The reviewed record of science sign in
Pith

arxiv: 2606.28474 · v1 · pith:PQ7XGNGL · submitted 2026-06-26 · eess.IV · cs.CV

Anatomy-Grounded Synthetic Coronary Angiography for Geometry-Informed Multi-View Matching

In Kyu Lee , Sumin Seo , Jaesik Min This is my paper

Reviewed by Pith2026-06-30 01:13 UTCgrok-4.3pith:PQ7XGNGLopen to challenge →

classification eess.IV cs.CV
keywords synthetic data generationcoronary angiographymulti-view matchingdigital reconstructed radiographC-arm geometry3D reconstructiondeep learningcorrespondence learning
0
0 comments X

The pith

Synthetic DRRs from CCTA volumes generate dense 3D-to-2D labels for training multi-view coronary matching models at zero annotation cost.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a framework that creates synthetic angiographic images from real 3D coronary CT scans by simulating C-arm X-ray projections. This produces paired images with exact correspondence labels without needing human experts to annotate them. The authors then use this data to train a Geometry-Informed Matching Module that combines global image features with anatomical structure information. If successful, this approach removes the main obstacle to developing reliable deep learning tools for 3D reconstruction from multiple angiographic views. A sympathetic reader would care because accurate multi-view matching is essential for guiding heart interventions but has been limited by scarce labeled data.

Core claim

Our framework generates dense, highly accurate 3D-to-2D projection labels by simulating realistic C-arm acquisition geometry on patient anatomy at zero human cost. Leveraging this dense supervision, we propose a Geometry-Informed Matching Module (GIMM) that integrates global feature and anatomical structure into correspondence learning. Unlike real angiography where assessment relies on subjective human annotation, our dataset provides 2D correspondence labels with paired images, allowing human-free evaluation. We comprehensively evaluate our method on the proposed CT-derived DRR dataset and demonstrate improvements over other matching baseline models.

What carries the argument

Physically-grounded data generation framework that synthesizes high-fidelity DRRs from 3D CCTA volumes by simulating realistic C-arm acquisition geometry, producing dense 3D-to-2D projection labels.

If this is right

  • Enables training of robust deep learning models for multi-view correspondence without expensive human labeling.
  • Supports accurate 3D coronary reconstruction from angiographic views.
  • Provides a way to evaluate matching models without subjective human annotation.
  • Demonstrates improvements over baseline matching models on the synthetic dataset.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the synthetic data generalizes well, it could allow training on larger and more diverse patient anatomies than available in real datasets.
  • This method might be adapted to generate training data for other medical imaging tasks involving multi-view geometry.
  • A potential next step is to fine-tune models on a small amount of real data after pretraining on synthetics.

Load-bearing premise

The simulated DRRs and C-arm geometry produce images and labels sufficiently representative of real angiography that models trained on them will generalize to clinical data.

What would settle it

A model trained only on the synthetic dataset shows no improvement over baselines when evaluated on real angiographic image pairs with expert-annotated correspondences.

Figures

Figures reproduced from arXiv: 2606.28474 by In Kyu Lee, Jaesik Min, Sumin Seo.

Figure 1
Figure 1. Figure 1: (A) Overview of the CT DRR pipeline. (B) CT DRR data distribution. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overall architecture of GIMM. (ˆi, ˆj ′ ), and the coarse and fine correspondence sets by Mc and Mf . The coarse stage utilizes a shared ResNet backbone [6] and Transformer [17] to establish globally consistent coarse matches. The fine stage then processes local feature patches around each coarse match and estimates the final sub-pixel correspon￾dence as the expected location over the local matching probab… view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of our proposed GIMM against baseline methods at various [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative results. Predicted matches are shown in green, with magenta [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
read the original abstract

