Pith. sign in

REVIEW 4 major objections 6 minor 30 references

ContextLoss: Context Information for Topology-Preserving Segmentation

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read ContextLoss improves topological correctness by putting the whole context of each missed connection into the loss, repairing more gaps than skeleton-only masks.

desk verdict ContextLoss is a genuinely useful incremental loss with a new dataset and metrics, but the evaluation protocol is asymmetric and lacks error bars, so the headline claim about the context mask isn't yet solid. read the letter →

arxiv 2506.11134 v1 pith:INLQL2OQ submitted 2025-06-10 cs.CV eess.IV

Benedict Schacht , Imke Greving , Simone Frintrop , Berit Zeller-Plumhoff , Christian Wilms This is my paper
classification cs.CVeess.IV
keywords topology-preservingsegmentationlossfunctioncriticalpixelmaskskeletonizationdistancetransformmissedconnectionse0-GtAGS
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 proposes ContextLoss (CLoss), a training loss that helps segmentation networks repair topological errors such as broken vessels, roads, or bone cement lines. Its central claim is that the critical pixel mask should include the entire context of a missed or false connection, not just single critical pixels or the whole skeleton. To verify improved gap closing, the authors introduce two metrics, e0-Gt and AGS, that isolate connectivity and resist evaluation artifacts. On four datasets, including a new 3D nano-imaging bone cement line dataset, CLoss improves topology-aware metrics and closes up to 44% more missed connections than clDice on the HRF-Retina dataset.

What carries the argument

The load-bearing mechanism is the context-extended critical pixel mask. Skeletonization identifies topologically critical locations, and a distance transform extends the error part of the skeleton to all nearby pixels belonging to the label (for gaps) or prediction (for false positives), so the loss weighs the whole neighborhood of each topological error. Two complementary metrics, $e_0$-$Gt$ and AGS, are also part of the machinery: $e_0$-$Gt$ masks predictions with the ground truth before counting connected components, and AGS measures prediction accuracy on the ground-truth skeleton, making both less sensitive to cropping and contour artifacts than standard $e_0$ and clDice.

What would settle it

Run multiple training seeds for CLoss and compound clDice on HRF-Retina with all hyperparameters selected by the same criterion, for instance best e0-Gt or best combined Betti error, and compute e0-Gt and AGS with confidence intervals; if CLoss no longer closes more missed connections than compound clDice, the paper's attribution of the improvement to the context mask is refuted.

Watch

Extended reading notes

Core claim

The central claim is that context around topological errors, extracted by combining skeletonization with distance transforms, is what makes a topology-preserving loss effective. CLoss splits the label skeleton into the part at the topological error and the correctly predicted part, then uses the distance transform to keep only pixels closest to the error skeleton, cropped to label size; false-positive connections are handled symmetrically. This context-rich critical pixel mask $M$ enters the loss $L_{\text{context}} = (1-\gamma)L_{\text{pixel}} + \gamma L_{\text{pixel}} \odot M$. The paper argues that this mask, not the choice of pixel-wise loss, drives the improvement, supported by comparisons against compound clDice, which differs only in the mask. Training is staged: pretrain with a standard pixel-wise loss, fine-tune with CLoss, and post-process by keeping only fine-tuned foreground components that overlap pretraining predictions.

Load-bearing premise

The evaluation protocol is assumed to be reliable: baselines are tuned by their best clDice score while CLoss is reported at fixed gamma values, no error bars are given, and the headline 44% improvement is measured with the authors' own e0-Gt metric on a single dataset.

Editorial extensions

If this is right

  • Training with CLoss closes more gaps in elongated structures such as vessels, roads, and bone cement lines than training with clDice or Dice & Cross-Entropy.
  • CLoss works on 2D and 3D data and can be plugged into any segmentation network because it changes only the loss function.
  • The proposed topological post-processing removes small separate noise components without introducing new false-positive connections.
  • The metrics e0-Gt and AGS give a less artifact-prone reading of connectivity than e0 and clDice, which may change how topology-aware methods are compared.
  • CLoss runtime per epoch is comparable to clDice, so the topological gains do not require the high runtime of persistent-homology-based masks.

