Pith. sign in

REVIEW 3 major objections 6 minor 22 references

A segmentation can score high Dice yet still sever a bifurcation and reverse an FFR-CT treat/defer decision; BCS measures that connectedness gap.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-31 11:12 UTC pith:T42E6QE4

load-bearing objection Useful junction-level metric and a clean BCR–β0 dissociation; the severe-disease FFR OR is thinner than the abstract implies because of the gap-bridging 0D solver and an unreported stratum size. the 3 major comments →

arxiv 2607.28327 v1 pith:T42E6QE4 submitted 2026-07-30 cs.CV

Same Branches, Different Trees: A Bifurcation Connectedness Metric for Coronary Artery Segmentation and FFR-CT Decision Agreement

Maame Owusu-Ansah , Kelvin Lee , Dr Vinod Venugopal , Muhammad Moazzam Jawaid , Wenting Duan , James Brown This is my paper
classification cs.CV
keywords coronary artery segmentationbifurcation connectednessFFR-CTtopology-aware losssoft-BCSbranch correspondencevessel tree topology
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

FFR-CT decides whether a coronary narrowing needs treatment by simulating blood flow through a patient-specific vessel tree. That simulation fails if the segmentation severs a bifurcation, even when volumetric overlap looks excellent. This paper introduces the Bifurcation Connectedness Score (BCS), which checks, at each ground-truth junction, whether the outgoing branches remain mutually reachable in the prediction. BCS moves sharply when a junction is broken and stays nearly flat under pure narrowing that leaves the tree intact—behaviour standard overlap metrics do not share. Higher BCS tracks closer agreement between the treat/defer decisions a flow solver makes on predicted versus ground-truth geometry, especially in severe disease. Training experiments further show that two recall-based topology losses recover essentially the same branches yet assemble them into trees with different numbers of disconnected pieces. The authors therefore argue that branch recovery and junction connectedness are separable qualities and should both be reported.

Core claim

BCS isolates a property of coronary segmentations that Dice, clDice, HD95 and component count largely miss: whether the vessel tree stays connected at its bifurcations. Higher BCS accompanies higher agreement between FFR-CT decisions computed on predicted versus reference geometry (most clearly when ground-truth FFR is below 0.75), and this link is geometric rather than clinical because both decisions come from the same solver. In addition, soft-BCS and Skeleton Recall recover equivalent branch sets but produce systematically different tree fragmentation.

What carries the argument

Bifurcation Connectedness Score (BCS): the fraction of ground-truth bifurcations whose short outgoing stubs remain mutually reachable through a lightly dilated predicted skeleton; soft-BCS is its differentiable surrogate that weights the weakest stub at each junction.

Load-bearing premise

The claimed link between BCS and treatment decisions rests on a simplified zero-dimensional resistor-network flow solver whose outlet resistances are calibrated to a healthy ground-truth FFR of 0.85 and that bridges gaps smaller than 1.5 mm before solving.

What would settle it

If, on the same ImageCAS cases, BCS quartiles no longer stratify treat/defer agreement once decisions are taken from a full clinical FFR-CT pipeline or from invasive FFR rather than the paper’s calibrated 0D solver, the geometric-fidelity claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Coronary segmentation benchmarks should report a junction-level connectedness score alongside Dice and branch-recovery metrics.
  • Topology losses that only maximise skeleton recall can leave more disconnected components than losses that explicitly reward mutual reachability at junctions.
  • In severe-disease cohorts, segmentations with higher BCS are more likely to yield the same binary FFR-CT decision as the reference geometry under a fixed solver.
  • BCS can serve as a pre-computation flag for cases whose connectivity failures are likely to alter simulated flow paths.

