首页

International Medical Research and Translation(IMRT) August 2025 ,Volume 1, Issue 2

RV-DANet: Retinal Vessel Segmentation Using Vessel-oriented Information Enhancement and Global Relationship Modeling

Wen-Ze Zheng

School of Software Engineering, Xinjiang University, Urumqi 830046, China

Abstract: Accurate retinal vessel segmentation is essential for early ophthalmic diagnosis but remains challenging due to complex vascular structures and low image contrast. [1], [17] This paper proposes RV-DANet (Retinal Vessel Dual- Augmentation Network), a U-Net-based framework incor- porating hierarchically integrated dual-augmentation modules within skip connections. The Vessel-oriented Information En- hancement Module (VIEM) preserves structural details in shallow layers through Feature Dimension Enhancement and Regional Focus Enhancement. Concurrently, the Global Rela- tionship Module (GRM) captures vessel connectivity patterns in deeper layers via Contextual Pattern Encoding and Feature Signiﬁcance Modeling. This complementary design effectively addresses the multi-scale nature of retinal vasculature, where ﬁne capillaries and major vessels coexist. Our network em- ploys a ﬁve-level encoder-decoder architecture with strategi- cally positioned enhancement modules: VIEM enhances vessel boundaries and textures using Selective Information Sampling and Feature Transformation in shallow layers, while GRM ensures vascular continuity through Relationship Inference and Structural Reorganization in deeper layers. Extensive experiments on multiple public datasets demonstrate that our method outperforms state-of-the-art approaches across various evaluation metrics. Ablation studies conﬁrm both modules’ complementary contributions, with VIEM improving capil- lary detection and GRM enhancing vessel continuity. The Dual-Augmentation strategy optimally balances ﬁne capillary preservation with overall structural coherence while maintain- ing computational efﬁciency, offering substantial value for clinical applications and ophthalmic disease diagnosis.

Keywords: Retinal Vessel Segmentation, RV-DANet, Vessel- oriented Information Enhancement, Global Relationship Mod- eling

全文下载

上一期

Research Progress on the Effects of Physical Therapy in Post-Stroke Dysphagia Rehabilitation

Application of artificial intelligence-assisted ultrasound technology in the diagnosis and treatment of breast tumors

下一期

International Medical Research and Translation (IMRT)

International Education and Social Science Exploration (IESSE)

Technological Innovation and Science Management (TISM)

艺术探索与社会实践研究

中国经济与社会科学研究