Fundus Image Classification for Diabetic Retinopathy Using ResNet50V2 and InceptionV3

Diabetic Retinopathy (DR) is one of the diseases that can lead to blindness. According to WHO data, the number of patients with diabetes increased from 108 million people in 1980 to 422 million in 2014. Diabetic Retinopathy begins with common complications in individuals with diabetes, involving variations in the blood vessels of the retina. These variations can cause bleeding or fluid leakage, disrupting vision. Early detection in diabetes patients will help them find treatment to prevent blindness. Ophthalmologists with professional skills may take a long time to screen for DR patients, but with the help of computation, it can streamline the process and reduce costs for early diagnosis of DR patients. This study compares two deep learning methods used for the detection of Diabetic Retinopathy through classification of fundus retina images. Researchers used ResNet50V2 and InceptionV3 to classify 3 classes of diabetic retinopathy taken from the Kaggle dataset: Normal, DR level 1(Mild & Moderate), and DR level 2 (Severe & Proliferative DR). The models produced showed that the ResNet50V2 model performed better than InceptionV3, achieving an accuracy of 80%, sensitivity of 80%, and specificity of 92%. The ResNet50V2 model also processes data faster for each step compared to InceptionV3, at 12.42 seconds per step.