Pith. sign in

REVIEW 4 major objections 7 minor 1 cited by

Dynamic-Computed Tomography Angiography for Cerebral Vessel Templates and Segmentation

T0 review · 4 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Dynamic CTA can supply the training labels that make automatic artery and vein segmentation on conventional CTA feasible.

desk verdict Genuinely new CT angiographic atlases, but the DL-vs-atlas comparison is inflated by sensitivity-only metrics and shared iCafe labels. read the letter →

arxiv 2502.09893 v1 pith:KR436RPA submitted 2025-02-14 physics.med-ph cs.CV

Shrikanth Yadav , Jisoo Kim , Geoffrey Young , Lei Qin This is my paper · ORCID
classification physics.med-phcs.CV
keywords computedtomographyangiographydynamic4D-CTAcerebralvesselsegmentationartery-veinseparationangiographicatlasdeeplearninglabelingmodifiedDicecoefficient
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper asks whether the vessels visible in ordinary CT angiography can be segmented automatically, despite CTA's poor background suppression compared with magnetic resonance angiography. It exploits dynamic 4D-CTA, a time-resolved acquisition, to subtract bone and soft tissue and to separate arteries from veins by comparing contrast arrival voxel by voxel. Those separated volumes let the authors build what they describe as the first CT angiographic atlases for arteries and veins, and they let a semi-automatic MRA vessel-tracing tool generate training labels without full manual annotation. A deep learning model trained on bone-in CTA images with those labels outperformed atlas-based registration on an 11-patient test set, reporting average modified Dice of 0.856 for arteries and 0.743 for veins, versus 0.324 and 0.495 for the atlas approach. If the labels are trustworthy, the practical upshot is that automatic artery-vein segmentation on routine CTA is within reach.

What carries the argument

The central mechanism is a voxel-wise conditional subtraction rule. After rigidly registering the arterial-phase and venous-phase CTA series, the algorithm keeps a subtracted arterial voxel only when its intensity exceeds the corresponding venous value, suppressing veins, and keeps a subtracted venous voxel only in the reverse comparison, suppressing arteries. This produces an artery-only volume and a vein-only volume for every patient. Those volumes serve two roles: they are inputs to the iterative multivariate template-construction pipeline that yields the first arterial and venous CTA atlases, and they are the images on which the semi-automatic vessel tracer produces the labels used to train and evaluate the deep learning models.

What would settle it

Have two expert neuroradiologists manually annotate arteries and veins on the same 11 test scans, then recompute the modified Dice alongside a symmetric metric such as full Dice or precision; if precision is much lower than the reported sensitivity, or if inter-rater agreement is poor, the claimed accuracy is not a reliable measure of segmentation quality.

Watch

Extended reading notes

Core claim

The paper's central claim is that the temporal information in dynamic CTA can be converted into high-quality vascular ground truth for conventional CTA. By subtracting the pre-contrast baseline and then comparing arterial-phase and venous-phase intensities voxel by voxel, the authors isolate arteries and veins into separate volumes, which allows them to construct, for the first time, dedicated CT angiographic atlases of intracranial arteries and veins. Using these vessel-separated volumes as input to a semi-automatic MRA tracing tool, they generate 20 arterial labels plus venous labels with only modest manual correction, and train deep learning models to predict all vessels directly from bone-in CTA images. On an 11-patient test set, the deep learning models exceeded template-registration segmentation by a wide margin on the sensitivity-oriented modified Dice metric (0.856 versus 0.324 for arteries, 0.743 versus 0.495 for veins), with the largest gains on proximal vessels including the internal carotid, middle cerebral M1, posterior cerebral P1, and anterior cerebral A1 segments.

Load-bearing premise

The whole comparison depends on the labels produced by a semi-automatic MRA vessel-tracing tool applied to the subtracted CTA images and checked only qualitatively by a neuroradiologist, with a sensitivity-oriented metric that never penalizes predictions that add false vessels.

