Diabetic Macular Edema Grading by Using Deep Learning Algorithms

This research aimed to develop and evaluate convolutional neural network (CNN) models for predicting edema in retinal images using the IDRiD dataset. Edema, a medical condition characterized by the abnormal accumulation of fluid in tissues, is a significant factor in diabetic retinopathy (DR), leading to vision impairment and blindness worldwide. Early detection and accurate prediction of edema in retinal images are crucial for effective management and treatment of diabetic retinopathy. The IDRiD dataset, a publicly available collection of retinal fundus images, was used for model training. The methodology comprised data preprocessing, model development, training, and evaluation using various CNN architectures. VGG16 demonstrated the highest test accuracy (71.25%), outperforming other architectures like VGG19 and ResNet variants. The discussion highlighted the importance of model performance, architecture comparison, robustness, and future optimization strategies. The study's findings contribute to the advancement of automated diagnostic tools for diabetic retinopathy, emphasizing the potential for improved patient care outcomes through accurate edema prediction in retinal imaging.