List of Papers Browse by Subject Areas Author List
Abstract
Detailed automated artery and vein segmentation on brain CT angiography (CTA) enables the use of standardized quantitative collateral scores for prognosticating outcomes to support treatment decisions in acute ischemic stroke. However, current deep learning methods for vessel segmentation are inaccurate for small vessels, fail to separate arteries and veins, do not account for variations in contrast bolus timing, and require laborious manual annotations. To address these issues, we developed a deep learning method to separate intracranial artery and vein segmentation from semi-automatically annotated thin-slice CT perfusion (CTP) source data. These CTPs were used to simulate CTAs with variable contrast timing, thereby enabling contrast-phase augmentation during training. We compared models trained with and without contrast-phase augmentation and evaluated replicability across models trained from scratch and fine-tuned foundation models. Segmentation volumes in either hemisphere were used to automatically measure arterial and venous quantitative collateral scores (aQCS/vQCS) that were compared to expert-rated Tan scores (0-3) and cortical vein opacification scores (COVES; 0-6). Across external test sets of simulated and real CTAs, contrast-phase augmentation improved artery and vein segmentation performance for models trained from scratch, but yielded only a limited or no gain for foundation models. Models trained from scratch outperformed fine-tuned foundation models. aQCS and vQCS showed high agreement with expert rater scores. Our results demonstrate the quantitative benefits and clinical utility of contrast-phase augmentation for deep-learning-based artery and vein segmentation.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/4264_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{vanHen_Artery_MICCAI2026,
author = { van Voorst, Henk AND de Wilde, Daniel AND Canals, Pere AND Schwartz, Maya AND Kesten, Jamie AND Konduri, Praneeta R. AND Zamarud, Aroosa AND Jiang, Bin AND Christensen, Soren AND Kim, Donghoon AND Liu, Yongkai AND Albers, Greg AND Sharma, Gagan AND Parsons, Mark AND Zaharchuk, Greg AND Lansberg, Maarten AND Heit, Jeremy},
title = { { Artery and vein segmentation on brain CTA using CTP for contrast-phase augmentation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16884},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The authors developed a deep learning method to segment intracranial arteries and veins from semi-automatically annotated thin-slice CT perfusion (CTP) data. The CTPs were used to simulate CT angiography (CTA) images with varying contrast timing, enabling contrast-phase augmentation during training. They compared state-of-the-art existing models and trained them with and without this augmentation and assessed replicability across models trained from scratch and fine-tuned foundation models. Arterial and venous quantitative collateral scores (aQCS/vQCS) were automatically derived and compared with expert-rated Tan and COVES scores. Contrast-phase augmentation improved segmentation performance for models trained from scratch, with limited benefit for foundation models. They showed that models trained from scratch outperformed fine-tuned ones. These results highlight the clinical value of contrast-phase augmentation for artery and vein segmentation which is a gap in this field.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The problem and the gap are well defined and authors designed their method to address that. They use info from a modality in segmentation of regions in another modality.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
For the contrast-phase augmentation part they do the analysis in some regions of the brain but why those?
Among authors contributions is simulating CTA from CTP by simply summing three exposure-weighted consecutive CTP time frames. This is not assessed if CTAs can be simulated from CTP like this and how good are the results. This part is lacking from the paper.
Also their implementation is on a small set of dataset and this can affect their results.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
As said the problem is well defined and authors designed their method to address that. They use info from one modality (CTP) in segmentation of regions in another modality (CTA) which is novel.