Where Pith is reading between the lines

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

  • The same junction-level check could be applied to other thin tubular trees (cerebral arteries, airways, retinal vessels) where a single break removes an entire subtree from a downstream model.
  • If BCS-flagged breaks turn out to co-locate with known FFR-CT versus invasive-FFR discordances, the metric would give a cheap geometric filter before expensive physiologic validation.
  • Training objectives that jointly penalise both missing branches and excess components may be needed once recall and organisation are treated as separate axes.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes the Bifurcation Connectedness Score (BCS), a junction-level metric of whether ground-truth coronary bifurcations remain mutually reachable in a predicted segmentation, plus soft-BCS, a differentiable training surrogate. On ImageCAS (N=250 test) across three pretrained 3D backbones and three seeds, the authors argue that (i) BCS is largely independent of Dice/clDice/HD95/β0 (rank R²=0.21) and responds strongly to severing but not to connectedness-preserving narrowing; (ii) higher BCS accompanies closer same-solver FFR-CT treat/defer agreement on predicted vs ground-truth geometry, most clearly when FFR_GT<0.75 (OR 2.16, CI [1.23, 4.18]); and (iii) soft-BCS and Skeleton Recall recover equivalent branches (ΔBCR within noise) yet build differently connected trees (Δβ0=+2.94). They recommend reporting both branch recovery and connectedness.

Significance. If the claims hold, the work cleanly separates two properties that topology-aware vessel segmentation has often conflated—branch recovery vs junction-level assembly—and supplies a simple, interpretable metric tied to a downstream geometric decision task. Strengths include a controlled multi-architecture design (shared Cv init, fixed α=0.05, three seeds), perturbation specificity tests, patient-clustered bootstrap/GEE with Holm correction, an intentional ASOCA calibre-only OOD check, and explicit scoping that FFR agreement is geometric fidelity under one solver, not clinical accuracy. That package is useful for coronary segmentation benchmarking even if the clinical FFR-CT transfer remains open.

major comments (3)
  1. [Abstract; §2.2; Table 1] Abstract and §2.2 lead with the severe-disease OR 2.16, CI [1.23, 4.18], but the manuscript never reports the patient N (or number of architecture-seed triples) in the FFR_GT<0.75 stratum. Full-cohort BCS quartiles are non-monotonic (Table 1: Q2 74.7% < Q1 77.2%). Without stratum size, event counts, and a clear statement of how the OR was estimated (quartile contrast vs continuous; clustering unit), the headline clinical-facing number cannot be audited. Please report N, events, and the exact contrast, and temper the abstract if the stratum is small.
  2. [§1.5; §2.3; Fig. 2b] The FFR readout (§1.5, Fig. 2b) is a calibrated 0D Poiseuille–Kirchhoff solve that bridges gaps below 1.5 mm before flow. The paper’s own Cskel vs Ctopo comparison (§2.3) finds large Δβ0 (+2.94) with near-equal FFR agreement precisely because “fragmentation does not fully propagate.” That design choice compresses the junction-break signal BCS is meant to catch and makes the decision map partially topology-robust by construction. The geometric-fidelity claim is still defensible, but the manuscript should quantify how often BCS-detected breaks are gap-bridged away, and state more prominently that the BCS–decision link is measured under a gap-tolerant reduced-order solver, not a full 3D CFD FFR-CT pipeline.
  3. [Table 1; §2.2; Abstract] In the multivariable GEE (Table 1, right), BCS and Dice share the same point estimate (OR 0.81 per SD) with only BCS reaching p=0.002; clDice flips above 1 under collinearity. The text correctly declines to rank coefficients, yet the abstract still presents BCS as the metric that “accompanies closer agreement.” Please either show a model-comparison or partial-R² style decomposition that isolates BCS’s incremental contribution beyond Dice/BER, or rephrase the abstract/§2.2 claim to match the adjusted-association evidence actually reported.
minor comments (6)
  1. [Abstract; Fig. 1] Fig. 1 caption and body use Cv / Ctopo consistently, but the abstract never names the baseline or the competing losses; a half-sentence would help readers place soft-BCS.
  2. [§1.2] BCS hyperparameters (l=8, δ=3) are said to leave loss ordering unchanged across 10 combinations (§1.2); a small supplementary table of those combinations would make the robustness claim checkable.
  3. [§1.4; Eq. (3)] Eq. (3): clarify whether stub mean probability p̄_k is computed on the predicted soft mask or on a soft-skeletonised map; the surrounding text says “predicted foreground probability along every incident stub.”
  4. [Table 2] Table 2: FFR↑ column header is ambiguous (consistency % vs mean FFR); spell out “FFR decision consistency (%)” in the caption.
  5. [§2.2; §1.5] ASOCA is used only for calibre-blindness of BCS (§2.2); state explicitly that FFR was not run there because N=20 is underpowered, so readers do not expect a missing analysis.
  6. [Abstract; §1.1] Minor typography: “OR2 .16”, “CI[1 .23, 4.18]”, “1,000coronary”, “λ dLDice” spacing glitches appear in the compiled text; clean before camera-ready.

