ConPro: Contrast Projection Pretraining for Label-Efficient Vessel Segmentation in DSA Sequences

Xinge Guo; Yuanhao Wang; Liqi Shu; Yang Liu; Min Xu

cs.CVSep 24, 2026

ConPro: Contrast Projection Pretraining for Label-Efficient Vessel Segmentation in DSA Sequences

Authors: Xinge Guo, Yuanhao Wang, Liqi Shu, Yang Liu, Min Xu

Organizations: Duke University, Durham, NC, USA · Carnegie Mellon University, Pittsburgh, PA, USA · University of Pittsburgh Medical Center, Pittsburgh, PA, USA

Abstract

Dense vessel annotation in digital subtraction angiography (DSA) is labor-intensive, yet every unlabeled sequence records how contrast passes through the vessels. Semi-supervised methods take their targets from the current model, and generic self-supervised pretexts reconstruct static appearance, so this signal goes unused. We propose ConPro, a self-supervised pretraining scheme whose target is a contrast projection, the normalized drop of every pixel below its temporal median over the sequence. On DIAS and DSCA, with 10%, 20% and 50% of the training cases labeled, ConPro improves on training from scratch at every label fraction and is the best of the compared methods on DSCA at 20% and 50% labels. Controlled comparisons show that the gain comes from the target. A temporal-median target with the same input, loss and budget stays at scratch level, and using the projection directly instead of learning it, as an input channel or a pseudo-label, helps little or hurts. ConPro provides pretrained weights without changing the segmentation architecture, so it combines with semi-supervised training, and UniMatch, the strongest baseline, gains 0.5 to 2.0 Dice and 0.9 to 2.3 clDice at every label fraction when started from ConPro weights, reaching 75.4 Dice on DIAS and 81.3 on DSCA.

Figures & tables

Figure 1: Annotation-free pretraining. The network sees a subset of the frames (cyan) and predicts the contrast projection $C^{\star}$ of the whole sequence. Grey frames are withheld.

Figure 2: ConPro pipeline. Stage I predicts $C^{\star}$ from a subset of frames $X_{\Omega}$ , and Stage II transfers the backbone $F$ to vessel segmentation.

		DIAS			DSCA
Method	Type	10%	20%	50%	10%	20%	50%
Supervised	Sup.	65.1 / 58.7	71.8 / 65.2	73.5 / 67.2	79.4 / 75.2	80.2 / 76.1	80.2 / 76.1
CPS [ 6 ]	Semi	69.1 / 61.6	71.7 / 64.5	72.3 / 64.9	79.8 / 75.5	80.5 / 76.4	80.6 / 76.5
CorrMatch [ 7 ]	Semi	64.1 / 55.5	69.2 / 61.2	71.2 / 63.2	76.1 / 70.3	77.5 / 72.2	78.2 / 73.5
UniMatch [ 8 ]	Semi	71.2 / 65.0	73.1 / 66.6	74.9 / 68.8	80.0 / 76.0	79.6 / 76.1	79.3 / 75.5
RPST ∗ [ 2 ]	Semi	70.6 / 63.7	72.3 / 65.5	72.1 / 65.2	76.6 / 70.7	76.7 / 71.2	77.0 / 71.8

Table 1: Dice / clDice (%; three-seed means), matched backbone and Stage II update budget. Sup.: labeled sequences only; Semi: labeled and unlabeled sequences; Self: self-supervised pretraining, then supervised fine-tuning. The three

C^{\star}

rows use the contrast projection directly, without pretraining. The last three rows are ours, ConPro fine-tuned with labels only, and RPST ∗ and UniMatch started from ConPro weights. Bold : best; underline : second best.

Figure 3: Qualitative results at $20\%$ labels, one test sequence per dataset. Green marks correct vessel pixels, red marks errors.

