Pith. sign in

REVIEW 4 major objections 5 minor 21 references

Adaptive Particle-Based Shape Modeling for Anatomical Surface Correspondence

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read The paper claims that adaptive particle placement with a neighborhood correspondence loss and geodesic correction matches or beats uniform-particle PSM on distance and correspondence metrics at roughly half the particle budget.

desk verdict A solid incremental extension of PSM with a plausible half-particle-budget claim, but the empirical evidence lacks the error bars and parameter reporting needed to fully back it. read the letter →

arxiv 2507.07379 v1 pith:MI6ZHQNL submitted 2025-07-10 cs.CV

classification cs.CV
keywords statisticalshapemodelingparticle-basedanatomicalsurfacecorrespondenceradialbasisfunctionsgeodesicadaptivitypointdistributionmodels
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

The paper sets out to show that particle-based shape models can be made self-adaptive to local surface geometry without sacrificing cross-shape correspondence, and that this pays off in model efficiency. Its concrete claim is that a sampling loss with adaptivity weight $c=0.5$, paired with a new neighborhood correspondence loss and a geodesic correspondence correction algorithm, produces better or comparable maximum and mean two-way surface-to-surface distances and correspondence metrics than prior particle-based shape modeling, including a baseline run with twice the particle budget. The mechanism lets particles concentrate where the RBF-reconstructed surface deviates from the true surface while a periodic geodesic walk restores neighborhood consistency against a reference shape. On femur, liver, and right hip cohorts, the trade-off analysis shows moderate adaptivity improves surface representation, while excessive adaptivity degrades correspondence. If the claim holds, large medical cohort studies can build accurate shape models with fewer landmarks, cutting optimization cost.

What carries the argument

The load-bearing machinery is the interaction of four components: an implicit radial basis function surface that turns control point dipoles into a queryable signed-distance field; a sampling loss with adaptivity weight $c$ that uses RBF reconstruction error to attract particles to underrepresented regions; a neighborhood correspondence loss that compares scale-normalized local configurations instead of absolute positions; and the geodesic correspondence algorithm, which periodically walks mismatched particles toward reference neighbors along the surface and doubles as a convergence test. The algorithm identifies a particle as mismatched when its geodesic neighbors fall outside a factor of 1.5 times the closest-geodesic-neighbor distance, then fixes particles in order of most to least mismatched. These pieces work together to allow higher particle density on detailed anatomy without allowing the particle configuration to drift out of correspondence.

What would settle it

Take a cohort with one anatomically atypical shape, run the two-stage optimization at $c=0.5$, and then rerun it with the reference shape forced to be a typical cohort member instead of the ICP-selected one; if the max and mean surface-to-surface distances, or the correspondence mismatch count, change substantially, then the gains are tied to the specific regularization target rather than to the adaptivity mechanism itself.

Watch

Extended reading notes

Core claim

The central discovery is that adaptivity and correspondence are not inherently in conflict, provided correspondence is defined through local neighborhood configurations rather than absolute particle positions. The paper replaces the Frobenius-norm correspondence loss, which biases particles toward the mean shape, with a neighborhood correspondence loss that normalizes out translation, rotation, and scale before comparing each particle's neighborhood to the reference particle's neighborhood. The geodesic correspondence algorithm then fixes residual irregularities by moving the most mismatched particles along geodesic paths toward their reference neighbors, applied every 25 epochs and used as a convergence check. The paper reports that the $c=0.5$ configuration yields better or comparable maximum and mean two-way surface-to-surface distance and correspondence metrics than all tested alternatives, including PSM at double the particle count, with one femur mean-distance exception where PSM at 256 particles wins by a thin margin while the method still outperforms PSM at 128 particles.

Load-bearing premise

The load-bearing premise is that the ICP-selected reference shape and the fixed 1.5 times neighborhood threshold define the correct geodesic neighborhood structure for every shape in the cohort, so if the reference is unrepresentative or the threshold is biased, the regularization enforces correspondence to the wrong neighborhoods and the reported improvements may be an artifact of matching to that particular reference.

