pith. sign in

arxiv: 2605.16980 · v1 · pith:YMAZVBRDnew · submitted 2026-05-16 · 💻 cs.CV

Statistical Hand Shape Modeling from Clinical CT Scans Using Deep Learning and Implicit Skinning

Gokce Guven , Hasan Fehmi Ates , Deniz Karasahin , Kaan Erdogan This is my paper

Pith reviewed 2026-05-19 20:15 UTC · model grok-4.3

classification 💻 cs.CV
keywords statistical shape modelinghand anatomyCT segmentationdeep learningimplicit skinningnon-rigid registrationprincipal component analysisanthropometric validation
0
0 comments X

The pith

A pipeline cleans CT scans with AI, aligns hands via bone-driven skinning and registration, then builds a PCA shape model validated against army survey data.

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

The paper develops a full pipeline to turn clinical elbow-to-hand CT scans into statistical 3D models of hand shape variation. It first removes plaster casts and artifacts using a Pix2Pix GAN, segments skin and bones, then uses implicit skinning from the bone meshes to bring every hand into a common pose before applying non-rigid registration across subjects. Principal component analysis on the registered surfaces of 90 fully visible hands produces shape distributions that match measurements from the U.S. Army ANSUR II survey. A sympathetic reader would care because such models could supply population-level anatomical priors for ergonomics, prosthetics, and diagnostics without requiring new large-scale manual measurements.

Core claim

The authors show that an automated sequence of artifact removal by conditional GAN, mask extraction in 3D Slicer, skeletal implicit skinning for pose standardization, GBCPD++ non-rigid registration for correspondence, and PCA on 90 selected hand surfaces yields statistical shape distributions in strong agreement with the ANSUR II anthropometric survey, thereby supporting the anatomical validity of the reconstructed models.

What carries the argument

Implicit skinning driven by segmented bone meshes, which standardizes pose before GBCPD++ establishes point-wise skin correspondences for PCA shape modeling.

If this is right

  • The models can supply population priors for biomechanical hand simulations.
  • Ergonomic tools and workspaces can be sized against the captured shape variability.
  • Prosthetic and orthotic design can reference the principal modes of variation.
  • Precision diagnostics gain a quantitative baseline for detecting abnormal hand morphology.

Where Pith is reading between the lines

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

  • The same cleaning-plus-skinning-plus-registration steps could be applied to other articulated body segments such as feet or forearms from existing clinical archives.
  • Once correspondences exist, the pipeline could be retrained on larger or more diverse CT collections to test whether additional principal components appear in non-military populations.
  • The resulting shape space might serve as a generative prior for single-view 3D hand reconstruction from photographs or low-dose X-rays.

Load-bearing premise

The GBCPD++ non-rigid registration produces accurate point-wise correspondences across the 90 hands without systematic distortions that would alter the subsequent PCA components.

What would settle it

A direct comparison of hand length, width, and joint spacing statistics extracted from the PCA model against independent 3D surface scans or caliper measurements from a new cohort of at least several hundred adults would falsify the agreement claim if the distributions diverge beyond measurement error.

Figures

Figures reproduced from arXiv: 2605.16980 by Deniz Karasahin, Gokce Guven, Hasan Fehmi Ates, Kaan Erdogan.

Figure 1
Figure 1. Figure 1: Example skin and bone meshes extracted from Pix2Pix-cleaned elbow-to-hand CT volumes using the proposed [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Manual placement of skeletal landmarks used to construct the articulated hand skeleton for implicit skinning. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Example results of the Pix2Pix preprocessing stage. Top row: Original axial CT slices including background [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of pose-normalized hand models after implicit skinning across representative subjects. Top [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Representative hand shapes derived from the PCA-based statistical shape model. Top row: frontal views. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
read the original abstract

Accurate segmentation and statistical shape modeling of hand anatomy have significant implications for medical diagnostics, ergonomics, and biomechanics. This study proposes an AI-assisted reconstruction pipeline for segmenting and analyzing hand anatomy from 1,271 elbow-to-hand (e2h-CT) computed tomography scans. A Pix2Pix-based conditional generative adversarial network is first employed to remove plaster cast and background artifacts from CT volumes. The cleaned scans are then processed in 3D Slicer to extract skin and bone masks, which are converted into closed-surface mesh models. Segmented bone meshes are used to construct skeletal representations, enabling implicit skinning to align all hand models into a standardized anatomical configuration. Subsequently, non-rigid registration is performed on the hand skin surfaces using the Geodesic Based Coherent Point Drift++ (GBCPD++) algorithm to establish point-wise correspondence across subjects. Principal Component Analysis (PCA) is then applied to the registered models to quantify anatomical shape variability. The Pix2Pix preprocessing stage achieved a Dice coefficient of 0.9856 and an IoU of 0.9720 on the held-out test set. Statistical modeling was performed on a subset of 90 scans in which the fingers were fully visible and anatomically separated. The resulting statistical shape distributions demonstrate strong agreement with the U.S. Army Anthropometric Survey (ANSUR II), supporting the anatomical validity of the reconstructed models. The proposed methodology demonstrates significant potential for advancing biomechanical modeling, ergonomic optimization, prosthetic design, and precision medical diagnostics.