Reading between the lines

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

  • The context-mask design is independent of the pixel-wise loss and skeletonization routine, so it could be combined with adaptive or learned skeleton extraction to handle noisy 3D surfaces.
  • The reported comparison assumes a particular selection rule for baselines and fixed gamma values for CLoss; readers should check whether CLoss retains its margin when all methods, including CLoss, are tuned by the same criterion.
  • The proposed metrics, particularly e0-Gt, could serve as inexpensive connectivity checks for any segmentation pipeline, not only for training.
  • The thickening effect of CLoss suggests that combining it with a boundary- or thickness-aware term might recover Dice while keeping the topology gains.
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 proposes ContextLoss (CLoss), a topology-preserving segmentation loss whose critical pixel mask includes the context around topological errors (missed and false-positive connections), obtained by combining skeletonization with distance transforms. The authors also introduce two new metrics (e0-Gt and AGS) intended to measure missed-connection reparation, and present a new 3D bone cement-line dataset (CLD). CLoss is evaluated on Roads, HRF-Retina, Vessap, and CLD against nnU-Net, Dice+CE, clDice, and a newly introduced compound clDice that shares CLoss's pixel-wise loss. The central design claim is that the context mask, not the pixel-wise loss, drives the reported topological improvements.

Significance. The context-mask idea is a reasonable and potentially useful contribution to topology-preserving segmentation, and the new CLD dataset plus the two proposed metrics will likely interest the community. The paper ships public code, uses a standard nnU-Net protocol, and includes an ablation (Table 4) that isolates the critical pixel mask, which are strengths. However, the current evidence for the central claim is weakened by an asymmetric hyperparameter-selection protocol, by reliance on the authors' own metrics for the headline result, and by the absence of variance or significance information. If the evaluation is made symmetric and statistically grounded, the contribution would be solid; in its present form the main design claim is not yet established beyond reasonable doubt.