Editorial extensions

If this is right

  • A model trained on a modest dynamic-CTA cohort can segment proximal arteries and major veins on bone-in CTA images that were never manually annotated.
  • The first CT angiographic atlases can serve as population templates for automated labeling of the ICA, vertebrobasilar, M1, A1, and P1 territories, although distal branches are not reliably captured.
  • Artery-vein separation is achievable from temporal information alone, making dynamic acquisition a source of ground truth rather than requiring expert manual tracing on static CTA.
  • Registration-based segmentation is competitive only when aligned to subtracted or vessel-separated volumes, not to raw CTA, so atlas guidance for routine CTA will need the deep learning route or an intermediate subtraction step.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The evaluation is sensitivity-focused, so the reported gap would likely shrink under a symmetric metric that counts false-positive vessels; a precision-aware comparison could test this, and the paper does not report one.
  • Because the deep learning model is trained on bone-in CTA across multiple contrast phases, a direct deployment path is conventional single-phase CTA acquired near peak arterial timing; the paper demonstrates phase robustness but does not test that transfer.
  • Distal-vessel segmentation remains weak (M2+, A2+, P2+), suggesting that the next bottleneck is small-vessel annotation rather than artery-vein separation; centerline supervision or topology-aware losses are natural extensions.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper proposes two automatic vessel segmentation approaches for conventional CTA: (1) construction of population-averaged arterial and venous CT angiographic templates from 4D-CTA data, followed by nonlinear registration to segment vessels via template-derived ROIs, and (2) deep learning segmentation models (nnU-Net and NexToU) trained on labels produced by the semi-automatic MRA tool iCafe applied to vessel-separated dynamic CTA volumes. The authors evaluate both approaches on 11 test patients using branch-wise metrics (modified Dice coefficient mDC, average directed Hausdorff distance adHD, and topology sensitivity tSens), reporting that deep learning outperforms atlas-based registration (arterial amDC 0.856 vs 0.324; venous amDC 0.743 vs 0.495). The main contributions claimed are the first angiographic CT atlases and a practical pipeline for generating CTA vessel ground truth with reduced manual effort.

Significance. If the claims are substantiated, the work would provide a useful contribution: the first CT angiographic templates, a method for leveraging dynamic CTA to separate arteries and veins, and a deep learning approach that may reduce annotation burden for CTA vessel segmentation. The use of public tools (ANTs, nnU-Net, iCafe) and a clearly described preprocessing pipeline are strengths. However, the central quantitative claim (DL superiority over atlas) rests on a sensitivity-only metric and on evaluation labels derived from the same iCafe pipeline used for training labels. The paper does not report precision, standard Dice, or any false-positive measure, nor a statistical comparison between DL and atlas results. The small, single-scanner, all-female evaluation cohort further limits generalizability. The novelty of the atlas construction is credible, but the segmentation claim as currently supported is fragile.

