VesselBridge3D: A Foundation Model Adaptation Framework for Label-Efficient 3D Vessel Segmentation

Kirato Yoshihara; Yohei Sugawara; Yuta Tokuoka; Lihang Hong

eess.IVFeb 27, 2026

VesselBridge3D: A Foundation Model Adaptation Framework for Label-Efficient 3D Vessel Segmentation

Authors: Kirato Yoshihara, Yohei Sugawara, Yuta Tokuoka, Lihang Hong

Organizations: The University of Osaka · Preferred Networks, Inc.

Abstract

State-of-the-art vessel segmentation methods typically require large-scale annotated datasets and suffer from severe performance degradation under domain shifts. In clinical practice, however, acquiring extensive annotations for every new scanner or protocol is unfeasible. To address this, we propose VesselBridge3D, a foundation model adaptation framework that bridges frozen vision foundation models and volumetric vessel segmentation through lightweight 3D adaptation modules. The framework combines a lightweight 3D Adapter, a multi-scale 3D Aggregator, and Z-channel embedding for efficient adaptation to volumetric medical images. We instantiate VesselBridge3D with three frozen foundation encoders (DINOv3, MedSAM, and MedGemma) and evaluate it on the TopCoW (ID) and Lausanne (OOD) datasets. In the extreme low-data regime with 5 training samples, our method achieved a Dice score of 43.42%, marking a 30% relative improvement over the state-of-the-art nnU-Net (33.41%) and outperforming other Transformer-based baselines by up to 45%. The proposed framework was effective across all evaluated frozen foundation encoders, with DINOv3 yielding the best performance in the most label-efficient settings. Furthermore, in the out-of-distribution setting, our model demonstrated superior robustness, achieving a 50% relative improvement over nnU-Net (21.37% vs. 14.22%), which suffered from severe domain overfitting. Ablation studies confirmed the effectiveness of the proposed 3D adaptation modules. Our results demonstrate that VesselBridge3D is an effective framework for label-efficient 3D vessel segmentation under data scarcity and domain shifts.

Figures & tables

Figure 1: Overview of the proposed VesselBridge3D framework. (a) The overall architecture adapts a frozen vision foundation encoder to extract semantic features, while a parallel Lightweight 3D Adapter captures high-frequency volumetric details. (b) The 3D Aggregator fuses multi-scale features via factorized slice and global spatial attention. (c) The decoupled convolution block within the 3D Adapter efficiently models spatial and inter-slice dependencies.

Figure 2: Label efficiency analysis on the TopCoW (ID) dataset. (a) Comparison with conventional segmentation baselines. Our method consistently outperforms nnU-Net, SwinUNETR, UNETR, and 3D U-Net in the low-data regime, achieving a 30.0% relative improvement over nnU-Net with only five annotated volumes. (b) Comparison of VesselBridge3D instantiated with different frozen foundation encoders (DINOv3, MedSAM, and MedGemma). The proposed framework remains effective across all evaluated encoders, while DINOv3 provides the strongest performance in the most label-efficient settings. Shaded regions indicate the mean

\pm

standard deviation of the Dice score across the 13 TopCoW test volumes for each trained model.

Data	nnU-Net			Ours (DINOv3)
Data	Dice(%) $\uparrow$	clDice $\uparrow$	HD95(mm) $\downarrow$	Dice(%) $\uparrow$	clDice $\uparrow$	HD95(mm) $\downarrow$
5 volumes	14.22	0.053	22.61	21.37	0.084	15.94
10%	14.89	0.051	24.37	17.19	0.076	16.45
50%	11.23	0.037	31.25	24.05	0.108	14.21
100%	13.74	0.057	29.03	22.50	0.095	15.77

Table 1: Cross-domain generalization results on the Lausanne dataset (OOD). VesselBridge3D (DINOv3 instantiation) outperforms nnU-Net in all metrics across all data regimes. While nnU-Net degrades in shape fidelity (HD95 increases) as training data increases, our method maintains robust performance, highlighting the generalizability of the frozen backbone.

Figure 3: Qualitative comparison on the unseen OOD Lausanne dataset using models trained on 5 annotated TopCoW volumes. Samples were randomly selected to ensure unbiased evaluation. Top Row: A 2D slice where nnU-Net fails due to domain shift, while VesselBridge3D (DINOv3) correctly delineates the vessel. Bottom Row: 3D rendering of a different subject; the baseline produces fragmented results (blue circle), while VesselBridge3D (DINOv3) preserves vascular connectivity (green circle).

Method	Dice(%) $\uparrow$	clDice $\uparrow$	HD95(mm) $\downarrow$
w/o 3D Aggregator	22.35	0.279	60.97
w/o 3D Adapter	28.67	0.072	85.33
Only Last Layer (11)	42.88	0.284	53.37
Only Last Layer (11) w/o Z	41.10	0.304	52.87
VesselBridge3D (Full)	43.42	0.310	54.29

Table 2: Ablation study of VesselBridge3D (DINOv3) on the TopCoW dataset using 5 annotated training volumes. We evaluate the contribution of key components: 3D Adapter, 3D Aggregator, Multi-scale fusion, and Z-channel. The full model yields the best balance of volumetric overlap (Dice%) and topological preservation (clDice), while maintaining boundary accuracy (HD95) within 1.5 mm of the best variant.

