Pith. sign in

REVIEW 4 major objections 6 minor 48 references

Beyond scalar losses: calibrating segmentation models via gradient vector field surgery

T0 review · 4 major / 6 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read This paper claims that replacing the sigmoid derivative with an error-scaled vector field during backpropagation makes region-based segmentation losses yield well-calibrated probabilities while preserving Dice overlap.

desk verdict Genuinely new gradient-surgery method for calibrating Dice-type segmentation losses with large, consistent empirical gains; the main weakness is that the headline claim rests on top-k and best-of-N run reporting rather than full-run statistics. read the letter →

arxiv 2607.14338 v1 pith:ATD5GFHA submitted 2026-07-15 cs.CV cs.AIcs.LG

Laurin Lux , Alexander H. Berger , Moritz Knolle , Daniel Rückert , Johannes C. Paetzold This is my paper
classification cs.CVcs.AIcs.LG
keywords segmentationcalibrationDicelossregion-basedgradientsurgeryvectorfieldoverconfidencemedicalimaging
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

Segmentation models trained with region-based losses such as Dice are notoriously overconfident, which is a problem in medical imaging where probability maps guide decisions. The paper traces this to the gradient of Dice with respect to logits: confident predictions contribute almost nothing, and the only equilibrium between foreground and background gradients is at probabilities 0 and 1. The authors propose a "gradient surgery" that replaces the sigmoid derivative in the backward pass with a vector field whose magnitude scales linearly with prediction error |y-p|. This makes the equilibrium sit at the true foreground ratio of indistinguishable voxel groups, analogous to cross-entropy, while retaining the region-size weighting that makes Dice work on imbalanced data. Empirically, the surgery improves calibration metrics on every tested 2D and 3D medical segmentation task without sacrificing Dice score.

What carries the argument

The central object is the gradient vector field "surgery": an override of the backward pass through the sigmoid/softmax activation that replaces the standard derivative p(1-p) with a factor proportional to |y-p|, optionally multiplied by a sharp decline near 0 and 1 controlled by an exponent n. The paper proves no scalar loss function can produce these partial derivatives because mixed second partials are asymmetric, making the field non-conservative; stability is argued from small curl and from the field always pointing toward the ground truth. This surgery is what converts the Dice-style region weighting into a calibrated, error-scaled training signal.

What would settle it

Build a synthetic segmentation task with known latent groups: each group has a fixed foreground ratio r_k, and voxels within a group are truly indistinguishable to the network. Train with Dice and with the surgery, then measure the equilibrium probabilities per group. If Dice does not converge to p near 0 or 1 regardless of r_k, or if the surgery does not converge to p near r_k, the paper's mechanism is falsified.

Watch

Extended reading notes

Core claim

Region-based losses converge to miscalibrated solutions because their logit gradients vanish for confident predictions and only balance foreground versus background gradients at p=0 and p=1, forcing all probabilities to extremes. We define a logit-level vector field that scales gradient magnitude linearly with |y-p|, prove that no scalar loss can realize these partial derivatives (the field is non-conservative), and show in experiments that using this vector field as the backward pass for Dice, Tversky, and Combo losses improves calibration metrics across 2D and 3D segmentation tasks while maintaining Dice overlap.

Load-bearing premise

The root-cause story assumes that at convergence the model has pairs of pixels that look identical to the network but have different true labels, so their opposing gradients balance; if real training does not settle into that equilibrium picture, the diagnosis of Dice overconfidence does not follow even though the proposed fix may still work.

Editorial extensions

If this is right

  • Applied to Dice, Tversky, and Combo losses, the surgery improves NLL, ECE, MCE, and Brier on all four tested medical segmentation benchmarks, often cutting calibration error by factors of 4 to 6.
  • Dice overlap is preserved or improved: on the 2D mammography mass dataset it rose from about 65% (Dice) to 74% (Dice + surgery), while on the other datasets it stayed within about a point.
  • Combining the surgery with a CE+Dice Combo loss gives the best or second-best calibration scores on most benchmarks, making it a drop-in replacement for plain region losses.
  • The intervention sharply reduces logit magnitudes—on the retinal vessel dataset, average foreground logits fell from roughly 26 to 3.7 with Dice—consistent with removing pathological overconfidence.
  • The method also improves calibration when used with a transformer-based segmentation backbone, suggesting it is not tied to one architecture.

