Enhanced retinal blood vessels segmentation using deep learning and residual network

Abstract

Accurate segmentation of retinal vessels is essential for the early detection and diagnosis of various ocular and systemic diseases, such as diabetic retinopathy, glaucoma, and hypertension. In this study, the effectiveness of employing a U-Net-based deep learning architecture for retinal vessel segmentation has been investigated, the proposed approach incorporates residual blocks into the U-Net architecture, which introduce refined feature learning by leveraging shortcut connections to mitigate the vanishing gradient problem and improve convergence, the particular encoder-decoder structure of the U-net model and skip connections, is particularly well-suited for biomedical image segmentation tasks due to its ability to capture fine-grained details while preserving global contextual information. In this experiment, retinal scans from the DRIVE, CHASE_DB1, and HRF datasets, were processed using the contrast limited histogram equalization (CLAHE), in addition to various preprocessing and augmentation techniques to enhance the model generalization. The obtained scores of the proposed workflow demonstrate that the U-Net model with residual blocks achieves state-of-the-art performance, with high accuracy, precision, Dice and IOU scores, underlining its potential as a robust tool for automated retinal vessel analysis. This research highlights the importance of deep residual learning in advancing clinical diagnostic tools and supports further exploration into improving model efficiency and adaptability to diverse datasets. • Machine learning and AI are revolutionizing retinal image analysis, improving detection, efficiency, and personalized care in ophthalmology. • The study explores a U-Net architecture enhanced with residual blocks to refine feature learning, mitigate vanishing gradients, and improve convergence. • The DRIVE dataset was used, with CLAHE and various augmentation techniques applied to enhance model generalization and performance. • The proposed Residual U-Net model achieves high accuracy, precision, Dice, and IOU scores, demonstrating the effectiveness of deep residual learning for automated retinal vessel analysis.