Editorial extensions

If this is right

  • A model built with $c=0.5$ achieves better or comparable surface-to-surface distance and correspondence metrics than uniform-particle PSM, meaning the same model quality is reachable with roughly half the landmarks.
  • The neighborhood correspondence loss supports a wider adaptivity range than the Frobenius loss, but past a certain adaptivity weight, particles concentrate on noisy or highly variable features and correspondence metrics degrade.
  • The geodesic correspondence algorithm removes missed correspondences at thin edges and small features, as seen in the liver and hip results, and provides a principled convergence criterion for the optimization.
  • Surface snapping after each iteration keeps particles exactly on the surface and accounts for a direct improvement in surface-to-surface distances over the prior extension.

Reading between the lines

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

  • A testable extension of this work would be to make $c$ spatially varying, letting flat regions keep uniform particle spacing while only high-curvature regions receive extra particles; the current two-stage schedule suggests such a curriculum could sharpen the adaptivity-correspondence trade-off.
  • Because the geodesic correction depends on one ICP-selected reference shape and a fixed 1.5 times neighborhood threshold, cohorts with outliers may need a robust or ensemble reference; this is an inference from the algorithm's structure, not a result the paper reports.
  • The reported particle-budget advantage suggests that adaptive models could replace the common practice of doubling particle counts for fine anatomy, but the paper does not test whether the advantage persists for very low particle counts or extremely high shape variability.
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 / 5 minor

Summary. The manuscript proposes an adaptive particle-based shape modeling method for anatomical surface correspondence. It extends the authors' RBF-shape optimization with two mechanisms: a neighborhood correspondence loss (Eq. 3) that permits local adaptivity without the mean-bias of the earlier Frobenius loss, and a geodesic correspondence algorithm (Algorithm 1) that periodically corrects neighborhood mismatches against a reference shape. The optimization combines a sampling loss with an adaptivity weight c, a correspondence loss, and an eigenshape loss, followed by surface snapping. Experiments on proximal femur, liver, and right-hip datasets compare the method with PSM at 128 and 256 particles and with the authors' prior method [12], using two-way surface-to-surface distances and compactness/generalization/specificity. The central claim is that c=0.5 gives better or comparable metrics than all tested methods, even PSM at twice the particle budget, with a stated exception on femur mean distance versus PSM-256.

Significance. If the empirical claims are robust, the method is a useful contribution: it offers a concrete mechanism for increasing particle adaptivity while preserving correspondence, and the half-budget comparison against PSM would improve the scalability of statistical shape modeling. The evaluation is substantive: three real anatomical datasets, standard SSM metrics, comparison with a strong baseline at two particle counts, and qualitative mode-of-variation figures. The losses are defined explicitly, the geodesic correction is given as pseudocode, and the evaluation uses external distance transforms and ground-truth meshes, so I do not see a circularity problem. The main weaknesses are statistical: the headline comparison rests on single runs without error bars or significance tests, the adaptivity weight c=0.5 is selected on the same data used for evaluation, and several parameters needed for reproduction are not reported. These issues are fixable, but they are load-bearing for the quantitative claims.

major comments (4)
  1. [Section III.E, Figs. 3-9] The central claim that c=0.5 yields better or comparable metrics than all tested methods is supported only by single-run point estimates with no error bars, confidence intervals, or significance tests. Because the margins over PSM are often small and the manuscript itself reports an exception for femur mean distance versus PSM-256, the comparison should be repeated over multiple restarts (or at least bootstrapped) and the blanket 'better or comparable' claim qualified accordingly.
  2. [Sections III.D and III.E] The paper first explores c as an exploratory trade-off axis and then fixes c=0.5 on the same three datasets for the headline comparison. No validation split or pre-specified model-selection rule is given, so the reported advantage may reflect selection on the evaluation data. Please report results over the full range of c considered (or use an independent validation set) to show that the c=0.5 outcome is not a post hoc choice, and quantify how sensitive the conclusions are to c.
  3. [Section III.C and Eq. (2)] Several parameters that are load-bearing for reproducibility are not reported: the dipole offset / narrow-band half-width s, the number of near-surface sample points R, the neighborhood size q in Eq. (3) (which uses the notation N6 without defining q=6 in the text), and the convergence tolerance in the geodesic algorithm. Without these values, the method cannot be reimplemented from the manuscript despite the otherwise detailed algorithm description. Please list all free parameters and stopping criteria.
  4. [Section II.C.1, Algorithm 1] The geodesic correction regularizes every shape toward the neighborhoods of a reference shape, but the manuscript does not state how the reference shape is chosen for Algorithm 1 (the ICP-based template description in Section II.B.2 is for the correspondence loss, not for the algorithm) and does not analyze sensitivity to that choice or to the fixed 1.5x geodesic-neighborhood factor. If the reference is atypical, the regularization could anchor correspondence to an unrepresentative topology and inflate the reported correspondence improvements. Please specify the reference-selection rule and test at least two alternative references and neighborhood factors.
minor comments (5)
  1. [Eq. (2)] The soft-minimum term softmin(K_i) is written as exp(-k)/sum_j exp(k), but the normalization axis is not stated; please clarify whether normalization is over control points for each near-surface point and explain how this implements a soft minimum.
  2. [Algorithm 1, step (c)] The update rule 'move particles by geodesically walking towards neighbors' is ambiguous; please provide the exact displacement formula or pseudo-code for where a particle is moved on the geodesic path.
  3. [Eq. (3)] The notation N6 in Eq. (3) is inconsistent with the general definition N_q in the surrounding text; please state explicitly that q=6 (or rename N6 to N_q).
  4. [Section III.C] For the PSM baseline, the paper says 'a concerted effort was made to find the optimal PSM hyperparameters' but does not list those hyperparameters or the software version; please provide them to make the comparison reproducible.
  5. [Section III.B] The derivation of the right-hip dataset from TotalSegmentator MR images is described only as 'processed with [9]'; please cite or describe the actual preprocessing pipeline so the dataset construction is reproducible.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the method's losses and the geodesic correspondence algorithm are evaluated against external ground-truth surfaces and independent baselines; self-citations [11][12] are lineage and baseline references only, not load-bearing.