Reading between the lines

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

  • If the equilibrium diagnosis is correct, the same vector-field surgery should extend to multi-class soft Dice and other region-based losses (e.g. Unified Focal) by generalizing the one-vs-rest error scaling; the paper only demonstrates binary segmentation, so that is a testable next step.
  • The non-existence of a scalar loss means any fully calibrated region-based objective must be non-conservative; this implies optimizer choice and learning-rate schedules could matter more than for standard losses, and formal stability bounds would be a natural follow-up.
  • The sharp-decline term behaves like a soft label-smoothing analog for region losses; an explicit comparison against spatially varying label smoothing on boundary voxels would isolate whether the gain comes from the error scaling or the decline.
  • Because the surgery reduces logit magnitudes, it may interact with post-hoc temperature scaling; a testable combination is whether temperature scaling adds anything after surgery, and whether the resulting model is more robust to domain shift than post-hoc-only calibration.
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 / 6 minor

Summary. The paper addresses miscalibration of segmentation models trained with region-based losses (Dice, Tversky, Combo). The authors analyze the gradient of the Dice loss w.r.t. logits and argue that the sigmoid derivative p(1-p) suppresses gradients for confident errors, causing overconfident predictions. They propose a 'gradient surgery' that replaces the sigmoid derivative in the backward pass with an error-scaled factor (proportional to |y-p|), while retaining the region-imbalance-aware global term of the Dice loss. They show (Appendix A) that no scalar loss can produce the desired partial derivatives, so they define a non-conservative vector field and use it directly as the optimization objective. Experiments on 2D (INbreast, FIVES) and 3D (BraTS-METS, KiTS) datasets report improved calibration (NLL, ECE, MCE, Brier) relative to baseline region-based losses, with DSC largely preserved. The central claim is that the surgery 'improves calibration metrics compared to the respective baseline losses alone across all cases in all datasets' (Section 4.1).

Significance. If the empirical results hold across full run distributions, the proposed method is a simple and attractive plug-in for training calibrated segmentation models with region-based losses, addressing a clinically relevant problem. The paper has several strengths: it explicitly considers gradient geometry, provides an implementation listing, includes active-region calibration checks (Appendix H), an ablation of the exponential parameter n, logit-magnitude analysis, and a supplementary transformer experiment. However, the evidence is weakened by selective run reporting (top-5-of-25 in 2D, best-of-10 in 3D) and by a theoretical analysis that rests on an unproved equilibrium assumption. The paper's practical contribution could be significant for medical imaging, but the validity of the headline claim cannot be assessed from the current reported numbers.

major comments (4)
  1. [Section 4.1 / Tables 1-2] The central claim 'across all cases in all datasets' is not supported by the reported statistics. Table 1 reports the average of the 5 best runs out of 25, selected by validation Dice, and Table 2 reports a single best run out of 10. Top-k selection biases the comparison in favor of the intervention and removes variance information needed for significance. Please report full-run statistics (e.g., mean/median with confidence intervals, or per-run scatter plots) for all 25/10 runs, or show pairwise comparisons in which the surgery improves over the baseline for a preponderance of runs. Without this, the claim is not established.
  2. [Appendix A] The non-existence proof contains an algebraic sign error. For y_i=y_k=0, the derivative of the global term with respect to p_k should be +2(2I+ε)/(P+Y+ε)^3, not the negative value printed (the derivative of N_i/D is [2(y_i−y_k) − 2N_i/(P+Y+ε)]/(P+Y+ε)^2). While the equality condition p_k=p_i still implies non-conservation after correcting the sign, the printed proof is incorrect. In addition, the proof applies to Eq. (9), not to the actual implemented field of Eq. (10) with the exponential decline terms; this gap should be addressed.
  3. [Section 3.1] The 'indistinguishable voxels' equilibrium argument is asserted, not derived. The existence of sets of voxels with identical latent representations and coupled probabilities, and the reduction of optimization to a gradient-balance equilibrium, are assumptions. Since the paper claims a theoretical diagnosis of Dice overconfidence, this argument needs to be made rigorous (e.g., as a formal statement with conditions) or explicitly presented as a heuristic supported by empirical evidence, such as an analysis of representation similarity in trained networks.
  4. [Section 4.1 / Table 2] The claim of 'maintaining high prediction accuracy' is not uniformly supported. On KiTS, Dice - Surgery reduces DSC from 76.62 to 74.01, a drop of 2.61 points; this is not negligible in a medical segmentation context. Please quantify the accuracy/calibration trade-off and discuss the clinical acceptability of such a decrease. The current text characterizes the impact as 'negligible' for KiTS, which appears inconsistent with the reported numbers.
minor comments (6)
  1. [Table 3 caption] The caption refers to 'TunableGradSym', a name that appears nowhere else in the manuscript; likely a leftover from an earlier version.
  2. [Eq. (10) and Listing 1] The implementation uses a scaling constant 0.25 that is absent from Eq. (10). Please state the exact vector field used in experiments and clarify whether the 0.25 is a hyperparameter or a normalization factor.
  3. [Section 3.2] Typo 'R ⋉' should likely be 'R^N' or 'R'. Also, the proof in Appendix A is referenced before the appendix is introduced.
  4. [Figure 4] The caption is difficult to parse: the two added voxels have labels y1=1 and y2=2 (typo for 0?), and the description of the remaining 100 voxels could be clearer. The axes are in probability space while the text says the field is w.r.t. logits; please clarify.
  5. [Appendix C / Figure 5] The subplot label 'Validation NLL' is duplicated, and the caption contains 'training.objective'. Also, the reference for Brier (1950) is cited as 'Brier Glenn et al.', which should be corrected to 'Brier'.
  6. [Statistical significance] The asterisks in Table 1 are computed on the selected 5 best runs. Even as a secondary analysis, the selection invalidates the significance test; please compute significance on all runs or describe the selection-correction procedure.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the error-scaled gradient is a deliberate design choice, not a fitted prediction; the main claims rest on held-out test metrics.