					DIAS			DSCA
Protocol	Pretraining	Input	Target	Loss	10%	20%	50%	10%	20%	50%
Target	None				65.1 / 58.7	71.8 / 65.2	73.5 / 67.2	79.4 / 75.2	80.2 / 76.1	80.2 / 76.1
	Median, all frames	$X_{\Omega}$	$\operatorname{med}(X)$	BCE+Dice	64.1 / 57.6	70.7 / 63.7	71.7 / 64.3	78.8 / 74.3	79.9 / 75.8	80.1 / 76.2
	Median, all frames	$X_{\Omega}$	$\operatorname{med}(X)$	$L_{1}$	61.9 / 58.9	68.7 / 61.5	71.2 / 64.4	78.8 / 74.3	79.6 / 75.6	79.9 / 75.9
	Median, withheld frames	$X_{\Omega}$	$\operatorname{med}(X_{\bar{\Omega}})$	$L_{1}$	61.8 / 57.1	69.3 / 62.9	70.6 / 63.1	78.7 / 74.3	79.7 / 75.8	79.7 / 75.7
	Contrast projection	$X_{\Omega}$	$C^{\star}$	BCE+Dice	69.3 / 63.5	71.9 / 66.0	74.4 / 68.4	80.0 / 75.8	81.2 / 77.4	81.4 / 77.6

Table 2: Stage I ablations (Dice / clDice %; three-seed means). Target: input fixed to a random half

X_{\Omega}

, target varied. Input: contrast projection kept, frame feeding varied. Protocols trained independently; bold : highest within a protocol.

Figure 4: Input variants of Table 2 on the sequences of Fig. 3 , alone and with UniMatch. Colors as in Fig. 3 .

Explore similar work

ESRVS: Extreme Semi-Supervised Retinal Vessel Segmentation with a Single Annotated Image
Jul 27, 2026Mingzhi Xu, Yizhe ZhangVessel Segmentation Semi-Supervised Medical Image Segmentation
Nanjing University of Science and Technology, Nanjing, China
MSA-UNet3+: Multi-Scale Attention UNet3+ with New Supervised Prototypical Contrastive Loss for Coronary DSA Image Segmentation
Apr 7, 2025Rayan Merghani Ahmed, Adnan Iltaf, Mohamed Elmanna +5Coronary Angiography Semi-Supervised Medical Image Segmentation
Shenzhen Institutes of Advanced Technology (SIAT), Chinese Academy of Sciences (CAS), Shenzhen, 518055, China · University of Chinese Academy of Sciences (UCAS), Beijing, 101408, China · Department of Biomedical Engineering and Systems, Faculty of Engineering, Cairo University, Cairo, 12613, Egypt +2
UI-VISA: U-Net Initialized Vascular Image Segmentation Architecture
Sep 1, 2026Asees Kaur, Suzanne S. Sindi, Erica M. RutterVessel Segmentation Coronary Angiography
University of California, Merced

Jul 27, 2026, cs.CVJ/K move · Enter open · S save

ESRVS: Extreme Semi-Supervised Retinal Vessel Segmentation with a Single Annotated Image

Mingzhi Xu, Yizhe Zhang

Nanjing University of Science and Technology, Nanjing, China

Vessel Segmentation Semi-Supervised Medical Image Segmentation Semi-Supervised Learning Dinov3

Open paper AskPDF arXiv

Learning from minimal human supervision is a long-standing goal in medical image analysis, where dense expert annotations are costly. We study retinal vessel segmentation in an extreme semi-supervised setting with one annotated image and a pool of unlabeled images. We propose ESRVS, which selects a representative reference image for manual annotation and transfers vessel cues using target-domain-adapted DINOv3 features. ESRVS constructs a multi granular vessel prototype, combines prototype-similarity maps with a physics-inspired prior to generate initial pseudo-labels, and refines the transferred supervision through weighted pseudo-label training and adversarial refinement. Across eight public datasets, ESRVS achieves the best Dice and clDice on six datasets, and the best HD95 on all eight datasets among the compared semi-supervised methods, although those methods use 10 to 20% labeled data. With Mask2Former, ESRVS retains on average 93.7% of fully supervised Dice and 95.1% of fully supervised clDice. These results demonstrate the potential of foundation-model label propagation for highly label-efficient retinal vessel segmentation. Code is available at https://github.com/IAANNH/ESRVS.

Abstract

Explore similar work

ESRVS: Extreme Semi-Supervised Retinal Vessel Segmentation with a Single Annotated Image

MSA-UNet3+: Multi-Scale Attention UNet3+ with New Supervised Prototypical Contrastive Loss for Coronary DSA Image Segmentation

UI-VISA: U-Net Initialized Vascular Image Segmentation Architecture