major comments (4)
  1. [Sec. A.6 and Table 1] The comparison between CLoss and compound clDice is asymmetric. Compound clDice is reported after a weight search over 0.1 to 0.5 in 0.1 steps, selecting the result with the highest clDice score, while CLoss is reported at fixed gamma values 0.08, 0.1, and 0.2 with no stated selection rule. On HRF-Retina the headline e0-Gt improvement (CLoss 1.378 at gamma=0.2 versus compound clDice 2.429) shrinks to about 5% at gamma=0.08 (2.347 versus 2.429), and at gamma=0.2 CLoss has a worse standard e0 (0.172 versus 0.155) and a lower Dice (80.57 versus 82.33). On Vessap the clDice score at gamma=0.2 is 93.13 versus 95.80 for compound clDice, and on Roads e0 at gamma=0.2 is 0.205 versus 0.199. The conclusion in Sec. 4.1 that CLoss's mask is superior to the full-skeleton mask therefore depends on which gamma is chosen and which metric is used to select it. The authors should report CLoss under the same selection rule as compound clDice, or compare all methods over a shared gamma grid without per-method selection.
  2. [Sec. 4 and Sec. A.5] The paper reports five-fold cross-validation for all methods but gives no per-fold variances, confidence intervals, or paired statistical tests. Several reported differences are small enough to be plausibly within cross-validation noise (for example, Roads e0 0.185 for CLoss gamma=0.08 versus 0.195 for compound clDice; CLD clDice 86.22 for CLoss gamma=0.08 versus 86.22 for clDice). Without a measure of variability, the claims of consistent superiority in Sec. 4.1 and the conclusion are not verifiable. At minimum, per-fold metric distributions or standard deviations should be reported, and a paired test (e.g., Wilcoxon signed-rank over folds) should be provided for the main comparisons.
  3. [Sec. 2.2 and Sec. 4.1] The headline claim of repairing up to 44% more missed connections rests on e0-Gt, one of the two metrics proposed by the authors, evaluated on a single dataset (HRF-Retina). The standard topology metrics do not consistently support the claim: on HRF-Retina CLoss at gamma=0.2 has e0=0.172, which is worse than compound clDice's 0.155, and on Vessap CLoss at gamma=0.2 has a clDice score 93.13 versus 95.80 for compound clDice. Since e0-Gt and AGS both measure accuracy on or relative to the ground-truth skeleton, they can be improved by thicker predictions, and the reported Dice drop at higher gamma is consistent with such thickening. The paper should either present the 44% claim alongside the standard e0, e1, and clDice results at the same operating point, or explicitly justify why the self-proposed metrics are the primary evidence for the claim; the current presentation overstates the strength of the evidence.
  4. [Sec. 4.1] The sentence 'CLoss consistently outperforms compound clDice on the topology metrics' is not supported by Table 1. On Vessap the clDice score is worse for CLoss at gamma=0.2 (93.13 versus 95.80) and on Roads e0 at gamma=0.2 is worse (0.205 versus 0.199). The discussion in Sec. A.9 acknowledges the Vessap clDice discrepancy by attributing it to skeletonization artifacts, but this attribution is not empirically demonstrated. The claim of consistency should be qualified to specific metrics, datasets, and gamma values, or the conclusion should be rewritten to match the actual table.
minor comments (6)
  1. [Abstract] The phrase 'recently proposed solutions are loss functions based on critical pixel masks that consider the whole skeleton of the segmented structures in the critical pixel mask' is hard to parse; consider simplifying to 'recent loss functions use critical pixel masks that include the entire skeleton of the segmented structures'.
  2. [Sec. 2.2, Eq. (3)] The notation beta_0^{Xbin⊙Y} is not defined before use in Equation (3); please state that it denotes the number of connected components of the binary prediction masked by the ground truth.
  3. [Sec. 2.2, Eq. (4)] The definition of AGS as sum(Xbin ⊙ S_Y)/sum(S_Y) is clear, but the text should explicitly note that AGS is not a skeleton-based metric and therefore is expected to be insensitive to prediction-skeleton artifacts; this is currently only implied by Figure 4.
  4. [Sec. 3] The new CLD dataset is a significant asset, but the main text gives only its shape and voxel size. A brief statement of the inter-annotator agreement or annotation QA procedure appears only in the appendix; consider moving a one-sentence version into the main text.
  5. [Sec. 4.1] The sentence 'CLoss repairs 44% more missed connections than clDice (e0-Gt)' should state the gamma value and the comparison method (clDice vs compound clDice) in the same sentence, since Table 1 shows different values for these two baselines.
  6. [Sec. A.1, Algorithm 1] The pseudo-code for the critical pixel mask uses the soft-skeleton notation S_X and S_Y, but the text of Sec. 2.1 describes the mask on binary predictions; please clarify how the soft skeleton is binarized or used in the distance-transform formulas.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: CLoss is a hand-designed loss and the proposed metrics are not the training objective.

full rationale

CLoss is introduced as a hand-designed loss function (Eq. 2) combining a pixel-wise loss with the same loss applied to a critical pixel mask; the mask is computed by skeletonization, splitting, and distance transforms on the label and prediction (Sec. 2.1, Alg. 1). There is no fitted parameter that is later reported as a prediction, and no uniqueness theorem or ansatz is imported from the authors' own prior work to force the design. The two proposed metrics, e0-Gt and AGS (Eqs. 3 and 4), are defined from the binary prediction and ground-truth skeleton; they are used as evaluation criteria, not as the loss being optimized, so reporting them as evidence does not reduce to a definitional loop. The comparison against compound clDice contains an evaluation asymmetry (compound clDice is selected by best clDice in Sec. A.6 while CLoss is reported at fixed gamma values, and the headline 44% figure uses the authors' own e0-Gt on one dataset); this is a statistical and fairness concern, not a circularity reduction. Self-citations appear only for imaging and data-acquisition context (Refs. 17 and 18) and are not load-bearing for the central loss comparison. Hence no specific circular step can be exhibited, and the derivation chain is self-contained.

