openly available platforms like Kaggle, EyePACS, and Messidor. These images were carefully prepared before training through processes like resizing, enhancing contrast, and applying basic transformations. The system employs a locally developed Convolutional Neural Network along with transfer learning-based ResNet50. The CNN model is designed to gradually extract important features from the images using standard building blocks like activation functions and pooling techniques, while also including methods to improve stability and prevent overfitting. The model was developed using the Django framework, along with user access to real-time diagnostic feedback through a web-based interface.