major comments (4)
  1. [§2.6.1 and §3.3] The modified Dice coefficient defined in §2.6.1 is mDC(Aj,P) = |Aj∩P|/|Aj|, which is pure recall (sensitivity). The headline comparisons in the abstract and §3.3, e.g., amDC 0.856 vs 0.324 for arteries, therefore penalize the atlas for missing ground-truth voxels but do not penalize either method for over-prediction. A prediction covering the entire image volume would achieve an mDC of 1.0 for every label. To support the claim that DL 'outperforms' atlas-based segmentation, the paper must report a precision-aware metric (e.g., standard Dice, precision, or volume overlap) or a false-positive-sensitive surface metric (e.g., bidirectional Hausdorff distance). Without this, the observed difference could reflect different recall/precision trade-offs rather than overall segmentation quality.
  2. [§2.5, §2.6.2, §3.3] There is a partial circularity in the DL evaluation: the training labels and the evaluation ground truth are both produced by the same iCafe pipeline applied to vessel-separated dynamic CTA volumes (§2.5). The deep learning model is therefore trained and scored on fidelity to iCafe, not on independent anatomical truth. The atlas-based segmentation is built independently of iCafe, making the DL-vs-atlas comparison asymmetric in label source. The limitations section (§4.3.2) acknowledges that iCafe was designed for MRA, required manual editing for veins, and that inter-rater studies are needed, but the paper does not quantify how much of the reported 0.856 vs 0.324 gap is attributable to reproducing iCafe-specific biases. The authors should evaluate the DL models on an independent manually annotated test set (or a subset at least) and report the agreement between iCafe-derived labels and manual expert labels.
  3. [§3.2, §3.3] The headline claim that DL outperforms atlas registration is not accompanied by any statistical comparison. The atlas results are presented in Figure 8 and Supplemental Table 2 (per-label mDC for Groups 1–3), while the DL results are given in Table 2 (per-phase averages). No paired test, confidence interval, or per-patient comparison between the two methods is reported. Given N=11 and the high variance visible in the atlas box plots, the observed differences may not be statistically robust. The authors should report per-label, per-patient paired comparisons (e.g., Wilcoxon signed-rank test) or at least confidence intervals for the amDC differences.
  4. [§2.4, §4.2, Table 1] The number of training patients is inconsistent across the manuscript. The abstract states that iCafe was used to segment vessels in 29 patients; §4.2 states 'a dataset with 102 training images from 27 patients'; Table 1 lists 'Training (27)'. Section 2.5 and Figure 4 do not specify the training cohort size. These numbers must be reconciled. The discrepancy affects the reader's ability to assess the claims about data scale and generalizability.
minor comments (7)
  1. [Abstract] The sentence 'Dynamic CTA is a type of imaging that captures temporal information about the We aim to develop' is grammatically broken; it should be revised.
  2. [§2.2] 'NifTi' should be 'NIfTI'.
  3. [§2.5.1] The label 'V A' (with a space) is used inconsistently; elsewhere it is 'VA'. Also 'Pcomm' vs 'PComm' capitalization should be made consistent.
  4. [§4.1] 'Smaller distal vessels appeared to be diffused' should likely be 'diffuse'.
  5. [§4.2] The phrase 'metrics that focused on sensitively' contains a typo; it should be 'sensitivity'.
  6. [Figure 2] '1937 imaged by the institution' should be '1937 images'.
  7. [§2.6.2] The metric adHD is defined as a sum over ground-truth surface points only (directed toward the predicted surface), so it does not penalize spurious predictions. This should at least be stated explicitly in the main text, not only in the Discussion, to avoid over-interpretation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: held-out evaluation of DL against iCafe-derived labels is a standard supervised generalization test, not a derivation equivalent to its inputs; the atlas comparison is independent of training labels.

full rationale

Walking the derivation chain, the only candidate circular link is that both training and evaluation labels for the DL models are produced by the same semi-automatic iCafe pipeline (Sections 2.4, 2.5, 2.6.2). This is a limitation in anatomical validity, not a circular derivation: the DL model never sees the evaluation patients' labels during training; patient-level splitting is stated in Section 2.4, and the test mDC therefore measures out-of-sample generalization of the learned mapping from bone-in CTA to iCafe-style labels. No equation in the paper defines the predictions in terms of the test labels, and no fitted parameter is re-evaluated on the data that produced it. The atlas-based comparison is also not circular because the atlas is built from a separate 25-patient cohort and its ROIs are threshold/manual, not derived from iCafe labels. The sensitivity-only mDC and the MRA-oriented iCafe tool are validity concerns explicitly acknowledged in Sections 2.6.1 and 4.3.2, but they do not make the reported DL-vs-atlas comparison equivalent to its inputs. There are no load-bearing self-citations: iCafe [16], nnU-Net [17], NexToU [18], and the template pipelines [24, 10] are external prior work with no author overlap. Hence no circular step meets the quoted-evidence bar.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The main external inputs are the scanner's subtraction, iCafe's labeling, and the threshold and registration choices. No new physical entities, forces, or conserved quantities are introduced; the angiographic atlases are constructed image templates, not postulated entities.