Explore similar work

May 25, 2026

VesselSim: learning 3D blood vessel segmentation without expert annotations

Blood vessel segmentation is a core task in medical image analysis for the care of vascular diseases and surgical planning, yet the challenges of providing expert vascular annotations pose a major obstacle for the progress of related deep learning techniques. To address this, we propose VesselSim, a two-stage framework for universal 3D blood vessel segmentation that eliminates the need for real annotated data during training. First, we introduce a stochastic, geometry-driven vascular simulation framework that models recursive branching, curvature-controlled growth, and collision-aware topology, followed by domain-randomized intensity synthesis to generate 16,500 anatomically plausible 3D angiographic volumes. Second, a 3D U-Net is trained solely on this synthetic data. To bridge the domain gap from synthetic to real images at inference time, we introduce a test-time adaptation strategy via a self-supervised mask reconstruction decoder, enabling adaptation to unseen clinical scans without prior domain knowledge. We evaluate VesselSim in a zero-shot setting on multiple real-world datasets spanning MR and CT across several anatomical regions, including the brain and kidneys. Despite being trained exclusively on synthetic data, VesselSim achieves performance competitive with state-of-the-art vascular segmentation foundation models. These findings suggest that learning vessel geometry from synthetic tubular structures is effective for robust cross-domain generalization, substantially reducing the reliance on acquired medical imaging data and more importantly, expert annotations.

Erin Rainville, Melissa Ananian, Tristan Mirolla +2

Department of Electrical and Computer Engineering, Concordia University, Montreal, Canada · Department of Computer Science and Software Engineering, Concordia University, Montreal, Canada

Synthetic-to-Real Domain Adaptation Synthetic Data Augmentation Image Segmentation Medical Image Segmentation 3D Medical Image Segmentation

Ask PDF

Aug 7, 2026

DINO-3DRA: Leveraging 2D Foundation Model Semantics for 3D Cerebral Aneurysm Segmentation

Accurate aneurysm segmentation in 3D rotational angiography (3DRA) is hindered by extreme class imbalance, morphological similarity to vessels, and absent large-scale 3D pretraining. 2D vision foundation models encode dense structural priors from 1.7 billion images, yet naïve slice-wise transfer fragments anatomical continuity and destabilises optimisation. We propose DINO-3DRA, a dual-path framework achieving effective cross-dimensional semantic transfer by injecting frozen DINOv3 features into a 3D U-Net backbone via Room-Lite spatial mixing and calibrated residual fusion. On multi-centre 3DRA data, DINO-3DRA achieves state-of-the-art aneurysm segmentation (Dice: 0.758; HD95: 2.75 mm; +13% over nnU-Net) with only 5.72M trainable parameters. Ablation studies confirm that gains arise from structured cross-dimensional transfer rather than loss design alone, with bridged foundation features improving anatomical continuity between aneurysms and parent vessels. Without fine-tuning on CADA and SHINY-ICARUS, DINO-3DRA eliminates all catastrophic failure cases observed in baseline architectures, demonstrating robust generalisation across heterogeneous imaging protocols.

Jiayang Lu, Fengming Lin, Alejandro F. Frangi +1

Centre for Computational Imaging and Modelling in Medicine (CIMIM), University of Manchester, Manchester, UK · Christabel Pankhurst Institute, University of Manchester, Manchester, UK · Department of Computer Science, University of Manchester, Manchester, UK +2

Medical Imaging Foundation Models Transfer Learning Medical Image Segmentation 3D Medical Image Segmentation

Ask PDF

Aug 27, 2026

Unsupervised Adaptation of 3D CT Foundation Models for 3D CBCT Segmentation

Accurate 3D segmentation of cone-beam CT (CBCT) is critical for interventional and radiation therapy applications, yet it remains limited by two compounding challenges: the scarcity of annotated CBCT data and the large domain shift from diagnostic CT. Interventional CBCT exhibits fundamental modality differences from conventional CT, driven by acquisition and physics effects as well as contrast-specific vascular content, thereby limiting effective cross-modality model transfer. We propose a novel unsupervised domain adaptation (UDA) framework based on redundancy-reducing feature alignment, enabling 3D CBCT segmentation with no target-domain annotations or inference-time adaptation. Our framework is architecture-agnostic, seamlessly adapting both CNN-based and ViT-based foundation models. We evaluate our method on two challenging CT-CBCT liver segmentation benchmarks: one for interventional vascular procedures and one for radiation therapy, demonstrating that even large-scale pretrained segmentation networks require explicit feature-space bridging to generalize across acquisition modalities, and that our approach consistently outperforms existing pretrained foundation model and UDA strategies. To support reproducibility and benchmarking, we release the liver segmentations for a public CBCT dataset, along with the code, trained models, and weights.

Gauthier Miralles, Loic Le Folgoc, Vincent Jugnon +1

LTCI, Télécom Paris, Institut Polytechnique de Paris, Palaiseau, France · GE HealthCare, Buc, France

Cross-Modality Medical Image Segmentation 3D Medical Image Segmentation Unsupervised Domain Adaptation Vision Foundation Model Adaptation Cone-Beam CT

Ask PDF