full rationale

The theoretical part (Section 3.1) is derived from the actual partial derivatives of Dice and CE; the claim that Dice equilibria lie at 0/1 while CE equilibria lie at p=r does not presuppose the conclusion. The proposed vector field (Eqs. 5-10) is admittedly constructed so that gradient magnitude scales with |y-p|, so observing improved calibration is partly a consequence of the intervention's design. However, that is not circularity of the prohibited kind: no test calibration metric is fitted and then relabeled as a prediction; calibration is measured on held-out voxels, and DSC preservation together with the ablation on the exponential term provide independent downstream evidence. The paper explicitly labels its equilibrium argument as 'a simplified argument' (Section 3.1), which is a robustness caveat rather than a circular step. The only self-citation with overlapping authorship (Berger et al. 2025, used in Appendix I to exclude very small tumors as label noise) is peripheral and not load-bearing for the central claim. The reported top-5/best-of-10 run selection is a statistical reporting weakness, not circularity: it does not make the empirical outcome equal to the method's input by construction.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central contribution rests on a region-based gradient structure, a heuristic equilibrium model, and the assumption that a purpose-built non-conservative vector field trains stably. The exponential n and the 0.25 factor are the main free constants; no new physical entities are introduced.

free parameters (3)
  • exponential decline n = n~20 (best in FIVES ablation; value used in main runs not explicitly stated)
    Eq. (10) adds (1-(1-p)^n)(1-p^n); n is tuned in Section 4.2 to balance calibration and DSC, not derived.
  • gradient scale constant 0.25 = 0.25
    Listing 1 sets error_weight = 0.25 * error; the 0.25 factor is unexplained and does not appear in Eq. (10).
  • method hyperparameters (Tversky alpha, SVLS sigma, NACL lambda, optimizer/weight decay/learning rate) = not fully reported
    A random hyperparameter search is run per setup, but the optimal configurations are not fully tabulated, making exact replication harder.
assumptions (5)
  • domain assumption Binary segmentation with a final sigmoid/softmax activation; the analysis is carried out for y in {0,1}.
    Section 3 starts with 'We assume a binary segmentation problem using a final sigmoid activation'; the multi-class implementation in Appendix D assumes a softmax with binary foreground split.
  • domain assumption A single voxel's contribution to the global sums P, Y, I is negligible, so the region-based term G(i) may be treated as a constant G.
    Used after Eq. (2): 'a single voxel has negligible influence on G as the image size N increases'; the equilibrium picture in Figure 1 depends on this approximation.
  • domain assumption At convergence the network contains sets of voxels with shared latent representations that force coupled probabilities; calibration is set by gradient-magnitude equilibrium on these sets.
    Section 3.1: 'we make a simplified argument by considering... voxels a and b... indistinguishable through the network's latent representation'; no proof that real trained networks realize this regime.
  • domain assumption The chosen non-conservative vector field is safe to optimize with SGD/Adam because its curl is 'negligible compared to the diagonal terms'.
    Section 3.3 asserts this with a reference to GAN numerics but provides no bound; stability and convergence of the non-scalar surrogate are assumed.
  • standard math Smoothness and sign-definiteness of F imply favorable training behavior (no sign flips, continuity on R^N).
    Section 3.3 lists these as theoretical properties; they are standard calculus facts but fall short of training guarantees.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond scalar losses: calibrating segmentation models via gradient vector field surgery." pith.science (2026). https://pith.science/paper/ATD5GFHA

