Scale-specific auxiliary multi-task contrastive learning for deep liver vessel segmentation

Amine Sadikine; Bogdan Badic; Jean-Pierre Tasu; Vincent Noblet; Pascal Ballet; Dimitris Visvikis; Pierre-Henri Conze

eess.IVSep 18, 2024

Scale-specific auxiliary multi-task contrastive learning for deep liver vessel segmentation

Authors: Amine Sadikine, Bogdan Badic, Jean-Pierre Tasu, Vincent Noblet, Pascal Ballet, Dimitris Visvikis, Pierre-Henri Conze

Organizations: LaTIM UMR 1101, Inserm, Brest, France · University of Western Brittany, Brest, France · University Hospital of Brest, Brest, France · University Hospital of Poitiers, Poitiers, France · ICube UMR 7357, CNRS, Strasbourg, France · IMT Atlantique, Brest, France

Abstract

Extracting hepatic vessels from abdominal images is of high interest for clinicians since it allows to divide the liver into functionally-independent Couinaud segments. In this respect, an automated liver blood vessel extraction is widely summoned. Despite the significant growth in performance of semantic segmentation methodologies, preserving the complex multi-scale geometry of main vessels and ramifications remains a major challenge. This paper provides a new deep supervised approach for vessel segmentation, with a strong focus on representations arising from the different scales inherent to the vascular tree geometry. In particular, we propose a new clustering technique to decompose the tree into various scale levels, from tiny to large vessels. Then, we extend standard 3D UNet to multi-task learning by incorporating scale-specific auxiliary tasks and contrastive learning to encourage the discrimination between scales in the shared representation. Promising results, depicted in several evaluation metrics, are revealed on the public 3D-IRCADb dataset.

Figures & tables

Figure 1: Proposed pipeline for deep liver vessel segmentation using scale-specific auxiliary multi-task contrastive learning.

Panel of Figure 2 — Figure 2: ( a ) Visual aid to interpret notations related to multi-scale vessel clustering (Sect. 2.1 ). ( b ) Example of 3-scale vasculature clustering applied on a synthetic vasculature [ 8 ] .

Figure 3: Encoder-decoder architecture involved in our multi-task contrastive pipeline. PReLu stands for Parametric ReLU.

Models		DSC $\uparrow$ score ( $\%$ )	Jacc $\uparrow$ score ( $\%$ )	clDSC $\uparrow$ score ( $\%$ )	HD $\downarrow$ dist. ( $\mathrm{mm}$ )
3D ResUNet	binary	$54.25\pm 3.65$	$37.65\pm 3.46$	$46.84\pm 3.63$	$\boldsymbol{80.09}\pm 4.83$
3D ResUNet	multi-class	$50.31\pm 4.43$	$34.13\pm 3.98$	$44.34\pm 2.89$	$114.3\pm 10.9$
Ours	$\lambda_{c}=0$	$54.27\pm 4.16$	$37.73\pm 3.87$	$46.90\pm 2.44$	$86.07\pm 8.07$
Ours	$\lambda_{c}\neq 0$	$\boldsymbol{55.04}\pm 2.79$	$\boldsymbol{38.50}\pm 2.71$	$\boldsymbol{48.13}\pm 3.52$	$82.36\pm 12.3$

Table 1: Liver vessel segmentation on CT scans from the 3D-IRCADb [ 16 ] dataset using 3D ResUNet in binary and multi-class settings as well as the proposed approach without and with contrastive learning. Best results are displayed in bold.

Figure 4: Letter-value plot of branch radii

\hat{r}_{j}

(\mathrm{mm})

per ground truth volume

\boldsymbol{y}_{i}

from the 3D-IRCADb [ 16 ] dataset.

Explore similar work

VasGuideNet: Vascular Topology-Guided Couinaud Liver Segmentation with Structural Contrastive Loss
Feb 25, 2026Chaojie Shen, Jingjun Gu, Zihao Zhao +4Medical Image Segmentation 3D Medical Image Segmentation
Pujian Medical Technology Co., Ltd., Hangzhou, China · Keyi College of Zhejiang Sci-tech University, ShaoXing, China · College of Computer Science and Technology, Zhejiang University, Hangzhou, China
VEELA: A Clinically-Constrained Benchmark for Liver Vessel Segmentation in Computed Tomography Angiography
May 21, 2026Ziya Ata Yazıcı, N. Sinem Gezer, İlkay Öksüz +19Medical Image Segmentation Computed Tomography
Huawei Türkiye R&D Center, Istanbul, 34768, Türkiye · Department of Radiology, Dokuz Eylul University, Izmir, 35100, Türkiye · Department of Computer Engineering, Istanbul Technical University, Istanbul, 34469, Türkiye +8
SegKAN: High-Resolution Medical Image Segmentation with Long-Distance Dependencies
Dec 28, 2024Shengbo Tan, Rundong Xue, Shipeng Luo +7Image Segmentation Medical Image Segmentation
College of Computer Science and Artificial Intelligence, Southwest Minzu University, Chengdu 610225, China · Institute of Artificial Intelligence and Robotics, Xi’an Jiaotong University, Xi’an 710049, China · College of Mechanical and Electrical Engineering, Northeast Forestry University, Harbin 150040, China +4

Feb 25, 2026J/K move · Enter open · S save

VasGuideNet: Vascular Topology-Guided Couinaud Liver Segmentation with Structural Contrastive Loss

Chaojie Shen, Jingjun Gu, Zihao Zhao, Ruocheng Li+3

Pujian Medical Technology Co., Ltd., Hangzhou, China · Keyi College of Zhejiang Sci-tech University, ShaoXing, China · College of Computer Science and Technology, Zhejiang University, Hangzhou, China

Medical Image Segmentation 3D Medical Image Segmentation Topology-Preserving Image Segmentation Convolutional Neural Networks

Open paper AskPDF arXiv

Accurate Couinaud liver segmentation is critical for preoperative surgical planning and tumor localization.However, existing methods primarily rely on image intensity and spatial location cues, without explicitly modeling vascular topology. As a result, they often produce indistinct boundaries near vessels and show limited generalization under anatomical variability.We propose VasGuideNet, the first Couinaud segmentation framework explicitly guided by vascular topology. Specifically, skeletonized vessels, Euclidean distance transform (EDT)--derived geometry, and k-nearest neighbor (kNN) connectivity are encoded into topology features using Graph Convolutional Networks (GCNs). These features are then injected into a 3D encoder--decoder backbone via a cross-attention fusion module. To further improve inter-class separability and anatomical consistency, we introduce a Structural Contrastive Loss (SCL) with a global memory bank.On Task08_HepaticVessel and our private LASSD dataset, VasGuideNet achieves Dice scores of 83.68% and 76.65% with RVDs of 1.68 and 7.08, respectively. It consistently outperforms representative baselines including UNETR, Swin UNETR, and G-UNETR++, delivering higher Dice/mIoU and lower RVD across datasets, demonstrating its effectiveness for anatomically consistent segmentation. Code is available at https://github.com/Qacket/VasGuideNet.git.

Abstract

Explore similar work

VasGuideNet: Vascular Topology-Guided Couinaud Liver Segmentation with Structural Contrastive Loss

VEELA: A Clinically-Constrained Benchmark for Liver Vessel Segmentation in Computed Tomography Angiography

SegKAN: High-Resolution Medical Image Segmentation with Long-Distance Dependencies