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

1 major / 2 minor

Summary. The paper presents a pipeline for statistical hand shape modeling from 1,271 clinical elbow-to-hand CT scans. A Pix2Pix GAN cleans plaster cast artifacts, followed by skin and bone segmentation in 3D Slicer, implicit skinning for alignment, GBCPD++ non-rigid registration on a filtered subset of 90 scans with visible separated fingers, and PCA to derive shape variability. It reports Dice 0.9856 and IoU 0.9720 for the GAN stage and claims strong agreement of the resulting shape distributions with ANSUR II anthropometric data, supporting anatomical validity for applications in biomechanics and ergonomics.

Significance. If the non-rigid registration produces reliable anatomical correspondences, the approach could enable scalable construction of hand shape models from existing clinical CT data, offering value for prosthetic design, ergonomic optimization, and medical diagnostics. The strong reported segmentation metrics for the preprocessing stage provide a solid foundation for the pipeline's feasibility.

major comments (1)
  1. [Methods (GBCPD++ non-rigid registration)] In the non-rigid registration step using GBCPD++ on the skin surfaces of the 90 selected subjects (following implicit skinning), the manuscript provides no quantitative validation metrics such as landmark error, target registration error, or consistency across subjects. This is load-bearing for the central claim because inaccurate point-wise correspondences, especially around fingers and joints, could introduce non-anatomical distortions into the registered meshes, making the PCA-derived principal components and the claimed ANSUR II agreement potentially artifacts of the registration rather than true anatomical variability.
minor comments (2)
  1. [Abstract] The abstract states that the statistical shape distributions demonstrate 'strong agreement' with ANSUR II but supplies no quantitative details such as mean differences, standard deviations, or correlation coefficients for specific measurements; these should be added to substantiate the claim.
  2. [Methods (subject selection)] Additional details on the exact criteria and potential biases in selecting the 90 scans (out of 1,271) with fully visible and separated fingers would help assess whether the captured shape variability is representative.

Simulated Author's Rebuttal

1 responses · 1 unresolved

We thank the referee for their constructive and detailed review. The feedback on validation of the non-rigid registration is well-taken, and we address it directly below while clarifying the validation strategy used in the study.

read point-by-point responses

  1. Referee: [Methods (GBCPD++ non-rigid registration)] In the non-rigid registration step using GBCPD++ on the skin surfaces of the 90 selected subjects (following implicit skinning), the manuscript provides no quantitative validation metrics such as landmark error, target registration error, or consistency across subjects. This is load-bearing for the central claim because inaccurate point-wise correspondences, especially around fingers and joints, could introduce non-anatomical distortions into the registered meshes, making the PCA-derived principal components and the claimed ANSUR II agreement potentially artifacts of the registration rather than true anatomical variability.

    Authors: We agree that quantitative registration validation would strengthen the manuscript. However, the clinical CT dataset does not contain pre-existing anatomical landmarks, making direct computation of landmark error or target registration error infeasible without substantial additional manual annotation that was outside the original study scope. GBCPD++ was selected specifically because it preserves geodesic distances on the surface, which is advantageous for maintaining finger separation. The 90-subject subset was further filtered to those with clearly visible and separated fingers, followed by visual inspection of alignments. The resulting PCA model shows strong quantitative agreement with independent ANSUR II anthropometric distributions across multiple hand dimensions; such agreement is unlikely to arise from grossly distorted correspondences. In revision we will expand the Methods and Discussion sections with additional registration parameter details, qualitative before/after registration visualizations, and a clearer statement of the indirect validation approach via downstream anthropometric agreement. revision: partial

standing simulated objections not resolved
  • Direct quantitative metrics such as landmark error or target registration error for GBCPD++, because the clinical CT scans lack ground-truth anatomical landmarks.