Circularity Check

1 steps flagged

No construction-level circularity: BCS, soft-BCS, and the FFR readout are independently defined; only a mild non-load-bearing self-citation for the metric’s prior introduction.

specific steps
  1. self citation load bearing [Abstract; Sec. 1 intro paragraph on BCS; Ref. [12]]
    "BCS and its surrogate soft-BCS were introduced in preliminary form [12], which established the metric and loss on a single backbone. Here we characterise BCS across three pretrained architectures, relate it to FFR-CT decision agreement, and show that branch recovery and graph fragmentation dissociate."

    The metric and loss originate in the authors’ own prior work [12] (Owusu-Ansah, Brown, Duan, Jawaid overlap). This is ordinary self-citation for priority, not a uniqueness theorem or a fit renamed as prediction: definitions are restated in full (Sec. 1.2, Eq. 3), and the FFR/BCR/β0 results are new measurements. Not load-bearing for the claimed associations, hence only a minor mark.

full rationale

The paper’s load-bearing claims are empirical, not definitional. BCS is defined from ground-truth junctions and mutual stub reachability in a dilated predicted skeleton (Sec. 1.2, Fig. 2a), without reference to FFR, Dice, or the training loss. Soft-BCS is explicitly a differentiable surrogate that weights the weakest stub (Eq. 3), not equated with BCS; Table 2 even shows Skeleton Recall can outscore soft-BCS-trained models on BCS. The FFR-CT readout is a separate reduced-order Poiseuille–Kirchhoff solve on centreline graphs (Sec. 1.5, Fig. 2b), with outlet resistances calibrated so the GT tree has healthy median terminal FFR 0.85; both predicted and GT geometries then pass through the same solver, so decision agreement is a geometric comparison, not a quantity fitted from BCS. Calibrating the healthy baseline does not force the BCS–agreement association by construction—the association is measured post hoc (quartiles, GEE). Independence of BCS from standard metrics is likewise measured (R²=0.21), not assumed. The only mild circularity-adjacent element is citation [12] (overlapping authors) for the prior introduction of BCS/soft-BCS; the present manuscript re-states the definitions and supplies new multi-backbone, FFR, and BCR/β0 results, so the self-citation is historical credit rather than a load-bearing uniqueness or uniqueness-from-authors step. Score 1 reflects that minor self-citation only.

Axiom & Free-Parameter Ledger

6 free parameters · 6 axioms · 3 invented entities

The paper is empirical CV/medical imaging: load-bearing structure is metric definitions, a simplified flow model, dataset labels as GT anatomy, and training hyperparameters—not a first-principles derivation. Free parameters are the BCS geometry knobs, loss weights, and FFR calibration targets. Axioms are standard segmentation/topology and hemodynamic modeling choices. Invented entities are the proposed scores/losses (BCS, soft-BCS, BCR usage here), which are definitions with empirical handles rather than physical postulates.