They have small sample of dataset that makes the evaluations week. Also the simulation part of CTA from CTP is not evaluated.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
The main contribution is the proposal of a contrast-phase augmentation method based on CTP source data to simulate CTA at different contrast agent phases, used to train an artery-vein segmentation model, and demonstrating that this augmentation improves segmentation performance. Furthermore, the authors used the segmentation results to automatically calculate arterial and venous collateral circulation scores, showing certain clinical application value.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1.The paper proposes a clear and sensible framework: using CTP source data to simulate CTA scans , and then using these simulated scans for contrast-phase augmentation. The overall pipeline is reasonable. 2.The authors compare training with and without contrast-phase augmentation, which shows the effective efficiency of the proposed method. 3.Separate artery and vein segmentation on brain CTA is clinically meaningful, and the paper also shows that the derived automated collateral scores correlate well with expert-based Tan and COVES ratings.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
1.If it is intended as a methodology paper, the technical novelty appears limited, since the main contribution is largely a task-specific augmentation and data construction strategy built on standard segmentation backbones. If it is positioned as an application-oriented paper, the experimental evidence on real CTA remains relatively limited to fully support broad clinical adoption. 2.Several parts of the pipeline, especially data generation, annotation refinement, and training details, would benefit from more precise description. 3.The visual differences between models trained with and without contrast-phase augmentation are not sufficiently clear from the current examples. 4.Reporting results only as integers is too coarse for a quantitative comparison. In addition, including the Canals et al. baseline in the current form is of limited value. 5.The overall design of Figure 1 makes the workflow harder to follow than necessary, especially for readers trying to understand the full pipeline quickly. The distribution of citations in this part is also somewhat awkward and interrupts the presentation. 6.The efficient datasets for evaluation are generally very limited.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not provide sufficient information for reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
1.If it is intended as a methodology paper, the technical novelty appears limited, since the main contribution is largely a task-specific augmentation and data construction strategy built on standard segmentation backbones. If it is positioned as an application-oriented paper, the experimental evidence on real CTA remains relatively limited to fully support broad clinical adoption. 2.Several parts of the pipeline, especially data generation, annotation refinement, and training details, would benefit from more precise description. 3.The visual differences between models trained with and without contrast-phase augmentation are not sufficiently clear from the current examples. 4.Reporting results only as integers is too coarse for a quantitative comparison. In addition, including the Canals et al. baseline in the current form is of limited value. 5.The overall design of Figure 1 makes the workflow harder to follow than necessary, especially for readers trying to understand the full pipeline quickly. The distribution of citations in this part is also somewhat awkward and interrupts the presentation. 6.The efficient datasets for evaluation is general very limited.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(3) Weak Reject — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Although the manuscript shows some promise, its limited contribution and insufficient experimental evaluation prevent it from being suitable for publication in its current form.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
I am willing to support acceptance, as the work addresses a clinically meaningful problem. However, I remain concerned that the evaluation is still limited and that the methodological/evaluative novelty is not strong, especially given the modest real-CTA test set.
Review #3
- Please describe the contribution of the paper
Artery and vein segmentation is important but remains challenging due to small vessel size, difficulty in separating arteries and veins, and the labor-intensive nature of manual annotation. This study introduces a method for segmenting arteries and veins from CTP data. The overall pipeline, from segmentation to clinical score, is robust.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1.Contrast phase augmentation is well evaluated, with comparisons on both simulated and real CTA for artery and vein 2.The ablation study demonstrates the effectiveness of using contrast-phase augmentation protocols during training on segmentation. It is well structured and supported by clear figures 3.The study goes beyond segmentation and extends to clinical score evaluation with solid validation
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
1.Figure 1 should be elaborated at a high level to better illustrate the overall pipeline (a brief explanation in the caption would improve clarity) 2.Patient information is missing, including demographics, selection criteria, and bias mitigation 3.Missing annotation details: number of annotators, level of expertise, and years of experience