full rationale

The derivation chain is self-contained. The sampling loss (Eq. 2) is evaluated pointwise against D_i, the signed distance transform of the true segmentation, so the optimization target is anchored to ground truth rather than to the particle system's own output. The reported surface-to-surface distances are computed between original meshes and meshes warped using particle correspondences, an external benchmark; compactness, generalization, and specificity are computed from PCA of the particle positions and leave-one-out reconstruction. The geodesic correspondence algorithm (Algorithm 1) is a regularizer that enforces neighborhood consistency with a chosen reference shape, but its success is measured by external metrics and by absence of warping artifacts (Figs. 7, 10, 11), not by the reference-neighborhood alignment it enforces. The adaptivity weight c is scanned and the c=0.5 result is reported as an experimental finding, not as a prediction of held-out data; although the selection is post hoc and error bars are absent, that is a reproducibility and statistical-support concern, not a circularity. Self-citations [11] and [12] are used to describe the lineage of RBF-based PSM and as baselines; they are not invoked as proof of the current method's claims, and the central comparisons are against external ground truths and independent approaches. No load-bearing argument reduces by construction to its own input.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The central claim depends on eight hand-set or dataset-tuned parameters, four of which are reported numerically and four of which are unreported in the manuscript. The method does not introduce new physical or computational entities; it combines established RBF representation, eigenshape optimization, and new regularization heuristics over a per-cohort reference shape.

free parameters (8)
  • c (adaptivity weight) = 0.5 (reported results); explored over [0,1]
    Adaptivity weight in Eq. (2); the final value was chosen after inspecting the trade-off between surface distance and correspondence in Figures 2-4, so it is fitted to the test data.
  • alpha (sampling loss weight) = 5 (second stage), 10 (first stage)
    Sampling loss weight in Eq. (5), set per dataset.
  • beta (eigenshape loss weight) = 0.05 (liver), 0.01 (hip and femur)
    Eigenshape loss weight in Eq. (5), set per dataset.
  • gamma (correspondence loss weight) = 5 (liver), 0.5 (femur), 0.1 (hip)
    Correspondence loss weight in Eq. (5), set per dataset.
  • s (dipole offset and narrow-band half-width) = not reported
    Defines dipole positions and the near-surface band for sampling loss in Eq. (2); never specified numerically.
  • R (number of near-surface sample points) = not reported
    Number of randomly sampled near-surface points B_i used in the sampling loss.
  • geodesic neighborhood factor = 1.5
    Algorithm 1 defines a neighbor as within 1.5 times the closest geodesic neighbor distance; this threshold directly controls the correction strength.
  • q (nearest neighbors in correspondence loss) = 6
    N6(pt,j) in Eq. (3) uses the 6 nearest neighbors; hand-specified.
assumptions (5)
  • standard math RBF interpolation with control points and dipoles produces an approximate signed distance field adequate for surface reconstruction.
    Invoked in Section II.A to build the RBF-shape from particles and dipoles; the correctness of this approximation is taken from Turk and O'Brien [13] and Carr et al. [14].
  • domain assumption The signed distance transform D_i derived from each binary segmentation is an accurate representation of the true anatomical surface.
    D_i is the ground truth for the sampling loss in Eq. (2); the paper does not analyze segmentation errors.
  • domain assumption All shapes in a cohort are pre-aligned into a common coordinate frame before optimization.
    The correspondence losses compare particle neighborhoods across shapes; the paper notes only that left femurs are reflected and that other datasets are processed with ShapeWorks [9], which includes alignment.
  • domain assumption PCA on corresponding particle positions yields clinically meaningful modes of shape variation.
    The eigenshape loss (Eq. 4) and the evaluation metrics (compactness, generalization, specificity) assume PCA is the right summary of population variability, standard in SSM.
  • domain assumption The automatically selected reference shape is representative of the cohort for geodesic regularization.
    Algorithm 1 and the neighborhood correspondence loss depend on a template shape t chosen via iterative closest point; an atypical template would bias the geodesic neighborhoods.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Adaptive Particle-Based Shape Modeling for Anatomical Surface Correspondence." pith.science (2026). https://pith.science/paper/MI6ZHQNL

