Pith. sign in

REVIEW 4 major objections 5 minor 20 references

From Motion to Meaning: Biomechanics-Informed Neural Network for Explainable Cardiovascular Disease Identification

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

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

Pith's one-line read A biomechanics-informed registration pipeline extracts local shear and bulk moduli from cine-MRI and, combined with volumetric features, classifies five cardiac conditions with 100% accuracy on the ACDC test set.

desk verdict The 100% test accuracy is not what it seems: the classifier is fed ground-truth segmentations at test time, and the paper's own Dice numbers don't match the abstract. read the letter →

arxiv 2507.05783 v1 pith:ZOQPVJYI submitted 2025-07-08 cs.LG

Comte Valentin , Gemma Piella , Mario Ceresa , Miguel A. Gonzalez Ballester This is my paper · ORCID
classification cs.LG
keywords cardiaccine-MRIdeformableimageregistrationNeo-Hookeanbiomechanicscardiovasculardiseaseclassificationstrain-basedfeaturesexplainableAIACDCdataset
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 tries to establish that cardiac disease can be classified from the motion of the heart rather than only from its shape. It builds a deep-learning registration pipeline that tracks how cine-MRI frames deform between end-diastole and end-systole, regularizes the deformation field with a Neo-Hookean strain-energy term, and converts the resulting deformation into local estimates of shear modulus, bulk modulus, and strain. These biomechanical markers, combined with volumetric features from expert segmentations, are fed to a logistic-regression classifier. On the ACDC benchmark, the pipeline reaches 98% training accuracy and 100% held-out test accuracy across five cardiac conditions. If that result survives independent testing, it would mean motion-derived mechanical features carry diagnostic information that static anatomy alone misses, and that the predictions come with a biomechanical explanation doctors can inspect.

What carries the argument

The load-bearing object is the Neo-Hookean strain-energy regularizer $\Phi(\varphi)=\frac{\mu}{2}(\hat I_1^C-3)+\frac{\kappa}{2}(J-1)^2$, where $\hat I_1^C$ is the deviatoric first invariant of the right Cauchy-Green tensor and $J=\det(F)$ is the local volume change. From the predicted deformation field $\varphi$, the model forms the deformation gradient $F$, splits its energy into distortional and volumetric parts, and inverts the strain-energy expression voxel-wise to produce local shear modulus $\mu(p)$ and bulk modulus $\kappa(p)$ estimates. Those per-voxel moduli, summarized by mean, standard deviation, and 10th/90th percentiles inside six anatomical labels, plus volumetric features from ground-truth segmentations, form the classification feature set. The regularizer does double duty: it constrains the registration to nearly volume-preserving, physically plausible motion and simultaneously produces the mechanical features used downstream for diagnosis.

What would settle it

Replace the ground-truth segmentations on the ACDC test set with segmentations produced by an automatic model, then rerun the same feature extraction and logistic-regression pipeline; if test accuracy drops from 100%, the perfect classification depends on test-time ground-truth masks rather than on the biomechanical features alone.

Watch

Extended reading notes

Core claim

The central claim is that biomechanics-informed image registration converts cardiac cine-MRI into physically interpretable features that separate normal hearts, myocardial infarction, dilated cardiomyopathy, hypertrophic cardiomyopathy, and abnormal right ventricle with perfect held-out accuracy on the ACDC dataset. The authors assert that embedding Neo-Hookean hyperelastic energy as a regularizer in a cascaded deformable-registration network yields physiologically plausible deformation fields (Dice 0.974/0.956 for LV, 0.912/0.857 for RV, 0.960/0.922 for myocardium under multi-frame propagation), and that local shear and bulk moduli estimated voxel-wise from those fields, added to volumetric features, give logistic regression a training accuracy of 0.98 and test accuracy of 1.0. They further claim the selected biomechanical features are not redundant with volume: removing them lowers accuracy, and each contributes above-chance accuracy alone, which supports the paper's thesis that cardiac mechanics carries classification-relevant signal.

Load-bearing premise

The pipeline's claimed 100% test accuracy assumes that expert segmentations of the heart chambers are available at test time, because the volumetric features and the six anatomical regions used to summarize the biomechanical features are computed from ground-truth masks.

