Pith. sign in

REVIEW 4 major objections 7 minor 32 references

SiliCoN: Simultaneous Nuclei Segmentation and Color Normalization of Histological Images

T0 review · 4 major / 7 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 SiliCoN performs simultaneous nuclei segmentation and color normalization in one deep generative model, reporting better results than all compared methods on both benchmarks.

desk verdict A useful joint color-normalization and nuclei-segmentation model with extensive experiments, but the segmentation comparison needs baseline training details and the joint-training benefit needs a sequential control. read the letter →

arxiv 2506.07028 v1 pith:SUXA454C submitted 2025-06-08 eess.IV cs.AIcs.CV

Suman Mahapatra , Pradipta Maji This is my paper
classification eess.IVcs.AIcs.CV
keywords nucleisegmentationcolornormalizationhistologicalimageanalysisdeepgenerativemodeltruncatednormalmixturespatialattentiondisentangledrepresentationstainconsistency
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to show that color normalization and nuclei segmentation should be solved together, not as separate preprocessing and analysis steps. It introduces SiliCoN, a deep generative model that takes an H&E-stained histology image and produces both a color-normalized version and a nuclei segmentation map in a single framework. The model keeps a latent color-appearance code independent of the segmentation and embedding maps, so a template image's color code can be substituted for a source image's color code without, in principle, disturbing the structure maps. Using a mixture of truncated normal distributions for the color prior and spatial attention for segmentation, SiliCoN is reported to outperform all compared color-normalization methods on a public breast-cancer benchmark and all compared segmentation models on a public histology benchmark.

What carries the argument

The load-bearing mechanism is the assumed factorization of the variational posterior as $Q(z_c)Q(y, z_\omega)$, which makes the latent color code independent of both the segmentation map and the embedding map. This is what permits Algorithm 1's inference trick: encode the template image's color code, encode the source image's structure maps, and decode to obtain a normalized image whose nuclei boundaries should be unchanged. The color code is drawn from a mixture of truncated normal distributions, chosen to model overlapping stains without letting the tails of the mixture dominate, and the segmentation generator is a residual attention U-Net that takes the hematoxylin channel as input. An adversarial discriminator on the quadruplet $(x, z_c, y, z_\omega)$ pushes the reconstructed encodings toward real ones.

What would settle it

Take a trained SiliCoN model, generate a segmentation map from a source image's own color code, then generate another map after swapping in a different image's template color code; if the Dice score between the two maps falls systematically below retest noise across many images, the color code is carrying structure information and the independence assumption fails.

Watch

Extended reading notes

Core claim

The central claim is that a single generative model can normalize stain color and segment nuclei simultaneously, and that doing so outperforms both the established color-normalization methods and the established segmentation networks. Each image is encoded into a latent color code $z_c$, a nuclei segmentation map $y$, and an embedding map $z_\omega$ that carries non-nucleus cellular detail; the decoder $G$ rebuilds an image from these three inputs. The key move is to treat $z_c$ as independent of $(y, z_\omega)$, so that the reconstruction objective factors, and so that at inference the model can replace the source color code with a template color code while keeping the structure maps fixed. The paper reports that SiliCoN achieves the best color-constancy indices on the breast-cancer set and the best Dice, Jaccard, precision, and recall on the public histology set among all compared methods.

Load-bearing premise

The load-bearing premise is that the part of the model's representation that captures color carries no information about where nuclei are; if it does, replacing a source image's color code with a template's could blur or move the segmentation boundaries.

Editorial extensions

If this is right

  • A single SiliCoN model can replace the two-stage workflow of normalizing color first and then segmenting nuclei, since both outputs come from one pass.
  • Stain normalization becomes a template swap at inference: any source image can be re-colored to match any chosen template image without retraining on that template.
  • Because the color code is supposed to carry no structure information, segmentation should remain stable across large stain variations, which matters for multi-center and multi-scanner histology data.
  • The reported comparisons suggest that joint training yields better segmentation accuracy than applying existing segmentation networks to separately normalized images.