free parameters (6)
  • BCS stub length l and dilation δ = l=8, δ=3
    Junction preservation test uses l=8 voxels and δ=3-voxel dilation of the predicted skeleton; ordering claimed robust across 10 combinations but values are chosen, not derived.
  • soft-BCS temperature T and topology weight α = T=0.2, α=0.05
    Stub weighting uses T=0.2; all topology losses share α=0.05 without per-loss tuning—hand-set training knobs that affect Ctopo vs baselines.
  • BCR match distance d_match and min branch length = d_match=3 mm; drop <4 voxels
    Branches match if symmetric chamfer ≤3 mm; branches <4 voxels dropped as noise—thresholds that define the 'same branches' claim.
  • FFR healthy terminal calibration and treat threshold = terminal FFR_GT=0.85; treat ≤0.80
    Outlet resistances calibrated so GT median terminal FFR=0.85; positive decision if min FFR≤0.80 (clinical convention) under Poiseuille resistances R∝L/r^4.
  • Solver gap-bridging length = 1.5 mm
    FFR solver bridges gaps below 1.5 mm before flow compute, which dampens propagation of β0 fragmentation into decisions—modeling choice affecting the clinical-facing endpoint.
  • Baseline loss weights λd, λCE, λv = Cv: 0.35/0.15/0.35; topo stage: 0.5/0.5/0
    Cv uses λd=0.35, λCE=0.15, λv=0.35; topology stage uses λd=λCE=0.5, λv=0—fixed recipe shared across comparisons.
axioms (6)
  • domain assumption ImageCAS manual (or dataset-provided) vessel masks are adequate ground truth for both volumetric metrics and junction topology.
    All BCS/BCR/FFR comparisons treat dataset labels as reference anatomy (Sec. 1.1, 1.2).
  • domain assumption 26-connected skeleton junctions (≥3 neighbors), stub reachability after dilation, and β0 component counts are the right discrete model of clinically relevant connectedness.
    BCS construction and connectedness summaries in Sec. 1.2–1.3.
  • domain assumption Reduced-order Poiseuille segment resistances plus lumped outlets and a 0.80 min-FFR rule are a sufficient decision model to study geometric FFR-CT fidelity.
    Sec. 1.5 FFR readout; authors note it is not physiological validation against invasive FFR.
  • standard math Standard differentiable segmentation losses (Dice, CE) and named topology losses (clDice, Skeleton Recall) behave as in the cited literature.
    Training objective Eq. 1 and loss configurations in Sec. 1.1.
  • domain assumption Patient-clustered bootstrap/GEE with Holm correction appropriately handles repeated measures across architectures, losses, and seeds on the same 250 cases.
    Statistical analysis Sec. 1.6 and Results 2.2.
  • ad hoc to paper Fixed α=0.05 and shared Cv initialization make the three topology losses fairly comparable.
    Explicit design choice in Sec. 1.1 to avoid per-loss tuning advantage.
invented entities (3)
  • Bifurcation Connectedness Score (BCS) independent evidence
    purpose: Fraction of GT bifurcations whose stubs remain mutually reachable in the (dilated) prediction.
    Core metric; preliminary form in [12], characterized here against standard metrics and FFR decisions.
  • soft-BCS loss independent evidence
    purpose: Differentiable surrogate emphasizing the weakest stub probability at each GT junction (Eq. 3).
    Training objective compared to clDice and Skeleton Recall; empirical handle via BCS/BCR/β0/FFR tables.
  • Branch Correspondence Ratio (BCR) independent evidence
    purpose: Harmonic mean of GT/pred branch match rates under chamfer matching to separate recovery from assembly.
    Operational definition used to claim soft-BCS and Skeleton Recall recover the same branches.

pith-pipeline@v1.2.0-daily-grok45 · 14607 in / 4569 out tokens · 82120 ms · 2026-07-31T11:12:32.714749+00:00 · methodology

0 comments
read the original abstract

Fractional flow reserve derived from CT angiography (FFR-CT) simulates flow through a patient-specific vessel model, so its accuracy depends on the connectedness of the segmented tree, not only on volumetric overlap: a segmentation can reach high Dice yet sever a bifurcation, dropping the downstream subtree and reversing the treatment decision. Topology-aware losses such as clDice and Skeleton Recall act on the global centreline and can miss localised breaks. We study the Bifurcation Connectedness Score (BCS), which scores connectedness at each ground-truth bifurcation, and soft-BCS, its differentiable training surrogate. BCS captures a property of segmentation quality the standard metrics miss: it responds strongly to breaks in connectedness while staying largely unchanged under connectedness-preserving narrowing. Higher BCS accompanies closer agreement between the FFR-CT decisions a solver makes on predicted versus ground-truth geometry, most clearly in severe disease (OR 2.16, CI [1.23, 4.18]). Both decisions come from the same solver, so this reflects geometric, not clinical, fidelity. In training, soft-BCS and Skeleton Recall recover the same branches but build different trees. Recovering branches and keeping them connected are separable properties, so we recommend reporting a measure of each.

Figures

Figures reproduced from arXiv: 2607.28327 by Dr Vinod Venugopal, James Brown, Kelvin Lee, Maame Owusu-Ansah, Muhammad Moazzam Jawaid, Wenting Duan.