Editorial extensions

If this is right

  • If the 100% test accuracy holds up, biomechanical features derived from registered motion can serve as diagnostic markers for five cardiac conditions with a transparent mechanism: each prediction is traceable to local stiffness and strain patterns rather than to a black-box image statistic.
  • The registration results imply that a single physics-regularized network can propagate segmentations across cardiac phases with Dice scores comparable to or better than a dedicated segmentation network for the left ventricle and myocardium, enabling temporal label propagation without retraining.
  • Physicians could inspect which anatomical region and which modulus drove a given classification, because the selected features are named mechanical quantities, such as the 10th percentile of $\mu$ in a specific myocardial segment at end-systole, rather than latent activations.
  • Because the model maintains accuracy above 0.9 with training sets as small as 60 subjects, the feature set may transfer to smaller clinical cohorts where deep classifiers usually overfit.

Reading between the lines

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

  • The paper leaves untested whether the 100% test figure survives when ground-truth segmentations are replaced by automatic ones; the volumetric features encode disease-related chamber geometry, so a segmenter that errs on diseased hearts could lower the real-world ceiling.
  • The fixed material constants $\mu = 2$ kPa and $\kappa/\mu = 50$ are a hidden sensitivity knob; varying them could reveal whether classification relies on relative strain patterns or on absolute moduli, and the authors' proposed learnable, data-driven parameters could turn the pipeline into a stiffness-mapping diagnostic.
  • The same registration-plus-strain pipeline could be pointed at other dynamic imaging tasks, such as left-atrial strain in atrial fibrillation or motion tracking in other organs, wherever a deformation field can be estimated and tissue mechanics are disease-relevant; the paper stops at cardiac ventricles.
  • Because feature selection was performed on the same 100-subject training set whose accuracy is reported, the specific feature ranking may be tuned to this benchmark; an external-cohort replication is needed before treating the list as general.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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. This manuscript proposes a cascaded deep-learning image registration model regularized by a Neo-Hookean strain-energy term and applies it to cine-MRI from the ACDC dataset. The registration is evaluated by propagating ground-truth labels and comparing Dice scores with nnU-Net. The authors then extract features from locally defined shear and bulk moduli within six anatomical labels, together with volumetric features, and use a greedy feature-selection procedure with five classifiers. Logistic regression reportedly achieves 98% training and 100% test accuracy. The paper claims that this constitutes a biomechanics-informed, explainable cardiovascular disease identification pipeline.

Significance. If the claims held, the work would be significant: it would connect physics-informed registration with downstream classification and propose interpretable biomechanical features for cardiac diagnosis. The multi-frame label-propagation strategy and the per-feature ablation analysis are useful elements. However, the current evidence does not support the headline claims. Classification is oracle-assisted by ground-truth segmentations at test time, the so-called biomechanical parameters are algebraic rescalings of the regularization terms, and the registration comparison is against a different task. These are load-bearing validity issues, not presentation problems. The paper also provides no code or detailed training configuration, which further limits reproducibility.

major comments (4)
  1. [Section 2.4 and Table 4] The classification features are computed from ground-truth segmentations at test time: volumetric features are "extracted from the ground-truth segmentation at ED and ES," and biomechanical features are computed within six anatomical labels. No automatic segmentation module is described for inference, and the registration-based multi-frame label propagation of Section 2.3 is not used to produce test-time masks for the classifier. The reported 98% training and 100% test accuracy is therefore obtained with oracle access to disease-relevant chamber geometry, so the abstract's claim of an automatic diagnostic system from cine-MRI is not supported. To make the claim load-bearing, the authors would need to run the full pipeline with automatically produced segmentations and report classification performance on those.
  2. [Section 2.2.3, Eqs. (23)-(25)] Equations (23)-(25) define mu(p) and kappa(p) as the neighborhood average of Phi_dis (resp. Phi_vol) divided by the local value, multiplied by the global constants mu and kappa. Since Phi_dis and Phi_vol are exactly the two terms of the Neo-Hookean regularization in Eq. (21), the extracted "biomechanical parameters" are by construction deterministic functions of the model's own objective and of the assumed literature values mu = 2 kPa and kappa/mu = 50. They are not measured tissue properties, and the classification features therefore do not independently establish that the model has learned biomechanical characteristics of the tissue. At minimum, the manuscript should rename these quantities (e.g., "deformation-energy features") and temper the physical-interpretation claims.
  3. [Section 3.1, Table 2] The registration evaluation compares Dice of propagated ground-truth labels (propagation and multi-frame) against the Dice of nnU-Net's direct segmentation output. These are different tasks: label propagation transforms existing expert masks, while nnU-Net predicts segmentations from intensities. The table can only support a claim that propagated labels achieve certain Dice, not that the proposed registration is more accurate than a state-of-the-art segmentation method. A registration baseline (e.g., VoxelMorph or a classical DIR method) evaluated by the same label-propagation protocol is needed for the stated comparison.
  4. [Algorithm 1 and Section 3.2] Algorithm 1 repeatedly computes "classification accuracy acc" without specifying which split is used, and Table 4 reports "Validation" and "Testing" columns although no validation set is defined in the dataset description (which gives 100 training and 50 test cases). If the feature selection or model selection used the test set, the reported accuracies would be optimistically biased. The authors must specify the exact data splits used for selection and for reporting Table 4 and Figure 4, and ideally perform nested cross-validation.
minor comments (5)
  1. [Abstract and Table 2] The abstract reports Dice scores of 0.945, 0.908, and 0.905 for LV, RV, and myocardium, but Table 2's multi-frame results are 0.974/0.956, 0.912/0.857, and 0.960/0.922; the abstract numbers do not match any row of the table.
  2. [Table 4] Headings such as "phi 1,3 ratio ED" and "V all ratio ES" are not defined; a feature glossary or formula list is needed to understand which quantities are being ablated.
  3. [Section 3.2] Calling 100% test accuracy on 50 patients "unprecedented" is overstated; the corresponding confidence interval is wide, and the claim should be tempered.
  4. [Section 2.2.3] The abstract states that local strains are estimated, but the methods only compute Neo-Hookean energy terms; the strain tensor is never defined or reported, so the wording should be aligned with the actual computation.
  5. [Reproducibility] The paper does not provide code, learning rates, patch sizes, the number of cascaded networks, or the training epochs used for the ACDC experiment; Section 2.1 refers to prior work, but the specific configuration for this study should be reported.