free parameters (2)
  • Template ROI threshold = 30 HU
    Section 2.3: binary arterial and venous ROIs are extracted by thresholding the atlases at 30 HU after largest-connected-component cleanup. No sensitivity analysis is provided, and this threshold directly sets the registration-based segmentation output.
  • Template convergence selection rule = DSC cutoff 98%; iteration with lowest RMSE, iteration 10
    Section 2.3 and Figure 7: the stopping rule is a choice made by the authors and influences which template is used for all downstream registration results.
assumptions (4)
  • domain assumption Scanner baseline subtraction removes bone and soft tissue accurately across all dynamic phases.
    Section 2.1 and Figure 1(a): the first pre-contrast frame is used by the scanner to subtract background; the whole pipeline operates on these subtracted images.
  • domain assumption A voxel brighter in the arterial phase than in the venous phase is artery, and vice versa.
    Algorithm 1, Steps 5-6: the conditional subtraction assumes no significant residual motion after rigid alignment and that arterial and venous phases have no overlapping contrast dynamics.
  • domain assumption iCafe vessel traces on subtracted CTA are valid ground truth for arteries and veins.
    Section 2.5: a tool designed for MRA is applied to CTA; radiologist validation is qualitative editing, not an independent quantitative standard.
  • ad hoc to paper Sensitivity-only modified Dice is a sufficient comparison metric.
    Section 2.6.1: mDC uses only the ground-truth volume in the denominator and never penalizes false positives, making it possible for over-predicted segmentations to score high.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dynamic-Computed Tomography Angiography for Cerebral Vessel Templates and Segmentation." pith.science (2026). https://pith.science/paper/KR436RPA

@misc{pith2026250209893,
  author       = {Pith},
  title        = {Pith review of: Dynamic-Computed Tomography Angiography for Cerebral Vessel Templates and Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KR436RPA}},
  note         = {Machine review of arXiv:2502.09893}
}
read the original abstract

Background: Computed Tomography Angiography (CTA) is crucial for cerebrovascular disease diagnosis. Dynamic CTA is a type of imaging that captures temporal information about the We aim to develop and evaluate two segmentation techniques to segment vessels directly on CTA images: (1) creating and registering population-averaged vessel atlases and (2) using deep learning (DL). Methods: We retrieved 4D-CT of the head from our institutional research database, with bone and soft tissue subtracted from post-contrast images. An Advanced Normalization Tools pipeline was used to create angiographic atlases from 25 patients. Then, atlas-driven ROIs were identified by a CT attenuation threshold to generate segmentation of the arteries and veins using non-linear registration. To create DL vessel segmentations, arterial and venous structures were segmented using the MRA vessel segmentation tool, iCafe, in 29 patients. These were then used to train a DL model, with bone-in CT images as input. Multiple phase images in the 4D-CT were used to increase the training and validation dataset. Both segmentation approaches were evaluated on a test 4D-CT dataset of 11 patients which were also processed by iCafe and validated by a neuroradiologist. Specifically, branch-wise segmentation accuracy was quantified with 20 labels for arteries and one for veins. DL outperformed the atlas-based segmentation models for arteries (average modified dice coefficient (amDC) 0.856 vs. 0.324) and veins (amDC 0.743 vs. 0.495) overall. For ICAs, vertebral and basilar arteries, DL and atlas -based segmentation had an amDC of 0.913 and 0.402, respectively. The amDC for MCA-M1, PCA-P1, and ACA-A1 segments were 0.932 and 0.474, respectively. Conclusion: Angiographic CT templates are developed for the first time in literature. Using 4D-CTA enables the use of tools like iCafe, lessening the burden of manual annotation.

Figures

Figures reproduced from arXiv: 2502.09893 by the authors.