Accurate correspondence matching across multiple angiographic views is the prerequisite for 3D coronary reconstruction and interventional guidance. However, the development of robust deep learning models for this task has been stifled by a fundamental data bottleneck. Obtaining ground truth for matching tasks in angiography pairs is prohibitively expensive and hard to scale. To overcome this barrier, we introduce a physically-grounded data generation framework that synthesizes high-fidelity Digital Reconstructed Radiographs (DRRs) from 3D Coronary CT Angiography (CCTA) volumes. Our framework generates dense, highly accurate 3D-to-2D projection labels by simulating realistic C-arm acquisition geometry on patient anatomy at zero human cost. Leveraging this dense supervision, we propose a Geometry-Informed Matching Module (GIMM) that integrates global feature and anatomical structure into correspondence learning. Unlike real angiography where assessment relies on subjective human annotation, our dataset provides 2D correspondence labels with paired images, allowing human-free evaluation. We comprehensively evaluate our method on the proposed CT-derived DRR dataset and demonstrate improvements over other matching baseline models.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces a physically-grounded framework that synthesizes high-fidelity DRRs from patient CCTA volumes by simulating realistic C-arm acquisition geometry, thereby generating dense, accurate 3D-to-2D projection labels at zero annotation cost. It proposes a Geometry-Informed Matching Module (GIMM) that incorporates global features and anatomical structure for multi-view correspondence learning, and reports improvements over baseline matching models when evaluated on the resulting CT-derived DRR dataset.

Significance. If the synthetic supervision generalizes, the framework could mitigate the labeled-data bottleneck for 3D coronary reconstruction and interventional guidance tasks. The patient-anatomy-driven label generation and human-free evaluation protocol are clear strengths. However, the practical significance remains conditional on untested transfer to clinical angiograms whose appearance differs in contrast dynamics, scatter, and overlapping structures.

major comments (2)
  1. [Abstract and Evaluation section] Abstract and Evaluation section: the central claim that the framework 'overcomes this barrier' for robust clinical models is load-bearing yet unsupported, because all quantitative results (including GIMM gains over baselines) are confined to the synthetic DRR dataset; no transfer experiments or domain-adaptation results on real multi-view angiographic pairs are presented.
  2. [Methods and Results] Methods and Results: the assumption that DRR appearance (uniform attenuation, absent iodine flow, simplified noise) is representative enough for models to generalize is not examined; the paper provides no ablation or comparison that quantifies the domain gap or tests whether GIMM trained on DRRs retains accuracy on clinical data.
minor comments (2)
  1. [Abstract] The abstract states 'improvements over other matching baseline models' without supplying any numerical values, ablation details, or dataset statistics; moving these metrics into the abstract would improve readability.
  2. [Methods] Notation for the GIMM module and the precise form of the geometry-informed loss could be clarified with an equation or pseudocode block to aid reproducibility.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive comments. We address each major point below, acknowledging the evaluation scope and agreeing to revisions where the manuscript can be strengthened without altering its core contributions.

read point-by-point responses

  1. Referee: [Abstract and Evaluation section] Abstract and Evaluation section: the central claim that the framework 'overcomes this barrier' for robust clinical models is load-bearing yet unsupported, because all quantitative results (including GIMM gains over baselines) are confined to the synthetic DRR dataset; no transfer experiments or domain-adaptation results on real multi-view angiographic pairs are presented.

    Authors: We agree the abstract phrasing implies broader clinical readiness than the presented evidence supports. All reported metrics, including GIMM improvements, are confined to the CT-derived DRR dataset. The work's focus is the zero-cost label generation pipeline and the module's performance under that supervision. We will revise the abstract and evaluation section to clarify that the framework addresses the annotation barrier for training data and to note the absence of real angiogram transfer results. revision: partial

  2. Referee: [Methods and Results] Methods and Results: the assumption that DRR appearance (uniform attenuation, absent iodine flow, simplified noise) is representative enough for models to generalize is not examined; the paper provides no ablation or comparison that quantifies the domain gap or tests whether GIMM trained on DRRs retains accuracy on clinical data.