Figure 1
Figure 1. Figure 1: Two segmentations with near-identical Dice scores, but vastly different BCS values. (a) Vesselness baseline (Cv): Dice 0.87, BCS 0.43. (b) Soft-BCS model (Ctopo): Dice 0.88, BCS 0.71. Voxels coloured by FFR; red marks ischaemic segments (≤0.80) [20]. In (a) a disconnected junction drops the downstream branch from the solver, deferring the case; in (b) the junction is preserved and the case is treated. Desp… view at source ↗
Figure 2
Figure 2. Figure 2: Methods overview. (a) BCS construction: ground-truth junctions are detected, l=8-voxel stubs are taken along each outgoing branch and tested through the δ=3-voxel￾dilated prediction; a junction is preserved only if all stubs remain mutually reachable, and BCS is the fraction preserved. (b) FFR-CT readout: the segmentation is reduced to a centreline graph, each segment is assigned a Poiseuille resistance, l… view at source ↗
Figure 3
Figure 3. Figure 3: Same branches, different trees. (a, b) For one representative case, Cskel and Ctopo recover the same ground-truth branches (equal BCR) but Cskel scatters them into more disconnected components while Ctopo keeps a single tree. (c, d) Paired distributions: BCR is equivalent across losses, while β0 is higher for Cskel [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

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

22 extracted references · 5 canonical work pages

  1. [1]

    In: Proceed- ings of the 5th International Joint Conference on Artificial Intelligence (IJCAI)

    Barrow, H.G., Tenenbaum, J.M., Bolles, R.C., Wolf, H.C.: Parametric correspon- dence and chamfer matching: Two new techniques for image matching. In: Proceed- ings of the 5th International Joint Conference on Artificial Intelligence (IJCAI). pp. 659–663. Cambridge, MA (1977)

    1977

  2. [2]

    Burzotta, F., Lassen, J., Lefèvre, T., et al.: Percutaneous coronary intervention for bifurcation coronary lesions: the 15th consensus document from the European Bifurcation Club. EuroIntervention: Journal of EuroPCR in Collaboration with the Working Group on Interventional Cardiology of the European Society of Cardiology 16(16), 1307–1317 (Mar 2021).http...

    work page doi:10.4244/eij-d-20-00169 2021

  3. [3]

    In: Wells, W.M., Colchester, A., Delp, S

    Frangi, A.F., Niessen, W.J., Vincken, K.L., Viergever, M.A.: Multiscale vessel enhancement filtering. In: Wells, W.M., Colchester, A., Delp, S. (eds.) Medical Image Computing and Computer-Assisted Intervention — MICCAI’98. pp. 130–137. Springer, Berlin, Heidelberg (1998).https://doi.org/10.1007/BFb0056195

    doi:10.1007/bfb0056195 1998

  4. [4]

    Computerized Medical Imaging and Graphics97, 102049 (Apr 2022).https://doi

    Gharleghi, R., Adikari, D., Ellenberger, K., Ooi, S.Y., Ellis, C., Chen, C.M., Gao, R., He, Y., Hussain, R., Lee, C.Y., Li, J., Ma, J., Nie, Z., Oliveira, B., Qi, Y., Skandarani, Y., Vilaça, J.L., Wang, X., Yang, S., Sowmya, A., Beier, S.: Automated segmentation of normal and diseased coronary arteries – The ASOCA challenge. Computerized Medical Imaging a...

    arXiv 2022

  5. [5]

    Hu, X., Fuxin, L., Samaras, D., Chen, C.: Topology-Preserving Deep Image Segmentation (Jun 2019).https://doi.org/10.48550/arXiv.1906.05404, http: //arxiv.org/abs/1906.05404, arXiv:1906.05404 [cs.CV]

    Pith/arXiv arXiv doi:10.48550/arxiv.1906.05404 2019

  6. [6]

    Huang, Z., Wang, H., Deng, Z., Ye, J., Su, Y., Sun, H., He, J., Gu, Y., Gu, L., Zhang, S., Qiao, Y.: STU-Net: Scalable and Transferable Medical Image Segmentation Models Empowered by Large-Scale Supervised Pre-training (2023).https://doi. 10 M. Owusu-Ansah et al. org/10.48550/ARXIV.2304.06716, https://arxiv.org/abs/2304.06716, version Number: 1