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The topic is relevant and potentially useful to the artery and vein segmentation community.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Author Feedback
We thank the reviewers for their feedback. The main concerns relate to the size of the training and test sets, the novelty of our contribution, and a lack of methodological details. Dataset size: Although our cohort is modest in size (44 training CTPs; 17 CTPs and 20 CTAs for testing), it is the largest and most detailed dataset for this task to date. Prior studies included fewer subjects, covered only limited vascular territories, and did not distinguish between arteries and veins (PMID:38235066, PMID:41494368, PMID:31944937, PMID:35810508). Small datasets in this task reflect that detailed artery-vein annotation requires more than a day per scan. Despite the modest test set size, Table 1 shows considerable differences between models (rows) and with/without contrast-phase augmentation (columns), with small standard errors. In addition, we validated the agreement of quantitative collateral scores with conventional scores in a reader study of 309 CTAs. Contribution: We acknowledge that our approach is limited to single-phase CTA, but emphasize its importance for detailed cerebrovascular analysis. As reflected by the reviewers’ differing interpretations, our main contribution combines both methodological and application aspects. 1) Application: We present the first accurate deep learning artery-vein segmentation method for single-phase brain CTA, presenting the largest detailed dataset to date. 2) Method: We are the first to show that CTP-based training and contrast augmentation enable and improve single-phase CTA based artery-vein segmentation. 3) Method: We used semi-automated label generation for CTP, speeding up the laborious annotation process. 4) Application: This is the first method of quantitative collateral scoring for veins for which there is a growing interest. Annotation details: CTP/simulated CTA annotation is described in detail in Section 2.2, and real CTA annotation is briefly described in Section 3.1.We will add the following: Real CTA annotations were performed by one trained rater (2 years of experience), refined by a second experienced rater (6 years of experience), and finalized by a board-certified interventional neuroradiologist (15+ years of experience). Population details: We apologize for not adding demographic data. Inclusion criteria are in section 3.1.Upon acceptance, we will provide references to the source and annotated datasets, which were omitted here to preserve anonymity. Training details are described in Section 3.2, as well as open-sourcing our code. Contrast phase augmentation is described in section 2.3.It is unclear to us which specific details are missing, but we would be happy to provide further clarification. Figures 1 and 2: The references in Figure 1 refer to text descriptions in sections 2.1, 2.2, and 2.3.We would like to point out the subscript that clarifies each color and the narrative description in Section 3.3.We will add a more extensive figure subscripts. Reviewer 1 questioned the motivation for using a specific region of interest: Reader-based arterial (Tan) and venous (COVES) scores assess specific vascular regions affected by ischemic stroke, while our segmentations cover the entire brain. For fair comparison, we restricted artery-vein segmentations for calculating quantitative collateral scores to the corresponding Tan and COVES regions using atlas-based masks. All other uses – training and results (Table 1) – consider the entire brain. We will add this motivation to our manuscript. Reviewer 2 questioned the necessity of comparing with the method by Canals et al.: Our method’s performance in Table 1 may appear lower than that reported in other studies (PMID:38235066, PMID:41494368, PMID:31944937, PMID:35810508). By comparing against Canals et al., the only published method available to us, we show that previous studies may reflect data-specific performance, with a greater emphasis on larger vessels. We will add a description of this motivation.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
The paper proposes a deep learning framework to segment intracranial arteries and veins by simulating CT angiographies from multi-phase CT perfusion data. The resulting method is validated with vascular segmentation and automated collateral score measures.
Overall, the paper has a good clinical value and the proposed CTA simulation is sound with the effectiveness of the strategy well demonstrated. However, the reviewers also raise concerns in a few major aspects: 1.The novelty of the contribution is limited as the primary focus is data generation and augmentation 2.The data size is small to effectively evaluate the performance of the proposed segmentation method 3.Several key points of the methodologies (e.g., model training, data annotation) should be further detailed
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
Although the novelty of the manuscript primarily focuses on data augmentation, the clinical relevance is high and the proposed training technique is effective. Vascular segmentation is a challenging and important task. After rebuttal, all reviewers have given their positive scores. Thus, an acceptance is recommended. The authors should revise the paper accordingly as mentioned in the rebuttal when preparing the camera-ready version.
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
All reviewers agree on acceptance of the manuscript after rebuttal, despite the evaluation remaining still limited given the modest real-CTA test set.
Meta-review #3
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
agree with the reviewers