Circularity Check

1 steps flagged · score 5.0 of 10

Local shear/bulk moduli are algebraic rescalings of the Neo-Hookean regularizer in the training loss, so the biomechanical-feature claim is self-definitional; classification labels are independent, yielding partial circularity.

  1. self definitional [Section 2.2.3, Equations (23)-(25)]
    "Hence we obtain: µ(p) = µ· P pi Φdis(pi,φ(pi)) / (d3·Φdis(p,φ(p))), κ(p) = κ· P pi Φvol(pi,φ(pi)) / (d3·Φvol(p,φ(p))). These estimates allow for the biomechanical properties of the tissue to be inferred at voxel level, providing a more accurate representation of cardiac mechanics."

    Equations (23)-(25) define the 'estimated local moduli' µ(p) and κ(p) as normalized versions of Φdis and Φvol, which are exactly the two terms of the Neo-Hookean energy Φ(φ) in Eq. (17)-(19) used as the regularizer λ·Φ(φ) in the training loss L = Lsim + λ·Φ(φ) (Eqs. (20)-(21)). Thus the 'inferred biomechanical properties' are, by construction, reciprocal-weighted functions of the model's own regularization objective, not independent measurements of tissue stiffness. The disease labels are not involved, so the classification result itself is not tautological; what reduces by definition is the claim that voxel-level shear/bulk moduli are being estimated.

full rationale

The paper's registration architecture is adapted from the authors' prior work (Comte et al., 2025), but that self-citation is incidental and validated on ACDC; it is not load-bearing. The genuine circularity is in Section 2.2.3: Eq. (25) defines local μ(p) and κ(p) as rescaled versions of Φdis and Φvol, the very energy terms minimized by the Neo-Hookean regularizer in the training loss (Eqs. (17)-(21)). Consequently, the 'biomechanical features' in Table 4 are not independent physical measurements but algebraic transforms of the model's objective, weakening the explainability claim. Because classification labels are independent of the regularizer, the 98%/100% accuracy is not forced by this construction. A separate, non-circular validity concern is that volumetric features are computed from ground-truth segmentations at ED and ES (Section 2.4), giving the classifier oracle access to disease-correlated geometry at test time; this is a leakage issue, not a derivation-cycle, and is not counted in the score.

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

The central claim depends on a handful of chosen constants (lambda, kappa/mu, mu, patch sizes), unvalidated physical modeling assumptions, and an ad hoc definition of local moduli. No new physical entities are introduced.

