Automated detection of diabetic retinopathy using convolutional neural networks on a small dataset

This study presents an automated approach for the detection of diabetic retinopathy (DR) using Convolutional Neural Networks (CNNs) applied to a small dataset. We employed the DenseNet architecture, specifically DenseNet-121, for its proven effectiveness in image classification tasks. The methodology involved several key pre-processing steps, including image resizing, grayscale conversion, noise removal, and contrast enhancement, to improve the quality of the input images. We implemented segmentation techniques to isolate relevant features, followed by data augmentation to enhance the dataset's diversity and robustness. The resulting model was trained to classify images into four distinct categories: Mild DR, Moderate DR, No DR, and Proliferative DR. Each category's classification accuracy was rigorously evaluated, demonstrating the model's capability to effectively identify varying degrees of diabetic retinopathy. The findings highlight the potential of using DenseNet in the automated analysis of retinal images, paving the way for rapid and accurate DR screening in clinical settings, especially in scenarios with limited data availability. This approach not only addresses the challenges posed by small datasets but also contributes to the early detection and management of diabetic retinopathy, ultimately improving patient outcomes in diabetes care.

Keywords: Diabetic Retinopathy Dataset, Pre-Processing, DenseNet, Deep learning, Classification, Accuracy.