@misc{pith2026260714338,
  author       = {Pith},
  title        = {Pith review of: Beyond scalar losses: calibrating segmentation models via gradient vector field surgery},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ATD5GFHA}},
  note         = {Machine review of arXiv:2607.14338}
}
read the original abstract

Region-based loss functions, such as the Dice loss, have established themselves as the de facto standard for highly class- and region-imbalanced segmentation tasks. However, models trained using region-based loss functions are notoriously miscalibrated and typically yield over-confident predictions. In medical imaging applications, such as defining tumor resection margins, this miscalibration is hindering clinical adoption. In this work, we outline a novel gradient perspective on this overconfidence and show how it affects region-based loss functions. We propose a "surgery" on the gradient vector field as a simple, yet effective intervention to mitigate calibration issues. This surgery adds a factor to the loss's partial derivative, scaling the gradient's magnitude linearly with the prediction error. In empirical evaluations across 2D and 3D medical segmentation tasks, we demonstrate the effectiveness of this intervention while maintaining high prediction accuracy when used in conjunction with any region-based loss function.

Figures

Figures reproduced from arXiv: 2607.14338 by the authors.

Figure 1
Figure 1. Partial derivatives of dice loss (top row) vs. cross entropy loss (bottom row) for single voxels. Sub-panels show the absolute value of: δL δpi , δp δzi , and δL δzi as a function of the predicted probability p for a foreground (y = 1, yellow) and a background (y = 0, purple) voxel. Red squares indicate intersection points where the magnitude of foreground and background derivatives is in equilibrium. For cross-entr… view at source ↗
Figure 2
Figure 2. Visualization of the gradient w.r.t. the voxel-wise logits. Purple circles indicate confident errors, where gradients vanish through the activation function for Dice. towards convergence without reaching zero loss. In this scenario, a and b influence the gradient w.r.t. the network parameters in opposite directions through the opposing ground truth labels for these ”indistinguishable” voxels. For ya = 1, the network… view at source ↗
Figure 3
Figure 3. Visualization of the predicted probability maps as heat maps on the BraTS dataset (best viewed zoomed in). Yellow and red indicate medium and high foreground probability, respectively. Blue arrows indicate regions of overconfi￾dence of the Dice model, while our approach exhibits well-calibrated predictions. The Dice model overconfidently predicts background (A, both arrows and B, right arrow) and foreground (B, left… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Gradient/vector fields w.r.t. logits of different loss functions and our proposed vector field. For better visualization, the axes are displayed as probabilities p instead of logits z. 3414 [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]
Figure 5
Figure 5. Figure 5: shows training and validation curves with different optimizers. Training soft dice / dice loss Dice Loss + SGD Dice Loss + Adam Dice Loss - Surgery + SGD Dice Loss - Surgery + Adam Validation dice score Validation NLL Validation NLL [PITH_FULL_IMAGE:figures/full_fig_p…
Figure 6
Figure 6. Figure 6: Visualization of the normalized partial derivatives derived from different loss func￾tions. 3418 [PITH_FULL_IMAGE:figures/full_fig_p022_6.png]
Figure 7
Figure 7. Figure 7: Comparison of the probability maps for two samples from a Dice loss trained model and our adapted Dice vector field approach. The overconfidence phe￾nomenon on the Dice probabilities is apparent, with probabilities almost appear￾ing binarized. In comparison, our method…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

48 extracted references · 5 linked inside Pith

  1. [1]

    1986 , isbn =

    Lamport, Leslie , title =. 1986 , isbn =

    1986

  2. [2]

    Distilling the knowledge in a neural network , author=

  3. [3]

    Journal of Digital Imaging , volume=

    Calibrating the dice loss to handle neural network overconfidence for biomedical image segmentation , author=. Journal of Digital Imaging , volume=. 2023 , publisher=

    2023

  4. [4]