free parameters (4)
  • regularization weight lambda = not reported
    Balances image similarity and Neo-Hookean energy in Eq. (20); no sensitivity analysis is provided.
  • bulk-to-shear modulus ratio kappa/mu = 50
    Assumed for nearly incompressible cardiac tissue in Section 2.2.3; no sensitivity analysis.
  • shear modulus mu = 2 kPa
    Taken from Pislaru et al. (2014) as global anchor for local modulus estimates; external input not fitted here.
  • patch size d3 and neighborhood sizes = not reported
    Used in Eq. (3) and Eqs. (23)-(24) for multi-scale NCC and modulus averaging; values not given.
assumptions (4)
  • domain assumption Neo-Hookean hyperelastic model represents cardiac tissue mechanics
    Section 2.2 assumes isotropic, nearly incompressible hyperelastic behavior; authors acknowledge anisotropy is ignored in limitations.
  • domain assumption The registration deformation field equals physical cardiac tissue motion
    Section 2.2.2 treats the predicted DF as the deformation of cardiac tissue, ignoring through-plane motion in stacked 2D slices and registration error.
  • standard math VoxelMorph-style local cross-correlation is an adequate image similarity measure
    Loss from Balakrishnan et al. (2019), treated as standard in the field.
  • ad hoc to paper Equation (25) yields valid local moduli
    The local moduli are defined as rescaled regularization energies, not derived from independent mechanical measurements.

how reviews work

0 comments
Cite this review

Pith. "Pith review of From Motion to Meaning: Biomechanics-Informed Neural Network for Explainable Cardiovascular Disease Identification." pith.science (2026). https://pith.science/paper/ZOQPVJYI

@misc{pith2026250705783,
  author       = {Pith},
  title        = {Pith review of: From Motion to Meaning: Biomechanics-Informed Neural Network for Explainable Cardiovascular Disease Identification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZOQPVJYI}},
  note         = {Machine review of arXiv:2507.05783}
}
read the original abstract

Cardiac diseases are among the leading causes of morbidity and mortality worldwide, which requires accurate and timely diagnostic strategies. In this study, we introduce an innovative approach that combines deep learning image registration with physics-informed regularization to predict the biomechanical properties of moving cardiac tissues and extract features for disease classification. We utilize the energy strain formulation of Neo-Hookean material to model cardiac tissue deformations, optimizing the deformation field while ensuring its physical and biomechanical coherence. This explainable approach not only improves image registration accuracy, but also provides insights into the underlying biomechanical processes of the cardiac tissues. Evaluation on the Automated Cardiac Diagnosis Challenge (ACDC) dataset achieved Dice scores of 0.945 for the left ventricular cavity, 0.908 for the right ventricular cavity, and 0.905 for the myocardium. Subsequently, we estimate the local strains within the moving heart and extract a detailed set of features used for cardiovascular disease classification. We evaluated five classification algorithms, Logistic Regression, Multi-Layer Perceptron, Support Vector Classifier, Random Forest, and Nearest Neighbour, and identified the most relevant features using a feature selection algorithm. The best performing classifier obtained a classification accuracy of 98% in the training set and 100% in the test set of the ACDC dataset. By integrating explainable artificial intelligence, this method empowers clinicians with a transparent understanding of the model's predictions based on cardiac mechanics, while also significantly improving the accuracy and reliability of cardiac disease diagnosis, paving the way for more personalized and effective patient care.

Figures

Figures reproduced from arXiv: 2507.05783 by the authors.

Figure 1
Figure 1. Overview of the cascaded registration model, adapted from [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Labels used to compute the local biomechanical features: 1) right segment of the left ventricular myocardium, [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Confusion matrices of the classification on the training set and testing set. [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Accuracy as a function of the training set size. [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 17 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  3. [3]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  4. [4]

    VoxelMorph: a learning framework for deformable medical image registration

    Balakrishnan G, Zhao A, Sabuncu MR, Guttag J, Dalca AV. VoxelMorph: a learning framework for deformable medical image registration. IEEE Transactions on Medical Imaging 2019;38(8):1788--1800

    work page 2019

  5. [5]

    Deep learning techniques for automatic MRI cardiac multi-structures segmentation and diagnosis: is the problem solved? IEEE Transactions on Medical Imaging 2018;37(11):2514--2525

    Bernard O, Lalande A, Zotti C, Cervenansky F, Yang X, Heng P, et al. Deep learning techniques for automatic MRI cardiac multi-structures segmentation and diagnosis: is the problem solved? IEEE Transactions on Medical Imaging 2018;37(11):2514--2525

    work page 2018

  6. [6]

    Finite element analysis of air supported membrane structures

    Bonet J, Wood RD, Mahaney J, Heywood P. Finite element analysis of air supported membrane structures. Computer Methods in Applied Mechanics and Engineering 2000;190(5-7):579--595

    work page 2000

  7. [7]

    A radiomics approach to computer-aided diagnosis with cardiac cine-MRI

    Cetin I, Sanroma G, Petersen SE, Napel S, Camara O, Ballester Gonzalez MA, et al. A radiomics approach to computer-aided diagnosis with cardiac cine-MRI. In: Statistical Atlases and Computational Models of the Heart. ACDC and MMWHS Challenges: 8th International Workshop, STACOM 2017, Held in Conjunction with MICCAI 2017, Quebec City, Canada, September 10-...