    Pith/arXiv arXiv doi:10.48550/arxiv.2304.06716 2023

  7. [7]

    Kirchhoff, Y., Rokuss, M.R., Roy, S., Kovacs, B., Ulrich, C., Wald, T., Zenk, M., Vollmuth, P., Kleesiek, J., Isensee, F., Maier-Hein, K.: Skeleton Recall Loss for Connectivity Conserving and Resource Efficient Segmentation of Thin Tubular Structures (Apr 2024),https://arxiv.org/abs/2404.03010v2

    Pith/arXiv arXiv 2024

  8. [8]

    Journal of Cardiovascular Computed Tomography 8(5), 342–358 (Sep 2014).https://doi.org/10.1016/j.jcct.2014.07.003

    Leipsic, J., Abbara, S., Achenbach, S., Cury, R., Earls, J.P., Mancini, G.J., Nieman, K., Pontone, G., Raff, G.L.: SCCT guidelines for the interpretation and reporting of coronary CT angiography: A report of the Society of Cardiovascular Computed To- mography Guidelines Committee. Journal of Cardiovascular Computed Tomography 8(5), 342–358 (Sep 2014).http...

    work page doi:10.1016/j.jcct.2014.07.003 2014

  9. [9]

    The Lancet

    Lin, A., Manral, N., McElhinney, P., et al.: Deep learning-enabled coronary CT angiography for plaque and stenosis quantification and cardiac risk prediction: an international multicentre study. The Lancet. Digital health4(4), e256–e265 (Apr 2022).https://doi.org/10.1016/S2589-7500(22)00022-X

    work page doi:10.1016/s2589-7500(22)00022-x 2022

  10. [10]

    Nature methods21(2), 195–212 (Feb 2024).https: //doi.org/10.1038/s41592-023-02151-z

    Maier-Hein, L., Reinke, A., Godau, P., et al.: Metrics Reloaded: Recommendations for image analysis validation. Nature methods21(2), 195–212 (Feb 2024).https: //doi.org/10.1038/s41592-023-02151-z

    doi:10.1038/s41592-023-02151-z 2024

  11. [11]

    JAMA308(12), 1237–1245 (Sep 2012)

    Min, J., Leipsic, J., Pencina, M., et al.: Diagnostic accuracy of fractional flow reserve from anatomic CT angiography. JAMA308(12), 1237–1245 (Sep 2012). https://doi.org/10.1001/2012.jama.11274

    doi:10.1001/2012.jama.11274 2012

  12. [12]

    In: Proceedings of the Medical Image Understanding and Analysis (MIUA) Conference

    Owusu-Ansah, M., Brown, J., Duan, W., Jawaid, M.: Adapt or Preserve? Encoder Strategy for Topology-Aware CT Foundation Models. In: Proceedings of the Medical Image Understanding and Analysis (MIUA) Conference. Springer, Dublin, Ireland (2026)

    2026

  13. [13]

    https://doi.org/10.48550/arXiv.2501.09001, http://arxiv.org/ abs/2501.09001, arXiv:2501.09001 [eess]

    Pai, S., Hadzic, I., Bontempi, D., Bressem, K., Kann, B.H., Fedorov, A., Mak, R.H., Aerts, H.J.W.L.: Vision Foundation Models for Computed Tomography (Feb 2025). https://doi.org/10.48550/arXiv.2501.09001, http://arxiv.org/ abs/2501.09001, arXiv:2501.09001 [eess]

    Pith/arXiv arXiv doi:10.48550/arxiv.2501.09001 2025

  14. [14]

    Pfaller, M.R., Pham, J., Verma, A., Pegolotti, L., Wilson, N.M., Parker, D.W., Yang, W., Marsden, A.L.: Automated generation of 0D and 1D reduced-order models of patient-specific blood flow. International Journal for Numerical Methods in Biomedical Engineering38(10), e3639 (Oct 2022).https://doi.org/10.1002/ cnm.3639,http://arxiv.org/abs/2111.04878, arXiv...

