Dynamic Attention-Guided Convolutional Network for Retinal Vessel Segmentation.

Saved in:

Bibliographic Details
Title:	Dynamic Attention-Guided Convolutional Network for Retinal Vessel Segmentation.
Authors:	Zhang, Xihong¹ zhangxihong2025@163.com
Source:	IAENG International Journal of Computer Science. Jul2026, Vol. 53 Issue 7, p2546-2557. 12p.
Subjects:	Image segmentation, Convolutional neural networks, Fundus oculi, Feature extraction, Image processing, Artificial neural networks
Abstract:	Precise retinal vessel segmentation is a fundamental task in automated fundus image analysis and computeraided diagnosis for various ocular and systemic diseases. Despite the widespread adoption of U-Net-based architectures, accurate vascular delineation remains challenging due to the intricate morphology of thin vessels, low contrast against heterogeneous backgrounds, and inherent imaging noise. These factors often lead to discontinuous predictions and the omission of delicate capillary branches. To address these challenges, we propose DynamicMSAF-Net, a convolutional framework that integrates a Multi-Scale Attention Fusion (MSAF) module and a Layered Multi-Attention Fusion (LMFA) module. Specifically, MSAF adaptively aggregates vascular cues across diverse receptive fields to enhance multi-scale representation learning, while LMFA refines encoder-decoder feature integration through layered attention to reduce redundancy and preserve structural continuity. Experimental results on the DRIVE and CHASE DB1 benchmarks show that DynamicMSAF-Net achieves strong segmentation performance, yielding AUCs of 98.78% and 99.04%, accuracies of 97.04% and 97.60%, and sensitivities of 82.69% and 85.72%, respectively. The results indicate that the proposed method effectively preserves thinvessel topology and maintains high segmentation fidelity under diverse imaging conditions while remaining computationally efficient. [ABSTRACT FROM AUTHOR]
	Copyright of IAENG International Journal of Computer Science is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	Precise retinal vessel segmentation is a fundamental task in automated fundus image analysis and computeraided diagnosis for various ocular and systemic diseases. Despite the widespread adoption of U-Net-based architectures, accurate vascular delineation remains challenging due to the intricate morphology of thin vessels, low contrast against heterogeneous backgrounds, and inherent imaging noise. These factors often lead to discontinuous predictions and the omission of delicate capillary branches. To address these challenges, we propose DynamicMSAF-Net, a convolutional framework that integrates a Multi-Scale Attention Fusion (MSAF) module and a Layered Multi-Attention Fusion (LMFA) module. Specifically, MSAF adaptively aggregates vascular cues across diverse receptive fields to enhance multi-scale representation learning, while LMFA refines encoder-decoder feature integration through layered attention to reduce redundancy and preserve structural continuity. Experimental results on the DRIVE and CHASE DB1 benchmarks show that DynamicMSAF-Net achieves strong segmentation performance, yielding AUCs of 98.78% and 99.04%, accuracies of 97.04% and 97.60%, and sensitivities of 82.69% and 85.72%, respectively. The results indicate that the proposed method effectively preserves thinvessel topology and maintains high segmentation fidelity under diverse imaging conditions while remaining computationally efficient. [ABSTRACT FROM AUTHOR]
ISSN:	1819656X