Figure 1
Figure 1. Panel (a) Dynamic CTA from a sample patient. The axis on the left shows the acquisition time of each [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 3
Figure 3. (a) Preprocessing steps for a dynamic CTA from a single patient. For both phases, a head ROI is used [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Dynamic CTA for all patients are collected (Step (a)). The subtracted images are processed to suppress [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figures from the paper (9 more)
Figure 5
Figure 5. Figure 5: An overview of extracting the ground truth segmentations of veins and arteries using iCafe. Traces are [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Overview of the evaluation procedure. We register angiographic CT template to the multiphase CTA data [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: RMSE and DSC for each iteration in the template construction pipeline for all images [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Modified dice coefficient (mDC) for each arterial and venous label. Three registrations are tested for each [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 1
Figure 1. Figure 1: Axial images of arterial phase CTA (Xa), bone subtracted arterial and venous phase (Sa and Sv), and the vessel separated arterial and venous phase (S ∗ a and S ∗ v ) vol￾umes in a sample patient. 2 [PITH_FULL_IMAGE:figures/full_fig_p019_1.png]
Figure 2
Figure 2. Figure 2: Lightbox view of the 3D surface rendering of ROI∗ a (identi￾fied from T ∗ as) in red and ROI∗ v (identified from T ∗ vs) in blue. 3 [PITH_FULL_IMAGE:figures/full_fig_p020_2.png]
Figure 3
Figure 3. Figure 3: Lightbox view of the 3D surface rendering of the ground truth annotations from iCafe. 4 [PITH_FULL_IMAGE:figures/full_fig_p021_3.png]
Figure 4
Figure 4. Figure 4: Figure shows two examples where registration-based seg￾mentation succeeded and failed, respectively. Panel (a): 3D surface rendering of a poor registration where the ICAs and the vertebral ar￾teries are poorly aligned with the ROI. ICA (R), ICA (L), and M1 (R) were mod…
Figure 5
Figure 5. Figure 5: Modified Dice, Topology sensitivity, and average directed Hausdorff distance for [PITH_FULL_IMAGE:figures/full_fig_p025_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Automated anatomy-based post-processing reduces false positives and improved interpretability of deep learning intracranial aneurysm detection

    eess.IV 2025-07 conditional novelty 6.0 of 10

    Applying brain, artery, and vein segmentation masks after deep learning detection cut false positives by 51-71% in two aneurysm models with no loss of true positives on the test set.

Reference graph

Works this paper leans on

32 extracted references · 30 canonical work pages · cited by 1 Pith paper

  1. [1]

    Walluscheck, L

    S. Walluscheck, L. Canalini, H. Strohm, S. Diekmann, J. Klein, and S. Heldmann, MR-CT multi-atlas registration guided by fully automated brain structure segmentation with CNNs, International Journal of Computer Assisted Radiology and Surgery 18, 483–491 (2023)

    work page 2023

  2. [2]

    Dundamadappa, K

    S. Dundamadappa, K. Iyer, A. Agrawal, and D. Choi, Multiphase CT Angiography: A Useful Technique in Acute Stroke Imaging—Collaterals and Beyond, AJNR: American Journal of Neuroradiology 42, 221–227 (2021)

    work page 2021

  3. [3]

    P. S. Rajiah, N. Weber, J. Loewen, H. Kasten, E. Williamson, A. Moore, and S. Leng, Dynamic CT Angiog- raphy in Vascular Imaging: Principles and Applications, RadioGraphics 42, E224–E225 (2022), Publisher: Radiological Society of North America

    work page 2022

  4. [4]

    Yang et al., TopCoW: Benchmarking Topology-Aware Anatomical Segmentation of the Circle of Willis (CoW) for CTA and MRA, 2024, arXiv:2312.17670 [cs, q-bio] version: 2

    K. Yang et al., TopCoW: Benchmarking Topology-Aware Anatomical Segmentation of the Circle of Willis (CoW) for CTA and MRA, 2024, arXiv:2312.17670 [cs, q-bio] version: 2

    arXiv 2024

  5. [5]

    L. Rist, O. Taubmann, F. Thamm, H. Ditt, M. Sühling, and A. Maier, Bifurcation matching for consistent cerebral vessel labeling in CTA of stroke patients, International Journal of Computer Assisted Radiology and Surgery (2022)

    work page 2022

  6. [6]

    W. L. Nowinski, Evolution of Human Brain Atlases in Terms of Content, Applications, Functionality, and Availability, Neuroinformatics 19, 1–22 (2021)

    work page 2021

  7. [7]