    Pith/arXiv arXiv 2022

  15. [15]

    In: Greenspan, H., Madabhushi, A., Mousavi, P., Salcudean, S., Duncan, J., Syeda-Mahmood, T., Taylor, R

    Qiu,Y.,Li,Z.,Wang,Y.,Dong,P.,Wu,D.,Yang,X.,Hong,Q.,Shen,D.:CorSegRec: A Topology-Preserving Scheme for Extracting Fully-Connected Coronary Arteries from CT Angiography. In: Greenspan, H., Madabhushi, A., Mousavi, P., Salcudean, S., Duncan, J., Syeda-Mahmood, T., Taylor, R. (eds.) Medical Image Computing and Computer Assisted Intervention – MICCAI 2023. pp...

    work page doi:10.1007/978-3-031-43898-1_64 2023

  16. [16]

    Journal of the American College of Cardiology76(25), 2982–3021 (Dec 2020)

    Roth, G.A., Mensah, G.A., Johnson, C.O., et al.: Global Burden of Cardiovascular Diseases and Risk Factors, 1990-2019: Update From the GBD 2019 Study. Journal of the American College of Cardiology76(25), 2982–3021 (Dec 2020). https: //doi.org/10.1016/j.jacc.2020.11.010

    doi:10.1016/j.jacc.2020.11.010 1990

  17. [17]

    In: 2021 IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR)

    Shit, S., Paetzold, J.C., Sekuboyina, A., Ezhov, I., Unger, A., Zhylka, A., Pluim, J.P.W., Bauer, U., Menze, B.H.: clDice – A Novel Topology-Preserving Loss Function for Tubular Structure Segmentation. In: 2021 IEEE/CVF Con- ference on Computer Vision and Pattern Recognition (CVPR). pp. 16555– 16564 (Jun 2021). https://doi.org/10.1109/CVPR46437.2021.01629...

    arXiv 2021

  18. [18]

    https://doi.org/10.48550/arXiv.2111.14791, http://arxiv

    Tang, Y., Yang, D., Li, W., Roth, H., Landman, B., Xu, D., Nath, V., Hatamizadeh, A.: Self-Supervised Pre-Training of Swin Transformers for 3D Medical Image Anal- ysis (Mar 2022). https://doi.org/10.48550/arXiv.2111.14791, http://arxiv. org/abs/2111.14791, arXiv:2111.14791 [cs.CV]

    Pith/arXiv arXiv doi:10.48550/arxiv.2111.14791 2022

  19. [19]

    Journal of the American College of Cardiology61(22), 2233–2241 (Jun 2013).https://doi.org/10.1016/j.jacc.2012.11.083

    Taylor, C.A., Fonte, T.A., Min, J.K.: Computational fluid dynamics applied to cardiac computed tomography for noninvasive quantification of fractional flow reserve: scientific basis. Journal of the American College of Cardiology61(22), 2233–2241 (Jun 2013).https://doi.org/10.1016/j.jacc.2012.11.083

    doi:10.1016/j.jacc.2012.11.083 2013

  20. [20]

    The New England Journal of Medicine360(3), 213–224 (Jan 2009)

    Tonino, P., De Bruyne, B., Pijls, N., et al.: Fractional flow reserve versus angiography for guiding percutaneous coronary intervention. The New England Journal of Medicine360(3), 213–224 (Jan 2009). https://doi.org/10.1056/NEJMoa0807611

    work page doi:10.1056/nejmoa0807611 2009

  21. [21]

    Yang, K., Musio, F., Ma, Y., et al.: Benchmarking the CoW with the TopCoW Challenge: Topology-Aware Anatomical Segmentation of the Circle of Willis for CTA and MRA (Jul 2025).https://doi.org/10.48550/arXiv.2312.17670, http: //arxiv.org/abs/2312.17670, arXiv:2312.17670 [cs] version: 4

    Pith/arXiv arXiv doi:10.48550/arxiv.2312.17670 2025

  22. [22]

    Computerized Medical Imaging and Graphics 109, 102287 (Oct 2023).https://doi.org/10.1016/j.compmedimag.2023.102287

    Zeng, A., Wu, C., Lin, G., Xie, W., Hong, J., Huang, M., Zhuang, J., Bi, S., Pan, D., Ullah, N., Khan, K.N., Wang, T., Shi, Y., Li, X., Xu, X.: ImageCAS: A large- scale dataset and benchmark for coronary artery segmentation based on computed tomography angiography images. Computerized Medical Imaging and Graphics 109, 102287 (Oct 2023).https://doi.org/10....

    arXiv 2023