    work page 2017

  8. [8]

    Deep cascaded registration and weakly-supervised segmentation of fetal brain MRI

    Comte V, Alenya M, Urru A, Recober J, Nakaki A, Crovetto F, et al. Deep cascaded registration and weakly-supervised segmentation of fetal brain MRI. Heliyon 2025;11(1)

    work page 2025

Show all 20 references
  1. [9]

    A deep learning framework for unsupervised affine and deformable image registration

    De Vos BD, Berendsen FF, Viergever MA, Sokooti H, Staring M, I s gum I. A deep learning framework for unsupervised affine and deformable image registration. Medical Image Analysis 2019;52:128--143

    2019

  2. [10]

    High-Resolution Maps of Left Atrial Displacements and Strains Estimated with 3D CINE MRI and Unsupervised Neural Networks

    Galazis C, Shepperd S, Brouwer E, Queir \'o s S, Alskaf E, Anjari M, et al. High-Resolution Maps of Left Atrial Displacements and Strains Estimated with 3D CINE MRI and Unsupervised Neural Networks. arXiv preprint arXiv:231209387 2023

    2023

  3. [11]

    Automatic cardiac disease assessment on cine-MRI via time-series segmentation and domain specific features

    Isensee F, Jaeger PF, Full PM, Wolf I, Engelhardt S, Maier-Hein KH. Automatic cardiac disease assessment on cine-MRI via time-series segmentation and domain specific features. In: Statistical Atlases and Computational Models of the Heart. ACDC and MMWHS Challenges: 8th Interna...

    2017

  4. [12]

    nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation

    Isensee F, Jaeger PF, Kohl S, Petersen J, Maier-Hein KH. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nature Methods 2021;18(2):203--211

    2021

  5. [13]

    Densely Connected Fully Convolutional Network for Short-Axis Cardiac Cine MR Image Segmentation and Heart Diagnosis Using Random Forest

    Khened M, Varghese A, Krishnamurthi G. Densely Connected Fully Convolutional Network for Short-Axis Cardiac Cine MR Image Segmentation and Heart Diagnosis Using Random Forest. In: STACOM@MICCAI; 2017. ://api.semanticscholar.org/CorpusID:42508148

    2017

  6. [14]

    Intelligent cardiovascular disease prediction empowered with gradient descent optimization

    Nawaz MS, Shoaib B, Ashraf MA. Intelligent cardiovascular disease prediction empowered with gradient descent optimization. Heliyon 2021;7(5)

    2021

  7. [15]

    Reference ranges for cardiac structure and function using cardiovascular magnetic resonance (CMR) in Caucasians from the UK Biobank population cohort

    Petersen SE, Aung N, Sanghvi MM, Zemrak F, Fung K, Paiva JM, et al. Reference ranges for cardiac structure and function using cardiovascular magnetic resonance (CMR) in Caucasians from the UK Biobank population cohort. Journal of Cardiovascular Magnetic Resonance 2016;19(1):18

    2016

  8. [16]

    Viscoelastic properties of normal and infarcted myocardium measured by a multifrequency shear wave method: comparison with pressure-segment length method

    Pislaru C, Urban MW, Pislaru SV, Kinnick RR, Greenleaf JF. Viscoelastic properties of normal and infarcted myocardium measured by a multifrequency shear wave method: comparison with pressure-segment length method. Ultrasound in Medicine & Biology 2014;40(8):1785--1795

    2014

  9. [17]

    Biomechanics-informed neural networks for myocardial motion tracking in MRI

    Qin C, Wang S, Chen C, Qiu H, Bai W, Rueckert D. Biomechanics-informed neural networks for myocardial motion tracking in MRI. In: Medical Image Computing and Computer Assisted Intervention--MICCAI 2020: 23rd International Conference, Lima, Peru, October 4--8, 2020, Proceedings...

    2020

  10. [18]

    Neural Ordinary Differential Equation based Sequential Image Registration for Dynamic Characterization

    Wu Y, Dong M, Jena R, Qin C, Gee JC. Neural Ordinary Differential Equation based Sequential Image Registration for Dynamic Characterization. arXiv preprint arXiv:240402106 2024

    2024

  11. [19]

    Learning correspondences of cardiac motion from images using biomechanics-informed modeling

    Zhang X, You C, Ahn S, Zhuang J, Staib L, Duncan J. Learning correspondences of cardiac motion from images using biomechanics-informed modeling. In: International Workshop on Statistical Atlases and Computational Models of the Heart Springer; 2022. p. 13--25

    2022

  12. [20]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

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