    Mouches and N

    P. Mouches and N. D. Forkert, A statistical atlas of cerebral arteries generated using multi-center MRA datasets from healthy subjects, Scientific Data 6, 29 (2019), Number: 1 Publisher: Nature Publishing Group. 14

    work page 2019

  8. [8]

    Comparing Vessel Imaging: Noncontrast Computed Tomography/Computed Tomographic Angiography Should Be the New Minimum Standard in Acute Disabling Stroke - PubMed

    work page

Show all 32 references
  1. [9]

    Rorden, L

    C. Rorden, L. Bonilha, J. Fridriksson, B. Bender, and H.-O. Karnath, Age-specific CT and MRI templates for spatial normalization, NeuroImage 61, 957–965 (2012)

    2012

  2. [10]

    Muschelli, A Publicly Available, High Resolution, Unbiased CT Brain Template, in Information Processing and Management of Uncertainty in Knowledge-Based Systems , edited by M.-J

    J. Muschelli, A Publicly Available, High Resolution, Unbiased CT Brain Template, in Information Processing and Management of Uncertainty in Knowledge-Based Systems , edited by M.-J. Lesot, S. Vieira, M. Z. Reformat, J. P. Carvalho, A. Wilbik, B. Bouchon-Meunier, and R. R. Yage...

    2020

  3. [11]

    Chilamkurthy, R

    S. Chilamkurthy, R. Ghosh, S. Tanamala, M. Biviji, N. G. Campeau, V . K. Venugopal, V . Mahajan, P. Rao, and P. Warier, Deep learning algorithms for detection of critical findings in head CT scans: a retrospective study, Lancet (London, England) 392, 2388–2396 (2018)

    2018

  4. [12]

    S. Liu, R. Su, J. Su, J. Xin, J. Wu, W. van Zwam, P. J. van Doormaal, A. van der Lugt, W. J. Niessen, N. Zheng, and T. van Walsum, An automated framework for brain vessel centerline extraction from CTA images, 2024, arXiv:2401.07041 [cs, eess]

    2024 arXiv

  5. [13]

    G. D. M. Talou, S. Safaei, P. J. Hunter, and P. J. Blanco, Adaptive constrained constructive optimisation for complex vascularisation processes, Scientific Reports 11, 6180 (2021), Publisher: Nature Publishing Group

    2021

  6. [14]

    Di Noto, G

    T. Di Noto, G. Marie, S. Tourbier, Y . Alemán-Gómez, O. Esteban, G. Saliou, M. B. Cuadra, P. Hagmann, and J. Richiardi, Towards Automated Brain Aneurysm Detection in TOF-MRA: Open Data, Weak Labels, and Anatomical Knowledge, Neuroinformatics 21, 21–34 (2023)

    2023

  7. [15]

    OASIS-3: Longitudinal Neuroimaging, Clinical, and Cognitive Dataset for Normal Aging and Alzheimer Dis- ease | medRxiv

  8. [16]

    L. Chen, M. Mossa-Basha, N. Balu, G. Canton, J. Sun, K. Pimentel, T. S. Hatsukami, J.-N. Hwang, and C. Yuan, Development of a quantitative intracranial vascular features extraction tool on 3D MRA using semiautomated open-curve active contour vessel tracing, Magnetic Resonance ...

    2018

  9. [17]

    Isensee, P

    F. Isensee, P. F. Jaeger, S. A. A. Kohl, J. Petersen, and K. H. Maier-Hein, nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation, Nature Methods 18, 203–211 (2021), Number: 2 Publisher: Nature Publishing Group

    2021

  10. [18]

    P. Shi, X. Guo, Y . Yang, C. Ye, and T. Ma, NexToU: Efficient Topology-Aware U-Net for Medical Image Segmentation, (2023), Publisher: arXiv Version Number: 1

    2023

  11. [19]

    F. Zhao, Y . Chen, Y . Hou, and X. He, Segmentation of blood vessels using rule-based and machine-learning- based methods: a review, Multimedia Systems 25, 109–118 (2019)

    2019

  12. [20]

    F. Fu, J. Wei, M. Zhang, F. Yu, Y . Xiao, D. Rong, Y . Shan, Y . Li, C. Zhao, F. Liao, Z. Yang, Y . Li, Y . Chen, X. Wang, and J. Lu, Rapid vessel segmentation and reconstruction of head and neck angiograms using 3D con- volutional neural network, Nature Communications 11, 482...

