Optimized Vessel Segmentation: A Structure-Agnostic Approach with Small Vessel Enhancement and Morphological Correction

TL;DR

This work proposes an optimized vessel segmentation framework: a structure-agnostic approach incorporating small vessel enhancement and morphological correction for multi-modality vessel segmentation, and demonstrates that this approach achieves superior segmentation accuracy, generalization, and a 34.6% improvement in connectivity.

Abstract

Accurate segmentation of blood vessels is essential for various clinical assessments and postoperative analyses. However, the inherent challenges of vascular imaging, such as sparsity, fine granularity, low contrast, data distribution variability, and the critical need for preserving topological structure, making generalized vessel segmentation particularly complex. While specialized segmentation methods have been developed for specific anatomical regions, their over-reliance on tailored models hinders broader applicability and generalization. General-purpose segmentation models introduced in medical imaging often fail to address critical vascular characteristics, including the connectivity of segmentation results. To overcome these limitations, we propose an optimized vessel segmentation framework: a structure-agnostic approach incorporating small vessel enhancement and morphological correction for multi-modality vessel segmentation. To train and validate this framework, we compiled a comprehensive multi-modality dataset spanning 17 datasets and benchmarked our model against six SAM-based methods and 17 expert models. The results demonstrate that our approach achieves superior segmentation accuracy, generalization, and a 34.6% improvement in connectivity, underscoring its clinical potential. An ablation study further validates the effectiveness of the proposed improvements. We will release the code and dataset at github following the publication of this work.

Figures (7)

• Figure 1: The distribution of vascular imaging data shows significant variation across different anatomical regions and acquisition protocols, posing challenges for the generalization of task-specific expert models.
• Figure 2: Framework diagram of OVS-Net, which comprises five main modules: the ViT-based macro vessel extraction module, the CNN-based micro vessel enhancement module, the mask prediction module, the prompt encoder and the morphological correction-based post-processing network.
• Figure 3: Comparison with other SAM-based methods. From top to bottom, the images are from XCAD, ORVS, DrSAM and OCTA500-3M datasets. Due to the micro vessel enhancement module and the post-processing network, our method is better able to handle detailed information, resulting in superior performance in processing micro-vessels and maintaining vessel connectivity compared to other methods.
• Figure 4: Radar chart of scores on complex vascular datasets compared with the suboptimal SAM model.
• Figure 5: Bar chart comparing the scores of OVS-Net and the expert model. Here we use the standard deviation as the error bar standard. For expert model selection, we select the expert model with the highest Dice score to plot the bar chart.