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

The central claim rests on a hand-designed loss with several hyperparameters (gamma, alpha, fine-tuning length) and on evaluation choices (patch sizes, two new metrics). The main assumptions are that skeleton/distance-transform context extraction identifies the right pixels, that the new metrics measure gap closing fairly, that the CLD annotations are reliable, and that the evaluation protocol does not favor CLoss. No new physical or conceptual entities are introduced.

free parameters (4)
  • gamma (topology weight) = 0.08, 0.1, or 0.2 per dataset
    Weight of the topology-sensitive term in Eq. 2. Chosen by validation per dataset; results are reported at multiple values, so the best-gamma selection is part of the claim.
  • alpha (pixel-wise loss weight) = 0.5
    Weight balancing Dice and cross-entropy in Eq. 1. Taken from prior work rather than fitted, but it is a hand-set hyperparameter that affects all methods equally.
  • Fine-tuning length = 50 epochs (5% of pretraining)
    Selected after ablations in Sec. A.10; the authors note there is no consistent trend between 50 and 100 epochs.
  • Evaluation patch sizes = 375^2, 292^2, full volume, 1024^2 x 64
    Betti metrics and e0-Gt are computed on author-chosen patch sizes (Sec. A.7), which changes the numerical values and can influence ranking.
assumptions (4)
  • domain assumption Skeletonization and distance transforms on binarized labels and predictions faithfully localize all topological errors and their context.
    The critical pixel mask in Sec. 2.1 and Fig. 3 depends entirely on this; no proof is given that the operations capture all topologically relevant pixels or exclude irrelevant ones.
  • domain assumption e0-Gt and AGS are valid indicators of gap-closing quality and do not systematically favor overprediction.
    e0-Gt masks predictions by GT and AGS measures recall on the GT skeleton, so both ignore false positives outside the GT and can reward overprediction; the 44% claim rests on e0-Gt (Sec. 2.2).
  • domain assumption Manual annotation of CLD is accurate enough to serve as ground truth for topology evaluation.
    CLD is a new benchmark with low contrast, diffuse borders, and interpolation-based annotation (Sec. A.4). If annotations contain breaks or merges, all topology metrics on CLD are affected.
  • domain assumption Topological post-processing does not remove genuine gap-closing repairs.
    Post-processing keeps only fine-tuning components connected to pretraining predictions (Eq. 5). If a newly closed gap produced a separate component, it would be discarded and the loss's benefit understated; the paper applies it to all methods.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ContextLoss: Context Information for Topology-Preserving Segmentation." pith.science (2026). https://pith.science/paper/INLQL2OQ

@misc{pith2026250611134,
  author       = {Pith},
  title        = {Pith review of: ContextLoss: Context Information for Topology-Preserving Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/INLQL2OQ}},
  note         = {Machine review of arXiv:2506.11134}
}
read the original abstract

In image segmentation, preserving the topology of segmented structures like vessels, membranes, or roads is crucial. For instance, topological errors on road networks can significantly impact navigation. Recently proposed solutions are loss functions based on critical pixel masks that consider the whole skeleton of the segmented structures in the critical pixel mask. We propose the novel loss function ContextLoss (CLoss) that improves topological correctness by considering topological errors with their whole context in the critical pixel mask. The additional context improves the network focus on the topological errors. Further, we propose two intuitive metrics to verify improved connectivity due to a closing of missed connections. We benchmark our proposed CLoss on three public datasets (2D & 3D) and our own 3D nano-imaging dataset of bone cement lines. Training with our proposed CLoss increases performance on topology-aware metrics and repairs up to 44% more missed connections than other state-of-the-art methods. We make the code publicly available.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