@misc{pith2026250707379,
  author       = {Pith},
  title        = {Pith review of: Adaptive Particle-Based Shape Modeling for Anatomical Surface Correspondence},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MI6ZHQNL}},
  note         = {Machine review of arXiv:2507.07379}
}
read the original abstract

Particle-based shape modeling (PSM) is a family of approaches that automatically quantifies shape variability across anatomical cohorts by positioning particles (pseudo landmarks) on shape surfaces in a consistent configuration. Recent advances incorporate implicit radial basis function representations as self-supervised signals to better capture the complex geometric properties of anatomical structures. However, these methods still lack self-adaptivity -- that is, the ability to automatically adjust particle configurations to local geometric features of each surface, which is essential for accurately representing complex anatomical variability. This paper introduces two mechanisms to increase surface adaptivity while maintaining consistent particle configurations: (1) a novel neighborhood correspondence loss to enable high adaptivity and (2) a geodesic correspondence algorithm that regularizes optimization to enforce geodesic neighborhood consistency. We evaluate the efficacy and scalability of our approach on challenging datasets, providing a detailed analysis of the adaptivity-correspondence trade-off and benchmarking against existing methods on surface representation accuracy and correspondence metrics.

Figures

Figures reproduced from arXiv: 2507.07379 by the authors.

Figure 1
Figure 1. Illustration of the global RBF shape model, where the surface is populated with control points and their respective dipoles, derived from surface normals. A system of equations is constructed from these control points and dipoles, enabling efficient querying of approximate distances for points near the surface. II. METHODS An SSM encodes shape variation into a set of parameters designed to be as compact as possible.… view at source ↗
Figure 2
Figure 2. Each pair of columns shows the first and second modes of variation obtained from our method with different adaptivity weights c. For the next to last column, α = 10, which is double the default of α = 5. The last column shows the ones obtained from [12]. From left to right, observe that adaptivity increases and more detailed features are captured, but correspondence deteriorates, which is evident in the surface arti… view at source ↗
Figure 4
Figure 4. Two-way surface-to-surface distances on a log scale between original meshes and ones warped using particles for the femur dataset [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figures from the paper (6 more)
Figure 5
Figure 5. Figure 5: Two-way surface-to-surface distances on a log scale between original meshes and ones warped using particles for the liver dataset [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 3
Figure 3. Figure 3: Compactness, generalization, and specificity of tested PDM approaches for femurs [PITH_FULL_IMAGE:figures/full_fig_p004_3.png]
Figure 7
Figure 7. Figure 7: Shows sample shapes from each dataset with their individual warped reconstructions and the ground truth shape. Good correspondence manifests as reconstructed surfaces without spurious artifacts and folds absent in the ground truth shape. Excellent correspondence result…
Figure 8
Figure 8. Figure 8: Compactness, generalization, and specificity of tested PDM approaches for livers [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 10
Figure 10. Figure 10: The first mode of variation for the liver dataset for different strategies. The geodesic correspondence algorithm handles missed correspondences at the thin edges of the shape, which also aids in spreading particles more evenly [PITH_FULL_IMAGE:figures/full_fig_p009_…
Figure 11
Figure 11. Figure 11: The first mode of variation for the right hip for different strategies. The geodesic correspondence algorithm for the proposed approach procures better results without artifacts than [12] and PSM. The Autograd functionality of PyTorch 1.12.1 was used to automatically …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 19 canonical work pages