    Sensors , volume=

    An analysis of loss functions for heavily imbalanced lesion segmentation , author=. Sensors , volume=. 2024 , publisher=

    2024

  5. [5]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    Average calibration error: A differentiable loss for improved reliability in image segmentation , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2024 , organization=

    2024

  6. [6]

    arXiv preprint arXiv:2205.14094 , year=

    Failure detection in medical image classification: A reality check and benchmarking testbed , author=. arXiv preprint arXiv:2205.14094 , year=

    arXiv

  7. [7]

    International conference on machine learning , pages=

    On calibration of modern neural networks , author=. International conference on machine learning , pages=. 2017 , organization=

    2017

  8. [8]

    IEEE transactions on medical imaging , volume=

    Confidence calibration and predictive uncertainty estimation for deep medical image segmentation , author=. IEEE transactions on medical imaging , volume=. 2020 , publisher=

    2020

Show all 48 references
  1. [9]

    arXiv preprint arXiv:1811.11226 , year=

    CT organ segmentation using GPU data augmentation, unsupervised labels and IOU loss , author=. arXiv preprint arXiv:1811.11226 , year=

    arXiv

  2. [10]

    2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018) , pages=

    Minimal annotation training for segmentation of microscopy images , author=. 2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018) , pages=. 2018 , organization=

    2018

  3. [11]

    Computerized Medical Imaging and Graphics , volume=

    Combo loss: Handling input and output imbalance in multi-organ segmentation , author=. Computerized Medical Imaging and Graphics , volume=. 2019 , publisher=

    2019

  4. [12]

    2016 fourth international conference on 3D vision (3DV) , pages=

    V-net: Fully convolutional neural networks for volumetric medical image segmentation , author=. 2016 fourth international conference on 3D vision (3DV) , pages=. 2016 , organization=

    2016

  5. [13]

    Journal of the American statistical Association , volume=

    Strictly proper scoring rules, prediction, and estimation , author=. Journal of the American statistical Association , volume=. 2007 , publisher=

    2007

  6. [14]

    Computerized Medical Imaging and Graphics , volume=

    Unified focal loss: Generalising dice and cross entropy-based losses to handle class imbalanced medical image segmentation , author=. Computerized Medical Imaging and Graphics , volume=. 2022 , publisher=

    2022

  7. [15]

    International workshop on machine learning in medical imaging , pages=

    Tversky loss function for image segmentation using 3D fully convolutional deep networks , author=. International workshop on machine learning in medical imaging , pages=. 2017 , organization=

    2017

  8. [16]

    Nature methods , volume=

    nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation , author=. Nature methods , volume=. 2021 , publisher=

    2021

  9. [17]

    Scientific data , volume=

    Fives: A fundus image dataset for artificial intelligence based vessel segmentation , author=. Scientific data , volume=. 2022 , publisher=

    2022

  10. [18]

    arXiv preprint arXiv:2504.12527 , year=

    Analysis of the MICCAI Brain Tumor Segmentation--Metastases (BraTS-METS) 2025 Lighthouse Challenge: Brain Metastasis Segmentation on Pre-and Post-treatment MRI , author=. arXiv preprint arXiv:2504.12527 , year=

    2025

  11. [19]

    Academic radiology , volume=

    Inbreast: toward a full-field digital mammographic database , author=. Academic radiology , volume=. 2012 , publisher=

    2012

  12. [20]

    arXiv preprint arXiv:1904.00445 , year=

    The kits19 challenge data: 300 kidney tumor cases with clinical context, ct semantic segmentations, and surgical outcomes , author=. arXiv preprint arXiv:1904.00445 , year=

    1904 arXiv

  13. [21]

    Advances in neural information processing systems , volume=

    The numerics of gans , author=. Advances in neural information processing systems , volume=

  14. [22]

    International Conference on Medical image computing and computer-assisted intervention , pages=

    U-net: Convolutional networks for biomedical image segmentation , author=. International Conference on Medical image computing and computer-assisted intervention , pages=. 2015 , organization=

    2015

  15. [23]

    arXiv preprint arXiv:2109.00903 , year=

    Effect of the output activation function on the probabilities and errors in medical image segmentation , author=. arXiv preprint arXiv:2109.00903 , year=

    arXiv

  16. [24]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Deep residual learning for image recognition , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  17. [25]

    Monthly weather review , volume=

    Verification of forecasts expressed in terms of probability , author=. Monthly weather review , volume=. 1950 , publisher=