Circularity Check

0 steps flagged

No significant circularity; pipeline grounded in external data and standard methods.

full rationale

The derivation proceeds from CT scans through Pix2Pix artifact removal (Dice 0.9856 on held-out test set), 3D Slicer segmentation, implicit skinning for pose standardization, GBCPD++ non-rigid registration on 90 selected meshes, and PCA to obtain shape distributions. These distributions are then compared to the independent external ANSUR II anthropometric survey. No equations or steps reduce by construction to fitted parameters, self-citations, or ansatzes imported from the authors' prior work; the central claim of anatomical validity rests on external benchmark agreement rather than internal redefinition or renaming of known results.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The paper applies standard algorithms and software tools without introducing new free parameters, axioms beyond routine statistical assumptions, or invented entities; the main addition is the integrated application to this dataset.

axioms (2)
  • standard math PCA on registered point clouds captures the dominant modes of anatomical shape variation
    Applied directly to the aligned skin meshes to quantify variability.
  • domain assumption GBCPD++ produces sufficiently accurate non-rigid correspondences for downstream statistical modeling
    Invoked to establish point-wise correspondence across subjects prior to PCA.

pith-pipeline@v0.9.0 · 5814 in / 1268 out tokens · 69823 ms · 2026-05-19T20:15:03.482872+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages

  1. [1]

    Napier, "Hands", revised ed

    J. Napier, "Hands", revised ed. Princeton, NJ: Princeton University Press, 1993

    work page 1993

  2. [2]

    Anthropometry of the Hand and Product Design,

    A. W. Moustafa and I. Gharib, “Anthropometry of the Hand and Product Design,”International Design Journal, vol. 14, no. 6, pp. 1–19, Nov. 2024. DOI: 10.21608/idj.2024.378407

    work page doi:10.21608/idj.2024.378407 2024

  3. [3]

    Ergonomics and equipment design

    Schmidtke, Heinz. "Ergonomics and equipment design." Ergonomic data for equipment design. Boston, MA: Springer US, 1984. 1-13

    work page 1984

  4. [4]

    U-Net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, "U-Net: Convolutional networks for biomedical image segmentation," inMedical Image Computing and Computer-Assisted Intervention (MICCAI), Springer, Cham, 2015, pp. 234–241

    work page 2015

  5. [5]

    Active shape models-their training and application,

    T. F. Cootes, C. J. Taylor, D. H. Cooper, and J. Graham, "Active shape models-their training and application," Computer Vision and Image Understanding, vol. 61, no. 1, pp. 38–59, 1995

    work page 1995

  6. [6]

    Statistical shape models: understanding and mastering variation in anatomy

    Ambellan, Felix, et al. "Statistical shape models: understanding and mastering variation in anatomy." Biomedi- cal Visualisation: V olume 3. Cham: Springer International Publishing, 2019. 67-84

    work page 2019

  7. [7]

    Geometric skinning with approximate dual quaternion blending,

    L. Kavan, S. Collins, J. Žára, and C. O’Sullivan, "Geometric skinning with approximate dual quaternion blending,"ACM Trans. Graph., vol. 27, no. 4, pp. 1–23, 2008

    work page 2008

  8. [8]

    Bounded biharmonic weights for real-time deformation,

    A. Jacobson, I. Baran, J. Popovi´c, and O. Sorkine, "Bounded biharmonic weights for real-time deformation," ACM Trans. Graph., vol. 30, no. 4, pp. 78:1–78:8, 2011

    work page 2011

  9. [9]

    Implicit skinning: Real-time skin deformation with contact modeling,

    R. Vaillant et al., "Implicit skinning: Real-time skin deformation with contact modeling,"ACM Trans. Graph., vol. 32, no. 4, pp. 125:1–125:12, 2013. 10 APREPRINT- MAY19, 2026

    work page 2013

  10. [10]

    (2013, July)

    Dionne, O., de Lasa, M. (2013, July). Geodesic voxel binding for production character meshes. In Proceedings of the 12th ACM SIGGRAPH/Eurographics Symposium on Computer Animation (pp. 173-180)

    work page 2013

  11. [11]

    2012 Anthropometric Survey of U.S. Army Personnel: Methods and Summary Statistics,

    C. C. Gordon, C. L. Blackwell, B. Bradtmiller, J. L. Parham, P. Barrientos, S. P. Paquette, B. D. Corner, J. M. Carson, J. C. Venezia, B. M. Rockwell, M. Mucher, and S. Kristensen, “2012 Anthropometric Survey of U.S. Army Personnel: Methods and Summary Statistics,” U.S. Army Natick Soldier Research, Development and Engineering Center, Natick, MA, Tech. Re...

    work page 2012

  12. [12]