Reading between the lines

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

  • Beyond the paper: the same template-swap mechanism should support arbitrary stain transfer between any pair of slides, not only normalization to one fixed template, provided the independence assumption holds.
  • Editorial inference: the factorization could be transferred to other histology tasks where appearance and morphology need to be separated, such as batch-effect removal before biomarker quantification.
  • A testable extension not reported here is to probe residual dependence between $z_c$ and $y$ by perturbing the color code adversarially and measuring how much the segmentation map changes.
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 / 7 minor

Summary. The paper proposes SiliCoN, a deep generative model that simultaneously performs nuclei segmentation and stain color normalization of H&E histology images. It combines a color appearance encoder, a segmentation map generator, an embedding map generator, a decoder, and a discriminator, and assumes that the latent color appearance code is independent of the segmentation map and the embedding map. The training objective is an adversarial term plus a reconstruction and ELBO-style term, with a mixture of truncated normal priors for the color code; at inference, the source color code is replaced with a template color code and the segmentation map is recomputed on the normalized image. Experiments on UCSB and TCGA report state-of-the-art color normalization and nuclei segmentation results, supported by ablations and statistical significance tests. The paper does not release code and gives no training protocol for the baseline segmentation methods.

Significance. If the reported results are reproducible, the joint formulation is a useful contribution to computational pathology, because stain variation is known to degrade nuclei segmentation and the two tasks are naturally intertwined. Strengths of the paper include an ablation of the five objective terms, a correlated-latent control for the independence assumption, statistical significance testing, and comparison with many existing methods. The innovation is, however, incremental relative to the authors' prior TredMiL model [19] and the first author's dissertation [23], and the central comparative claims currently rest on undocumented baseline protocols and at least one overstated significance result.

major comments (4)
  1. [III.B, Table II] The paper reports its own training setup (15 TCGA training images, 2,535 overlapping 256x256 patches, with 507 validation and 192 test patches) but gives no training protocol for the baseline segmentation methods U-Net, Mask-R-CNN, U-Net++, HoVer-Net, MoNS, Stardist, WNSeg, Swin-MIL, and BoNuS. If the Table II numbers are taken from the original publications, differences in organ, split, and preprocessing make the comparison invalid; if the baselines were retrained on the same split, that must be documented with hyperparameters. Without this information, the claim that SiliCoN outperforms all compared segmentation algorithms is unverifiable.
  2. [Conclusion, Section III.B] The paper's central claim is that simultaneous training improves both tasks, but no experiment isolates the benefit of joint training. A sequential baseline that first applies SiliCoN's color normalization using the template color code and then runs the same F_phi segmentation network on the normalized image would test whether joint training adds anything beyond normalization-then-segmentation. The concluding statement that 'nuclei segmentation on color normalized histological images enhances the segmentation accuracy' is not directly supported by any reported comparison.
  3. [Section III.B, Tables III and IV] The text states that SiliCoN performs significantly better 'in all the cases' with both the paired-t and Wilcoxon tests, but the MoNS row of Table III reports recall p=1.93E-01 and Table IV reports recall p=1.36E-01, both above the 0.05 significance level. The statistical claim is therefore internally inconsistent and should be corrected, or the non-significant comparisons should be discussed explicitly.
  4. [Section III.A.2 and Algorithm 1] The independence assumption between zc and (y, z_omega) is load-bearing for the reconstruction decomposition in Eqs. (10)-(17) and for the template-code swap in Algorithm 1. The correlated-counterpart experiment in Section III.A.2 evaluates only stain vector estimation and color normalization indices (NMI, BiCC, WsCC), not segmentation quality; if zc carries structure information, segmentation after template replacement could degrade without affecting those color metrics. The paper should report segmentation metrics for the correlated counterpart or otherwise quantify dependence between zc and the segmentation/embedding maps.
minor comments (7)
  1. [Section II.C, Eqs. (16)-(17)] The step from Eq. (16) to Eq. (17) uses the factorization Q(zc, y, z_omega) = Q(zc)Q(y, z_omega), but this factorization is never stated explicitly; please add it next to the independence assumption.
  2. [Section III, UCSB experiments] Algorithm 1 requires a template image xT, but the manuscript does not state which UCSB image was used as the template; normalized outputs are template-dependent and this choice affects reproducibility.
  3. [Section III.A.2] The correlated counterpart is described only verbally; please specify how dependence between zc and (y, z_omega) is enforced in the network architecture and objective.
  4. [Section III, training setup] No values are reported for lambda_Adv, lambda_Rec, the number of mixture components and truncation parameters of PZc, the dimensionality of zc and z_omega, the optimizer, learning rate, number of epochs, or batch size; these are needed to re-implement SiliCoN.
  5. [Figure 9 caption] The caption lists U-Net, Mask-R-CNN, U-Net++, HoVer-Net, MoNS, Stardist, Swin-MIL, BoNuS, and SiliCoN but omits WNSeg, which appears in Table II; align the figure label and table.
  6. [General] No code or trained models are provided; given the large number of undocumented hyperparameters, releasing the implementation would substantially increase reproducibility.
  7. [Section II] The text notes that some results were originally reported in the first author's doctoral dissertation [23]; please clarify which experimental results and model components are new to this paper to avoid ambiguity about originality.