    2020

  13. [21]

    Meijs and R

    M. Meijs and R. Manniesing, Artery and vein segmentation of the cerebral vasculature in 4D CT using a 3D fully convolutional neural network, inMedical Imaging 2018: Computer-Aided Diagnosis , volume 10575, pages 394–399, SPIE, 2018

    2018

  14. [22]

    R. Su, P. M. van der Sluijs, Y . Chen, S. Cornelissen, R. v. d. Broek, W. H. van Zwam, A. van der Lugt, W. Niessen, D. Ruijters, and T. van Walsum, CA VE: Cerebral Artery-Vein Segmentation in Digital Subtraction Angiography, 2023, arXiv:2208.02355 [eess]

    2023 arXiv

  15. [23]

    X. Li, P. S. Morgan, J. Ashburner, J. Smith, and C. Rorden, The first step for neuroimaging data analysis: DICOM to NIfTI conversion, Journal of Neuroscience Methods 264, 47–56 (2016). 15

    2016

  16. [24]

    B. B. Avants, P. Yushkevich, J. Pluta, D. Minkoff, M. Korczykowski, J. Detre, and J. C. Gee, The optimal template effect in hippocampus studies of diseased populations, NeuroImage 49, 2457–2466 (2010)

    2010

  17. [25]

    S. M. Ryan, B. Vestal, L. A. Maier, N. E. Carlson, and J. Muschelli, Template Creation for High Resolution Computed Tomography Scans of the Lung in R Software, Academic radiology 27, e204–e215 (2020)

    2020

  18. [26]

    Kikinis, S

    R. Kikinis, S. D. Pieper, and K. G. V osburgh, 3D Slicer: A Platform for Subject-Specific Image Analysis, Visualization, and Clinical Support, in Intraoperative Imaging and Image-Guided Therapy , edited by F. A. Jolesz, pages 277–289, Springer, New York, NY , 2014

    2014

  19. [27]

    Phansalkar, S

    N. Phansalkar, S. More, A. Sabale, and M. Joshi, Adaptive local thresholding for detection of nuclei in diversity stained cytology images, in 2011 International Conference on Communications and Signal Processing , pages 218–220, 2011

    2011

  20. [28]

    J. N. Kapur, P. K. Sahoo, and A. K. C. Wong, A new method for gray-level picture thresholding using the entropy of the histogram, Computer Vision, Graphics, and Image Processing 29, 273–285 (1985)

    1985

  21. [29]

    Y . Wang, A. Narayanaswamy, C.-L. Tsai, and B. Roysam, A broadly applicable 3-D neuron tracing method based on open-curve snake, Neuroinformatics 9, 193–217 (2011)

    2011

  22. [30]

    O. U. Aydin, A. A. Taha, A. Hilbert, A. A. Khalil, I. Galinovic, J. B. Fiebach, D. Frey, and V . I. Madai, An evaluation of performance measures for arterial brain vessel segmentation, BMC Medical Imaging 21, 113 (2021)

    2021

  23. [31]

    O. U. Aydin, A. A. Taha, A. Hilbert, A. A. Khalil, I. Galinovic, J. B. Fiebach, D. Frey, and V . I. Madai, On the usage of average Hausdorff distance for segmentation performance assessment: hidden error when used for ranking, European Radiology Experimental 5, 4 (2021)

    2021

  24. [32]

    S. Shit, J. C. Paetzold, A. Sekuboyina, I. Ezhov, A. Unger, A. Zhylka, J. P. W. Pluim, U. Bauer, and B. H. Menze, clDice - a Novel Topology-Preserving Loss Function for Tubular Structure Segmentation, in 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR...

    2021 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.