    C., Meyer, M

    Wasserthal, J., Breit, H. C., Meyer, M. T., Pradella, M., Hinck, D., Sauter, A. W., Segeroth, M. (2023). TotalSegmentator: robust segmentation of 104 anatomic structures in CT images. Radiology: Artificial Intelligence, 5(5), e230024

    work page 2023

  13. [13]

    A., Helgason, B., Palsson, H., Sigurdsson, S., Gudnason, V ., Ellingsen, L

    Bjornsson, P. A., Helgason, B., Palsson, H., Sigurdsson, S., Gudnason, V ., Ellingsen, L. M. (2021, February). Automated femur segmentation from computed tomography images using a deep neural network. In Medical Imaging 2021: Biomedical Applications in Molecular, Structural, and Functional Imaging (V ol. 11600, pp. 324-330). SPIE

    work page 2021

  14. [14]

    Deep learning for automated segmentation of pelvic muscles, fat, and bone from CT studies for body composition assessment,

    R. Hemke, C. G. Buckless, A. Tsao, B. Wang, and M. Torriani, “Deep learning for automated segmentation of pelvic muscles, fat, and bone from CT studies for body composition assessment,”Skeletal Radiology, vol. 49, no. 3, pp. 387–395, 2020

    work page 2020

  15. [15]

    A morphable model for the synthesis of 3D faces,

    V . Blanz and T. Vetter, “A morphable model for the synthesis of 3D faces,” inProc. 26th Annual Conf. Computer Graphics and Interactive Techniques (SIGGRAPH ’99), ACM Press, 1999, pp. 187–194

    work page 1999

  16. [16]

    Joshi, S., Davis, B., Jomier, M., Gerig, G. (2004). Unbiased diffeomorphic atlas construction for computational anatomy. NeuroImage, 23, S151-S160

    work page 2004

  17. [17]

    Shetty, K., Birkhold, A., Jaganathan, S., Strobel, N., Egger, B., Kowarschik, M., Maier, A. (2023). BOSS: Bones, organs and skin shape model. Computers in Biology and Medicine, 165, 107383

    work page 2023

  18. [18]

    SMPL: A skinned multi-person linear model,

    M. Loper, N. Mahmood, J. Romero, G. Pons-Moll, and M. J. Black, “SMPL: A skinned multi-person linear model,”ACM Trans. Graph., vol. 34, no. 6, pp. 248:1–248:16, Oct. 2015

    work page 2015

  19. [19]

    Y ., Zhou, T., Efros, A

    Isola, P., Zhu, J. Y ., Zhou, T., Efros, A. A. (2017) Image-to-image translation with conditional adversarial networks. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 1125-1134)

    work page 2017

  20. [20]

    Rhee, T., Neumann, U., Lewis, J. P. (2006, March). Human hand modeling from surface anatomy. In Proceedings of the 2006 symposium on Interactive 3D graphics and games (pp. 27-34)

    work page 2006

  21. [21]

    Endo, Y ., Tada, M., Mochimaru, M. (2014). Reconstructing individual hand models from motion capture data. Journal of Computational Design and Engineering, 1(1), 1-12

    work page 2014

  22. [22]

    S., Tajdari, F., Song, Y

    Yang, Y ., Yuan, T., Huysmans, T., Elkhuizen, W. S., Tajdari, F., Song, Y . (2021). Posture-invariant three dimensional human hand statistical shape model. Journal of Computing and Information Science in Engineering, 21(3), 031006

    work page 2021

  23. [23]

    (2004, August)

    Kurihara, T., Miyata, N. (2004, August). Modeling deformable human hands from medical images. In Proceed- ings of the 2004 ACM SIGGRAPH/Eurographics symposium on Computer animation (pp. 355-363)

    work page 2004

  24. [24]

    Luo, Y . (2024). Toward fully automated personalized orthopedic treatments: innovations and interdisciplinary gaps. Bioengineering, 11(8), 817

    work page 2024

  25. [25]

    Hirose, O. (2022). Geodesic-based bayesian coherent point drift. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(5), 5816-5832

    work page 2022

  26. [26]

    3D Slicer as an Image Computing Platform for the Quantitative Imaging Network

    Fedorov A., Beichel R., Kalpathy-Cramer J., Finet J., Fillion-Robin J-C., Pujol S., Bauer C., Jennings D., Fennessy F.M., Sonka M., Buatti J., Aylward S.R., Miller J.V ., Pieper S., Kikinis R. 3D Slicer as an Image Computing Platform for the Quantitative Imaging Network. Magnetic Resonance Imaging. 2012 Nov;30(9):1323-41. PMID: 22770690. PMCID: PMC3466397. 11

    work page 2012