Circularity Check

0 steps flagged · score 1.0 of 10

No material circularity: the joint model is evaluated on external UCSB/TCGA benchmarks, and the reused truncated-normal prior and TredMiL context are not load-bearing.

full rationale

The derivation chain does not reduce to its own inputs. The independence assumption between zc and (y, zomega) is stated in Section II.C as a modeling assumption, and the reconstruction objective in Eqs. (10)-(17) explicitly builds on that assumption; it is not imported from prior work as a proved theorem. The mixture-of-truncated-normals prior is likewise presented as a modeling choice in the same section, not justified by citing the authors' earlier TredMiL paper [19]. Although TredMiL and the first author's doctoral dissertation [23] are self-references, they are used as a comparison baseline and a provenance note, respectively, and neither carries the central claim. The color-normalization procedure in Algorithm 1, which replaces the source color code with the template color code, is the method's definition rather than a hidden prediction, and its utility is tested against external methods on UCSB using NMI, BiCC, and WsCC. Nuclei segmentation is benchmarked on TCGA against multiple independent baselines in Table II. Concerns about whether the baseline segmentation numbers in Table II used comparable training splits are about experimental comparability, not circularity. No quoted step exhibits the specific kind of reduction-by-construction required to claim circularity.

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

No new physical entities are proposed; the latent variables (zc, y, zω) are modeling constructs rather than independent evidence-bearing entities.

free parameters (5)
  • Loss weights λ_Adv and λ_Rec = not reported
    Relative weights of adversarial and reconstruction terms in Eq. (1), constrained by λ_Adv + λ_Rec = 1; no values or tuning procedure given.
  • Mixture components and truncation parameters of PZc = not reported
    Prior for latent color code in Section II.C.3; number of components, means, variances, and truncation bounds are not specified.
  • Dimensionality of zc and zω = not reported
    Latent code sizes for Ec and Eω affect model capacity; not stated.
  • Training hyperparameters (optimizer, learning rate, epochs, batch size) = not reported
    Required for reproduction; only the GPU and patch counts are given.
  • Template image xT = not specified
    Algorithm 1 requires a template to define target color appearance; selection criteria are not described.
assumptions (5)
  • domain assumption The latent color appearance code zc is independent of the nuclei segmentation map y and the embedding map zω.
    Stated in Section II and used to factorize the joint distributions in Eqs. (10)-(17); no quantitative independence metric is reported.
  • domain assumption The prior PZc(zc) is a mixture of truncated normal distributions, and the prior PYZω(y, zω) is a standard normal distribution.
    Assumed in Section II.C.3 to model stain overlap and embedding information; the number of mixture components and truncation bounds are not given.
  • ad hoc to paper The joint surrogate posterior Q(zc, y, zω) factorizes as Q(zc)Q(y, zω).
    Required to convert the KL terms in Eq. (16) into separable regularizers R1 and R2; this factorization is not stated explicitly as a separate assumption.
  • domain assumption The H-channel extracted through HED color deconvolution, combined with RA-UNet spatial attention, is sufficient to represent nuclei regions.
    Section II.B relies on HED conversion and the existing RA-UNet architecture to generate segmentation maps.
  • domain assumption The UCSB and TCGA datasets are representative benchmarks for evaluating the claimed simultaneous improvements.
    All quantitative conclusions for color normalization and nuclei segmentation are drawn from these two public datasets; generalization is not shown.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SiliCoN: Simultaneous Nuclei Segmentation and Color Normalization of Histological Images." pith.science (2026). https://pith.science/paper/SUXA454C