30 extracted references · 29 canonical work pages

  1. [1]

    ContextLoss: Context Information for Topology-Preserving Segmentation

    INTRODUCTION Loss functions are a core component of neural networks, which determine what the network is optimized for. In im- age segmentation, this translates directly to how well the predictions of the network can resemble the structural prop- erties of the ground truth. The two most commonly used loss functions in image segmentation are the pixel-wise...

    work page Pith review arXiv

  2. [2]

    ContextLoss We propose the novel loss function CLoss to promote topology- preserving segmentation with arbitrary segmentation net- works

    METHODOLOGY 2.1. ContextLoss We propose the novel loss function CLoss to promote topology- preserving segmentation with arbitrary segmentation net- works. CLoss extracts a critical pixel mask of topologically important locations, which is then considered in the total cost function. In addition, we introduce a pretraining and topological fine-tuning strate...

    work page

  3. [3]

    4: Evaluation artifacts

    Label 2) Prediction 3) False positive prediction Patch evaluation mask border Fig. 4: Evaluation artifacts. 1) has the same topology (e 0) as 2). However, the higher thickness of 2), compared to 1), leads to a cropping artifact in the patch-based evaluation (red area), which falsely suggests an unfavorable e 0. Our e 0-Gt is robust to this artifact from p...

    work page

  4. [4]

    Their topology is a key property for their function [15, 16], there- fore automated topology-preserving segmentation creates a Table 1: Quantitative results

    CEMENT LINE DATASET (CLD) Cement lines form and separate structural units in the bone, and are thought to play an essential role in crack propaga- tion, energy absorption, and bone mineralization [15]. Their topology is a key property for their function [15, 16], there- fore automated topology-preserving segmentation creates a Table 1: Quantitative result...

    work page

  5. [5]

    Following other work on topology-preserving segmentation [3–8, 13], we use the two public 2D datasets, Massachusetts Roads (Roads) [19] and HRF-Retina [20]

    EXPERIMENTS Datasets. Following other work on topology-preserving segmentation [3–8, 13], we use the two public 2D datasets, Massachusetts Roads (Roads) [19] and HRF-Retina [20]. Additionally, we evaluate on two 3D datasets, the public Vessap [21] dataset, and our CLD (Sec. 3). More details on the training splits are included in the supplement (Sec. A.5)....

    work page

  6. [6]

    We apply our proposed topological post-processing (Sec

    (50 iterations) is used for all corresponding methods. We apply our proposed topological post-processing (Sec. 2.1) to all fine-tuning results to highlight the impact of our proposed loss function. Ablations for our proposed topological post- processing are provided in the supplement (Sec. A.10). We perform five-fold cross-validations for all methods. Fur...

    work page

  7. [7]

    CLoss is based on a critical pixel mask, which considers the whole context of topological errors

    CONCLUSION In this work, we propose the novel topology-preserving loss function CLoss. CLoss is based on a critical pixel mask, which considers the whole context of topological errors. We implement CLoss with a dedicated training strategy that allows for topological post-processing. CLoss can be used on 2D and 3D datasets with any arbitrary segmentation n...

    work page

  8. [8]

    Loss odyssey in medical image seg- mentation,

    J. Ma, J. Chen, M. Ng, R. Huang, Y . Li, C. Li, X. Yang, and A. L. Martel, “Loss odyssey in medical image seg- mentation,” MedIA, vol. 71, 2021

    work page 2021

Show all 30 references
  1. [9]

    Do we really need dice? the hidden region-size biases of segmentation losses,

    B. Liu, J. Dolz, A. Galdran, R. Kobbi, and I. Ben Ayed, “Do we really need dice? the hidden region-size biases of segmentation losses,” MedIA, vol. 91, 2024