  1. [12]

    Optimization-driven statistical models of anatomies using radial basis function shape representation,

    H. Xu and S. Y . Elhabian, “Optimization-driven statistical models of anatomies using radial basis function shape representation,” in 21st IEEE International Symposium on Biomedical Imaging (ISBI 2024) . IEEE, 2024

  2. [1]

    Quantitative comparison of cortical bone thickness using correspondence-based shape modeling in patients with cam femoroacetabular impingement,

    P. R. Atkins, S. Y . Elhabian, P. Agrawal, M. D. Harris, R. T. Whitaker, J. A. Weiss, C. L. Peters, and A. E. Anderson, “Quantitative comparison of cortical bone thickness using correspondence-based shape modeling in patients with cam femoroacetabular impingement,” Journal of Orthopaedic Research, vol. 35, no. 8, pp. 1743–1753, 2017

  3. [2]

    Quantifying the severity of metopic craniosynostosis: A pilot study application of machine learning in craniofacial surgery,

    R. Bhalodia, L. A. Dvoracek, A. M. Ayyash, L. Kavan, R. Whitaker, and J. A. Goldstein, “Quantifying the severity of metopic craniosynostosis: A pilot study application of machine learning in craniofacial surgery,” Journal of Craniofacial Surgery , 2020

  4. [3]

    Statistical shape modeling of the talocrural joint using a hybrid multi-articulation joint approach,

    A. Lenz, N. Kr ¨ahenb¨uhl, A. Peterson, R. Lisonbee, B. Hintermann, C. Saltzman, A. Barg, and A. Anderson, “Statistical shape modeling of the talocrural joint using a hybrid multi-articulation joint approach,” Scientific Reports, vol. 11, pp. 7314, 04 2021

  5. [4]

    Statistical shape modeling of the hip and the association with hip osteoarthritis: a systematic review,

    M. van Buuren, N. Arden, S. Bierma-Zeinstra, W. Bramer, N. Casartelli, D. Felson, G. Jones, N. Lane, C. Lindner, N. Maffiuletti, J. van Meurs, A. Nelson, M. Nevitt, P. Valenzuela, J. Verhaar, H. Weinans, and R. Agri- cola, “Statistical shape modeling of the hip and the association with hip osteoarthritis: a systematic review,” Osteoarthritis and Cartilage...

  6. [5]

    High variability of acetabular offset in primary hip osteoarthritis influences acetabular reaming—a computed tomography–based anatomic study,

    C. Merle, M. M. Innmann, W. Waldstein, E. C. Pegg, P. R. Aldinger, H. S. Gill, D. W. Murray, and G. Grammatopoulos, “High variability of acetabular offset in primary hip osteoarthritis influences acetabular reaming—a computed tomography–based anatomic study,” The Journal of Arthroplasty, vol. 34, no. 8, pp. 1808–1814, 2019

  7. [6]

    A statistical shape modelling framework to extract 3d shape biomarkers from medical imaging data: assessing arch morphology of repaired coarctation of the aorta,

    J. L. Bruse, K. McLeod, G. Biglino, H. N. Ntsinjana, C. Capelli, T.-Y . Hsia, M. Sermesant, X. Pennec, A. M. Taylor, S. Schievano, et al., “A statistical shape modelling framework to extract 3d shape biomarkers from medical imaging data: assessing arch morphology of repaired coarctation of the aorta,” BMC medical imaging , vol. 16, pp. 1–19, 2016

  8. [7]

    Statistical shape and appearance models of bones,

    N. Sarkalkan, H. Weinans, and A. A. Zadpoor, “Statistical shape and appearance models of bones,” Bone, vol. 60, pp. 129–140, 2014

Show all 21 references
  1. [8]

    A minimum description length approach to statistical shape modeling,

    R. H. Davies, C. J. Twining, T. F. Cootes, J. C. Waterton, and C. J. Taylor, “A minimum description length approach to statistical shape modeling,” IEEE transactions on medical imaging , vol. 21, no. 5, pp. 525–537, 2002

  2. [9]