@misc{pith2026250607028,
  author       = {Pith},
  title        = {Pith review of: SiliCoN: Simultaneous Nuclei Segmentation and Color Normalization of Histological Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SUXA454C}},
  note         = {Machine review of arXiv:2506.07028}
}
read the original abstract

Segmentation of nuclei regions from histological images is an important task for automated computer-aided analysis of histological images, particularly in the presence of impermissible color variation in the color appearance of stained tissue images. While color normalization enables better nuclei segmentation, accurate segmentation of nuclei structures makes color normalization rather trivial. In this respect, the paper proposes a novel deep generative model for simultaneously segmenting nuclei structures and normalizing color appearance of stained histological images.This model judiciously integrates the merits of truncated normal distribution and spatial attention. The model assumes that the latent color appearance information, corresponding to a particular histological image, is independent of respective nuclei segmentation map as well as embedding map information. The disentangled representation makes the model generalizable and adaptable as the modification or loss in color appearance information cannot be able to affect the nuclei segmentation map as well as embedding information. Also, for dealing with the stain overlap of associated histochemical reagents, the prior for latent color appearance code is assumed to be a mixture of truncated normal distributions. The proposed model incorporates the concept of spatial attention for segmentation of nuclei regions from histological images. The performance of the proposed approach, along with a comparative analysis with related state-of-the-art algorithms, has been demonstrated on publicly available standard histological image data sets.

Figures

Figures reproduced from arXiv: 2506.07028 by the authors.

Figure 1
Figure 1. Block diagram of the proposed SiliCoN model to simult [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 3
Figure 3. Comparative performance analysis of SiliCoN model a [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 2
Figure 2. Comparative performance analysis of SiliCoN model a [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figures from the paper (4 more)
Figure 6
Figure 6. Figure 6: Analysis of comparative performances of proposed Si [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 5
Figure 5. Figure 5: Comparative performance analysis of SiliCoN model a [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 8
Figure 8. Figure 8: (a) Original images of UCSB data; and stain color norm [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 9
Figure 9. Figure 9: (a) Original image patches of TCGA data, (b) ground-tr [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

32 extracted references · 32 canonical work pages

  1. [19]

    Truncated Normal Mixture Pri or Based Deep Latent Model for Color Normalization of Histology Imag es,

    S. Mahapatra and P . Maji, “Truncated Normal Mixture Pri or Based Deep Latent Model for Color Normalization of Histology Imag es,” IEEE Transactions on Medical Imaging , vol. 42, no. 6, pp. 1746–1757, 2023

    work page 2023

  2. [23]

    Nuclei Segmentation and Color Normaliz ation of His- tological Images: Rough-Fuzzy Circular Clustering to Deep Generative Modeling,

    S. Mahapatra, “Nuclei Segmentation and Color Normaliz ation of His- tological Images: Rough-Fuzzy Circular Clustering to Deep Generative Modeling,” Doctoral Dissertation, Indian Statistical Ins titute, December, 2024

    work page 2024

  3. [1]

    Color Transfer Between Images,

    E. Reinhard et al., “Color Transfer Between Images,” IEEE Computer Graphics and Applications , vol. 21, no. 5, pp. 34–41, 2001

    work page 2001

  4. [2]

    Statistics of Cone Responses to Na tural Images: Implications for Visual Coding,

    D. L. Ruderman et al., “Statistics of Cone Responses to Na tural Images: Implications for Visual Coding,” Journal of the Optical Society of America A , vol. 15, no. 8, pp. 2036–2045, 1998

    work page 1998

  5. [3]

    A Method for Normalizing Histology Sl ides for Quantitative Analysis,

    M. Macenko et al., “A Method for Normalizing Histology Sl ides for Quantitative Analysis,” in Proceedings of IEEE International Symposium on Biomedical Imaging: From Nano to Macro , 2009, pp. 1107–1110

    work page 2009

  6. [4]

    A Complete Color Normalizat ion Approach to Histopathology Images Using Color Cues Computed from Sat uration- Weighted Statistics,

    X. Li and K. N. Plataniotis, “A Complete Color Normalizat ion Approach to Histopathology Images Using Color Cues Computed from Sat uration- Weighted Statistics,” IEEE Transactions on Biomedical Engineering , vol. 62, no. 7, pp. 1862–1873, 2015

    work page 2015

  7. [5]

