Image Filtering and Utilization of Deep Learning Algorithms to Detect the Severity of Diabetic Retinopathy

Diabetic retinopathy is a leading cause of blindness in adults, particularly in those with poorly controlled diabetes. It affects around 93 million people globally and is classified into two main categories: nonproliferative and proliferative. Early detection and treatment are crucial to prevent vision loss, but the current gold standard for diagnosis, fundus photography, has limitations such as requiring specialized equipment and trained personnel, and patient cooperation. To address these issues, this research develops a web application that uses CNN algorithms to effectively diagnose the severity of diabetic retinopathy. The web application, FundusNet, was developed through a multi-phase process using transfer learning and a dataset of fundus images. The proposed hybrid model (based on EfficientNETB7 and VGG-16 using Laplacian Sharpening) was found to have an accuracy of 91.8% and a F1 score of 91.5. FundusNet is free, easy to use, and does not require internet access, making it a convenient and scalable solution for diabetic retinopathy detection, especially in resource-limited settings.