    Shapeworks: particle-based shape correspondence and visualization software,

    J. Cates, S. Elhabian, and R. Whitaker, “Shapeworks: particle-based shape correspondence and visualization software,” in Statistical Shape and Deformation Analysis , pp. 257–298. Elsevier, 2017

  3. [10]

    Entropy-based particle correspondence for shape populations,

    I. Oguz, J. Cates, M. Datar, B. Paniagua, T. Fletcher, C. Vachet, M. Styner, and R. Whitaker, “Entropy-based particle correspondence for shape populations,” International journal of computer assisted radiology and surgery, vol. 11, pp. 1221–1232, 2016

  4. [11]

    Image2ssm: Reimagining statistical shape models from images with radial basis functions,

    H. Xu and S. Y . Elhabian, “Image2ssm: Reimagining statistical shape models from images with radial basis functions,” in Medical Image Computing and Computer Assisted Intervention – MICCAI 2023 , H. Greenspan, A. Madabhushi, P. Mousavi, S. Salcudean, J. Duncan, T. Syeda-Mahmoo...

  5. [13]

    Variational Implicit Surfaces,

    G. Turk and J. F. O’Brien, “Variational Implicit Surfaces,” Technical report GIT-GVU-99-15, GVU Center, Georgia Institute of Technology, Atlanta, GA, 1996

  6. [14]

    Reconstruction and representation of 3d objects with radial basis functions,

    J. C. Carr, R. K. Beatson, J. B. Cherrie, T. J. Mitchell, W. R. Fright, B. C. McCallum, and T. R. Evans, “Reconstruction and representation of 3d objects with radial basis functions,” in Proceedings of the 28th Annual Conference on Computer Graphics and Interactive Techniques ...

  7. [15]

    Automatic construction of eigenshape models by direct optimization,

    A. C. Kotcheff and C. J. Taylor, “Automatic construction of eigenshape models by direct optimization,” Medical Image Analysis , vol. 2, no. 4, pp. 303–314, 1998

  8. [16]

    The farthest point strategy for progressive image sampling,

    Y . Eldar, M. Lindenbaum, M. Porat, and Y . Zeevi, “The farthest point strategy for progressive image sampling,” IEEE Transactions on Image Processing, vol. 6, no. 9, pp. 1305–1315, 1997

  9. [17]

    The Hungarian Method for the Assignment Problem,

    H. W. Kuhn, “The Hungarian Method for the Assignment Problem,” Naval Research Logistics Quarterly , vol. 2, no. 1–2, pp. 83–97, March 1955

  10. [18]

    Benchmarking off-the-shelf statistical shape modeling tools in clinical applications,

    A. Goparaju, K. Iyer, A. B ˆone, N. Hu, H. B. Henninger, A. E. Anderson, S. Durrleman, M. Jacxsens, A. Morris, I. Csecs, N. Marrouche, and S. Y . Elhabian, “Benchmarking off-the-shelf statistical shape modeling tools in clinical applications,” Medical Image Analysis , vol. 76,...

  11. [19]

    Abdomenct-1k: Is abdominal organ segmentation a solved problem?,

    J. Ma, Y . Zhang, S. Gu, C. Zhu, C. Ge, Y . Zhang, X. An, C. Wang, Q. Wang, X. Liu, S. Cao, Q. Zhang, S. Liu, Y . Wang, Y . Li, J. He, and X. Yang, “Abdomenct-1k: Is abdominal organ segmentation a solved problem?,” IEEE Transactions on Pattern Analysis and Machine Intelligence...

  12. [20]

    Totalseg- mentator mri: Sequence-independent segmentation of 59 anatomical structures in mr images,

    T. Akinci D’Antonoli, L. Berger, A. Indrakanti, N. Vishwanathan, J. Weiß, M. Jung, Z. Berkarda, A. Rau, M. Reisert, T. K¨ustner, A. Walter, E. Merkle, M. Segeroth, J. Cyriac, S. Yang, and J. Wasserthal, “Totalseg- mentator mri: Sequence-independent segmentation of 59 anatomica...

  13. [21]

    nnu-net: Self-adapting framework for u-net-based medical image segmentation,

    F. Isensee, J. Petersen, A. Klein, D. Zimmerer, P. F. Jaeger, S. Kohl, J. Wasserthal, G. Koehler, T. Norajitra, S. Wirkert, et al., “nnu-net: Self-adapting framework for u-net-based medical image segmentation,” arXiv preprint arXiv:1809.10486 , 2018

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.