    Authors: The manuscript contains no domain-gap quantification or transfer tests to clinical angiograms. DRR generation omits dynamic contrast flow and certain scatter effects, as noted in the methods. We will add an explicit limitations paragraph in the revised manuscript discussing these simplifications and the need for future domain-adaptation work, while retaining the current focus on anatomically accurate label synthesis from CCTA. revision: yes

standing simulated objections not resolved
  • We cannot supply transfer experiments or domain-gap ablations on real multi-view angiographic pairs, as the study does not include such clinical datasets with ground-truth correspondences.

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The paper presents a data-generation pipeline that synthesizes DRRs and 3D-to-2D labels from CCTA volumes, followed by a new GIMM architecture trained and evaluated on that synthetic dataset. No equations, fitted parameters renamed as predictions, or self-citations appear in the provided text that would make any reported performance gain equivalent to the inputs by construction. The evaluation metrics are computed directly on the generated labels, but this is a standard supervised-learning setup rather than a self-referential reduction; the architecture and synthesis steps remain independent contributions with external content.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the domain assumption that C-arm geometry simulation on CCTA anatomy produces training data whose distribution is close enough to real angiography for the learned matcher to transfer. No explicit free parameters or new physical entities are named in the abstract.

axioms (1)
  • domain assumption Simulating realistic C-arm acquisition geometry on patient CCTA anatomy produces high-fidelity DRRs whose 2D projections yield accurate dense correspondence labels
    Invoked in the description of the data generation framework as the source of 'highly accurate' labels at zero human cost.
invented entities (1)
  • Geometry-Informed Matching Module (GIMM) no independent evidence
    purpose: Integrates global feature and anatomical structure into correspondence learning for multi-view matching
    New module proposed to leverage the dense supervision from the synthetic data.

pith-pipeline@v0.9.1-grok · 5722 in / 1418 out tokens · 37157 ms · 2026-06-30T01:13:43.614813+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

21 extracted references · 1 canonical work pages

  1. [1]

    In: European conference on computer vision

    Chen, H., Luo, Z., Zhou, L., Tian, Y., Zhen, M., Fang, T., Mckinnon, D., Tsin, Y., Quan, L.: Aspanformer: Detector-free image matching with adaptive span trans- former. In: European conference on computer vision. pp. 20–36. Springer (2022)

    2022

  2. [2]

    In: Proceedings of the IEEE conference on com- puter vision and pattern recognition workshops

    DeTone, D., Malisiewicz, T., Rabinovich, A.: Superpoint: Self-supervised interest point detection and description. In: Proceedings of the IEEE conference on com- puter vision and pattern recognition workshops. pp. 224–236 (2018)

    2018

  3. [3]

    In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition

    Edstedt, J., Sun, Q., Bökman, G., Wadenbäck, M., Felsberg, M.: Roma: Robust dense feature matching. In: Proceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition. pp. 19790–19800 (2024)

    2024

  4. [4]

    EuroIntervention8(12), 1451–1460 (2013)

    Girasis, C., Schuurbiers, J., Muramatsu, T., Aben, J.P., Onuma, Y., Soekhradj, S., Morel, M.A., Van Geuns, R.J., Wentzel, J.J., Serruys, P.W.: Advanced three- dimensional quantitative coronary angiographic assessment of bifurcation lesions: methodology and phantom validation. EuroIntervention8(12), 1451–1460 (2013)

    2013

  5. [5]

    Hartley,R.I.:Indefenseoftheeight-pointalgorithm.IEEETransactionsonpattern analysis and machine intelligence19(6), 580–593 (1997)

    1997

  6. [6]

    He,K.,Zhang,X.,Ren,S.,Sun,J.:Deepresiduallearningforimagerecognition.In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 770–778 (2016)

    2016

  7. [7]

    SIAM (2001)

    Kak, A.C., Slaney, M.: Principles of computerized tomographic imaging. SIAM (2001)

