Fundus images classification for Diabetic Retinopathy using machine learning technique

Diabetic retinopathy (DR) is a frequent consequence of diabetes that, if left untreated, can result in visual loss. Early identification and treatment are essential to avoid vision loss, and Fundus imaging is a popular technique for DR screening. In this study, we provide a deep-learning algorithm for categorizing fundus photos to identify DR. Our model, which employs a convolutional neural network (CNN) architecture, was developed using a sizable collection of fundus images from several sources. We use various picture preparation methods to enhance image quality and employ data augmentation approaches to expand and diversify our training dataset. Our technique successfully categorizes fundus pictures into several DR phases with high accuracy, highlighting its potential as a tool for DR screening.Medical data is susceptible and requires strict protection, making it difficult to collect and access. We create small datasets due to limited resources. With these limited data sources, we propose a novel method for extracting fundus image features, improving classification performance.A study to improve the efficiency of diabetic retinopathy image classifications utilizing various pre-processing and augmentation strategies to discover superior outcomes compared to the related studies in recent years. We utilized five pre-trained models to compare their performance. Using ANN, the elastic deformation techniques VGG16, VGG19, ResNet50, ResNet152, and InceptionV3 achieved adequate validation accuracy at 96%, 99%, 99%, 86%, and 93% respectively. Using five pre-trained models and CNN, photometric augmentation techniques obtain adequate validation accuracy at 98%, 99% , 99% , and 86% respectively.