    2024

  2. [10]

    clDice - a Novel Topology-Preserving Loss Function for Tubular Structure Segmentation,

    S. Shit, J. C. Paetzold, A. Sekuboyina, I. Ezhov, A. Unger, A. Zhylka, J. W. Pluim, U. Bauer, and B. H. Menze, “clDice - a Novel Topology-Preserving Loss Function for Tubular Structure Segmentation,” in CVPR, 2021

    2021

  3. [11]

    A skeletonization algorithm for gradient- based optimization,

    M. J. Menten, J. C. Paetzold, V . A. Zimmer, S. Shit, I. Ezhov, R. Holland, M. Probst, J. A. Schnabel, and D. Rueckert, “A skeletonization algorithm for gradient- based optimization,” in ICCV, 2023

    2023

  4. [12]

    Skeleton recall loss for connectivity conserving and resource efficient seg- mentation of thin tubular structures,

    Y . Kirchhoff, M. R. Rokuss, S. Roy, B. Kovacs, C. Ul- rich, T. Wald, M. Zenk, P. V ollmuth, J. Kleesiek, F. Isensee, and K. Maier-Hein, “Skeleton recall loss for connectivity conserving and resource efficient seg- mentation of thin tubular structures,” in ECCV, 2024

    2024

  5. [13]

    Topology- preserving deep image segmentation,

    X. Hu, F. Li, D. Samaras, and C. Chen, “Topology- preserving deep image segmentation,”NeurIPS, vol. 32, 2019

    2019

  6. [14]

    Structure-aware image segmentation with ho- motopy warping,

    X. Hu, “Structure-aware image segmentation with ho- motopy warping,” in NeurIPS, 2022, vol. 35

    2022

  7. [15]

    The centerline-cross entropy loss for vessel-like struc- ture segmentation: Better topology consistency without sacrificing accuracy,

    C. Acebes, A. H. Moustafa, O. Camara, and A. Galdran, “The centerline-cross entropy loss for vessel-like struc- ture segmentation: Better topology consistency without sacrificing accuracy,” in MICCAI, 2024

    2024

  8. [16]

    Embarrassingly simple scrib- ble supervision for 3d medical segmentation,

    K. Gotkowski, C. L ¨uth, P. F. J ¨ager, S. Ziegler, L. Kr¨amer, S. Denner, S. Xiao, N. Disch, K. H. Maier- Hein, and F. Isensee, “Embarrassingly simple scrib- ble supervision for 3d medical segmentation,” arXiv preprint arXiv:2403.12834, 2024

    2024 arXiv

  9. [17]

    nnU-Net: a self-configuring method for deep learning-based biomedical image segmenta- tion,

    F. Isensee, P. F. Jaeger, S. A. A. Kohl, J. Petersen, and K. H. Maier-Hein, “nnU-Net: a self-configuring method for deep learning-based biomedical image segmenta- tion,” Nature Methods, vol. 18, no. 2, 2021

    2021

  10. [18]

    Metrics reloaded: recommen- dations for image analysis validation,

    L. Maier-Hein and et al., “Metrics reloaded: recommen- dations for image analysis validation,” Nature Methods, vol. 21, no. 2, 2024

    2024

  11. [19]

    Pitfalls of topology-aware image segmentation,

    Alexander H. Berger, Laurin Lux, Alexander Weers, Martin Menten, Daniel Rueckert, and Johannes C. Paet- zold, “Pitfalls of topology-aware image segmentation,” arXiv preprint arXiv:2412.14619, 2024

    2024 arXiv

  12. [20]

    Topograph: An efficient graph-based framework for strictly topology preserving image segmentation,

    L. Lux, A. H. Berger, A. Weers, N. Stucki, D. Rueckert, U. Bauer, and J. C. Paetzold, “Topograph: An efficient graph-based framework for strictly topology preserving image segmentation,” in ICLR, 2025

    2025

  13. [21]