    2001

  8. [8]

    Scientific Reports14(1), 18630 (2024)

    Kim, H.W., Noh, S.C., Kim, S.H., Chu, H.W., Jung, C.H., Kang, S.H.: Effective descriptor extraction strategies for correspondence matching in coronary angiog- raphy images. Scientific Reports14(1), 18630 (2024)

    2024

  9. [9]

    In: European conference on computer vision

    Leroy, V., Cabon, Y., Revaud, J.: Grounding image matching in 3d with mast3r. In: European conference on computer vision. pp. 71–91. Springer (2024) 10 I.K. Lee et al

    2024

  10. [10]

    In: Proceedings of the seventh IEEE international conference on computer vision

    Lowe, D.G.: Object recognition from local scale-invariant features. In: Proceedings of the seventh IEEE international conference on computer vision. vol. 2, pp. 1150–

  11. [11]

    Journal of Cardiology69(1), 280–286 (2017)

    Nishi, T., Kitahara, H., Fujimoto, Y., Nakayama, T., Sugimoto, K., Takahara, M., Kobayashi, Y.: Comparison of 3-dimensional and 2-dimensional quantitative coronary angiography and intravascular ultrasound for functional assessment of coronary lesions. Journal of Cardiology69(1), 280–286 (2017)

    2017

  12. [12]

    In: Proceedings of the AAAI conference on artificial intelligence

    Perez, E., Strub, F., De Vries, H., Dumoulin, V., Courville, A.: Film: Visual rea- soning with a general conditioning layer. In: Proceedings of the AAAI conference on artificial intelligence. vol. 32 (2018)

    2018

  13. [13]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Sarlin, P.E., DeTone, D., Malisiewicz, T., Rabinovich, A.: Superglue: Learning feature matching with graph neural networks. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 4938–4947 (2020)

    2020

  14. [14]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Schonberger, J.L., Frahm, J.M.: Structure-from-motion revisited. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 4104–4113 (2016)

    2016

  15. [15]

    Medical physics12(2), 252–255 (1985)

    Siddon, R.L.: Fast calculation of the exact radiological path for a three-dimensional ct array. Medical physics12(2), 252–255 (1985)

    1985

  16. [16]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Sun, J., Shen, Z., Wang, Y., Bao, H., Zhou, X.: Loftr: Detector-free local fea- ture matching with transformers. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 8922–8931 (2021)

    2021

  17. [17]

    In: International Conference on Learning Representations

    Tang, S., Zhang, J., Zhu, S., Tan, P.: Quadtree Attention for Vision Transformers. In: International Conference on Learning Representations

  18. [18]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Weinzaepfel, P., Csurka, G., Cabon, Y., Humenberger, M.: Visual localization by learning objects-of-interest dense match regression. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 5634–5643 (2019)

    2019

  19. [19]

    arXiv preprint arXiv:2312.11593 (2023)

    Wu, Y., Jena, R., Gulsun, M., Singh, V., Sharma, P., Gee, J.C.: Towards Establish- ing Dense Correspondence on Multiview Coronary Angiography: From Point-to- Point to Curve-to-Curve Query Matching. arXiv preprint arXiv:2312.11593 (2023)

    work page arXiv 2023

  20. [20]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Yi, K.M., Trulls, E., Ono, Y., Lepetit, V., Salzmann, M., Fua, P.: Learning to find good correspondences. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 2666–2674 (2018)

    2018

  21. [21]

    Computerized Medical Imaging and Graphics109, 102287 (2023)

    Zeng,A.,Wu,C.,Lin,G.,Xie,W.,Hong,J.,Huang,M.,Zhuang,J.,Bi,S.,Pan,D., Ullah,N.,etal.:Imagecas:Alarge-scaledatasetandbenchmarkforcoronaryartery segmentation based on computed tomography angiography images. Computerized Medical Imaging and Graphics109, 102287 (2023)

    2023