    1950

  18. [26]

    international conference on machine learning , pages=

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning , author=. international conference on machine learning , pages=. 2016 , organization=

    2016

  19. [27]

    Advances in neural information processing systems , volume=

    Simple and scalable predictive uncertainty estimation using deep ensembles , author=. Advances in neural information processing systems , volume=

  20. [28]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Obtaining well calibrated probabilities using bayesian binning , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  21. [29]

    2019 IEEE 16th international symposium on biomedical imaging (ISBI 2019) , pages=

    A novel focal tversky loss function with improved attention u-net for lesion segmentation , author=. 2019 IEEE 16th international symposium on biomedical imaging (ISBI 2019) , pages=. 2019 , organization=

    2019

  22. [30]

    Advances in large margin classifiers , volume=

    Probabilistic outputs for support vector machines and comparisons to regularized likelihood methods , author=. Advances in large margin classifiers , volume=. 1999 , publisher=

    1999

  23. [31]

    Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining , pages=

    Transforming classifier scores into accurate multiclass probability estimates , author=. Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining , pages=

  24. [32]

    2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI) , pages=

    Post training uncertainty calibration of deep networks for medical image segmentation , author=. 2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI) , pages=. 2021 , organization=

    2021

  25. [33]

    Failure Detection in Medical Image Classification: A Reality Check and Benchmarking Testbed , author=

  26. [34]

    Medical Image Analysis , volume=

    Theoretical analysis and experimental validation of volume bias of soft dice optimized segmentation maps in the context of inherent uncertainty , author=. Medical Image Analysis , volume=. 2021 , publisher=

    2021

  27. [35]

    Medical imaging 2019: image Processing , volume=

    Towards increased trustworthiness of deep learning segmentation methods on cardiac MRI , author=. Medical imaging 2019: image Processing , volume=. 2019 , organization=

    2019

  28. [36]

    International MICCAI Brainlesion Workshop , pages=

    Optimization with soft dice can lead to a volumetric bias , author=. International MICCAI Brainlesion Workshop , pages=. 2019 , organization=

    2019

  29. [37]

    Journal of the American Medical Informatics Association , volume=

    Calibrating predictive model estimates to support personalized medicine , author=. Journal of the American Medical Informatics Association , volume=. 2012 , publisher=

    2012

  30. [38]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Rethinking the inception architecture for computer vision , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  31. [39]

    Advances in neural information processing systems , volume=

    When does label smoothing help? , author=. Advances in neural information processing systems , volume=

  32. [40]

    Medical Image Analysis , volume=

    Neighbor-aware calibration of segmentation networks with penalty-based constraints , author=. Medical Image Analysis , volume=. 2025 , publisher=

    2025

  33. [41]

    international conference on information processing in medical imaging , pages=

    Spatially varying label smoothing: Capturing uncertainty from expert annotations , author=. international conference on information processing in medical imaging , pages=. 2021 , organization=

    2021

  34. [42]

    International conference on medical image computing and computer-assisted intervention , pages=

    Boundary-weighted logit consistency improves calibration of segmentation networks , author=. International conference on medical image computing and computer-assisted intervention , pages=. 2023 , organization=

    2023

  35. [43]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    The devil is in the margin: Margin-based label smoothing for network calibration , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  36. [44]

    Medical Image Analysis , volume=

    Calibrating segmentation networks with margin-based label smoothing , author=. Medical Image Analysis , volume=. 2023 , publisher=

    2023

  37. [45]

    International Conference on Information Processing in Medical Imaging , pages=

    Pitfalls of topology-aware image segmentation , author=. International Conference on Information Processing in Medical Imaging , pages=. 2025 , organization=

    2025

  38. [46]

    arXiv preprint arXiv:2503.01835 , year=

    Primus: Enforcing attention usage for 3d medical image segmentation , author=. arXiv preprint arXiv:2503.01835 , year=

    arXiv

  39. [47]

    International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=

    nnu-net revisited: A call for rigorous validation in 3d medical image segmentation , author=. International Conference on Medical Image Computing and Computer-Assisted Intervention , pages=. 2024 , organization=

    2024

  40. [48]

    International conference on medical image computing and computer-assisted intervention , pages=

    Trust your neighbours: Penalty-based constraints for model calibration , author=. International conference on medical image computing and computer-assisted intervention , pages=. 2023 , organization=

    2023

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.