    Robust segmentation via topology viola- tion detection and feature synthesis,

    L. Li, Q. Ma, C. Ouyang, Z. Li, Q. Meng, W. Zhang, M. Qiao, V . Kyriakopoulou, J. V . Hajnal, D. Rueckert, and B. Kainz, “Robust segmentation via topology viola- tion detection and feature synthesis,” in MICCAI, 2023

    2023

  14. [22]

    Trabecular bone struc- tural units and their cement lines change with age, bone volume fraction, structure, and strength in female hu- man vertebrae,

    B. A. Lamarche, J. S. Thomsen, C. M. Andreasen, T. L. Andersen, and W. B. Lievers, “Trabecular bone struc- tural units and their cement lines change with age, bone volume fraction, structure, and strength in female hu- man vertebrae,” JBMR Plus, vol. 9, no. 1, 2024

    2024

  15. [23]

    Bone min- eral properties and 3D orientation of human lamellar bone around cement lines and the Haversian system,

    T. A. Gr ¨unewald, A. Johannes, N. K. Wittig, J. Palle, A. Rack, M. Burghammer, and H. Birkedal, “Bone min- eral properties and 3D orientation of human lamellar bone around cement lines and the Haversian system,” IUCrJ, vol. 10, no. 2, 2023

    2023

  16. [24]

    X-ray Zernike phase contrast tomography: 3D ROI visualization of mm-sized mice organ tissues down to sub-cellular com- ponents,

    E. Longo, L. Sancey, S. Flenner, A. Kubec, A. Bonnin, C. David, M. M ¨uller, and I. Greving, “X-ray Zernike phase contrast tomography: 3D ROI visualization of mm-sized mice organ tissues down to sub-cellular com- ponents,” Biomed. Opt. Express, vol. 11, no. 10, 2020

    2020

  17. [25]

    Multiscale morpho- logical analysis of bone microarchitecture around mg- 10gd implants,

    S. Sefa, J. Espiritu, H. ´Cwieka, I. Greving, S. Flen- ner, O. Will, S. Beuer, D.C F. Wieland, R. Willumeit- R¨omer, and B. Zeller-Plumhoff, “Multiscale morpho- logical analysis of bone microarchitecture around mg- 10gd implants,” Bioactive Materials, vol. 30, 2023

    2023

  18. [26]

    thesis, University of Toronto, 2013

    V olodymyr Mnih, Machine Learning for Aerial Image Labeling, Ph.D. thesis, University of Toronto, 2013

    2013

  19. [27]

    Robust Vessel Segmentation in Fundus Images,

    A. Budai, R. Bock, A. Maier, J. Hornegger, and G. Michelson, “Robust Vessel Segmentation in Fundus Images,” International Journal of Biomedical Imaging , vol. 2013, no. 1, 1 2013

    2013

  20. [28]

    Automated analy- sis of whole brain vasculature using machine learning,

    M. I. Todorov, J. C. Paetzold, O. Schoppe, G. Tetteh, V . Efremov, K. V¨olgyi, M. D¨uring, M. Dichgans, M. Pi- raud, B. Menze, and A. Ert ¨urk, “Automated analy- sis of whole brain vasculature using machine learning,” bioRxiv, 2019

    2019

  21. [29]

    nnU-Net Revisited: A Call for Rigorous Validation in 3D Medical Image Segmentation ,

    F. Isensee, T. Wald, C. Ulrich, M. Baumgartner, S. Roy, K. Maier-Hein, and P. F. J ¨ager, “ nnU-Net Revisited: A Call for Rigorous Validation in 3D Medical Image Segmentation ,” in MICCAI, 2024

    2024

  22. [30]

    U-net: Convo- lutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convo- lutional networks for biomedical image segmentation,” in MICCAI, 2015. A. APPENDIX A.1. Critical pixel mask Alg. 1 displays the pseudo-code for the extraction of our critical pixel mask. Algorithm 1CLoss critical pixel ma...

    2015

Pith tools

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