    Structure-Preserving Color Normal ization and Sparse Stain Separation for Histological Images,

    A. V ahadane et al., “Structure-Preserving Color Normal ization and Sparse Stain Separation for Histological Images,” IEEE Transactions on Medical Imaging , vol. 35, no. 8, pp. 1962–1971, 2016

    work page 1962

  8. [6]

    Adversarial Stain Transfe r for Histopathology Image Analysis,

    A. Bentaieb and G. Hamarneh, “Adversarial Stain Transfe r for Histopathology Image Analysis,” IEEE Transactions on Medical Imag- ing, vol. 37, no. 3, pp. 792–802, 2018

    work page 2018

Show all 32 references
  1. [7]

    Stain Normalization of Histopatho logy Images Using Generative Adversarial Networks,

    F. G. Zanjani et al., “Stain Normalization of Histopatho logy Images Using Generative Adversarial Networks,” in Proceedings of IEEE Inter- national Symposium on Biomedical Imaging (ISBI) , 2018, pp. 573–577

    2018

  2. [8]

    StainGAN: Stain Style Transfer for D igital Histological Images,

    M. T. Shaban et al., “StainGAN: Stain Style Transfer for D igital Histological Images,” in Proceedings of IEEE International Symposium on Biomedical Imaging (ISBI) , 2019, pp. 953–956

    2019

  3. [9]

    Unpaired Image-to-Image Translation using Cycle- Consistent Adversarial Networks,

    J.-Y . Zhu et al., “Unpaired Image-to-Image Translation using Cycle- Consistent Adversarial Networks,” in Proceedings of IEEE International Conference on Computer Vision (ICCV) , 2017, pp. 2242–2251

    2017

  4. [10]

    Circular Mixture Modeling of Color Distribu- tion for Blind Stain Separation in Pathology Images,

    X. Li and K. Plataniotis, “Circular Mixture Modeling of Color Distribu- tion for Blind Stain Separation in Pathology Images,” IEEE Journal of Biomedical and Health Informatics , vol. 21, no. 1, pp. 150–161, 2017

    2017

  5. [11]

    Circular Clustering in Fuzzy Approximation Spaces for Color Normalization of Histological Images,

    P . Maji and S. Mahapatra, “Circular Clustering in Fuzzy Approximation Spaces for Color Normalization of Histological Images,” IEEE Trans- actions on Medical Imaging , vol. 39, no. 5, pp. 1735–1745, 2020

    2020

  6. [12]

    U-Net: Convolutional Networks for Biomed ical Image Seg- mentation,

    O. R. et al., “U-Net: Convolutional Networks for Biomed ical Image Seg- mentation,” in Proceedings of Medical Image Computing and Computer- Assisted Intervention, 2015, pp. 234–241

    2015

  7. [13]

    UNet++: A Nested U-Net Architecture for Medical Image Segmentation,

    Z. Zhou et al., “UNet++: A Nested U-Net Architecture for Medical Image Segmentation,” in Proceedings of Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decisi on Support, 2018, pp. 3–11

    2018

  8. [14]

    Mask R-CNN,

    K. He et al., “Mask R-CNN,” in Proceedings of IEEE International Conference on Computer Vision , 2017, pp. 2961–2969

    2017

  9. [15]

    Hover-Net: Simultaneous Segmentati on and Classi- fication of Nuclei in Multi-Tissue Histology Images,

    S. Graham et al., “Hover-Net: Simultaneous Segmentati on and Classi- fication of Nuclei in Multi-Tissue Histology Images,” Medical Image Analysis, vol. 58, p. 101563, 2019

    2019

  10. [16]

    Nuclei Instance Segmentati on and Classi- fication in Histopathology Images with Stardist,

    M. Weigert and U. Schmidt, “Nuclei Instance Segmentati on and Classi- fication in Histopathology Images with Stardist,” in Proceedings of IEEE International Symposium on Biomedical Imaging Challenges , 2022, pp. 1–4

    2022

  11. [17]

    Transformer Based Multiple Instance Le arning for Weakly Supervised Histopathology Image Segmentation,

    Z. Qian et al., “Transformer Based Multiple Instance Le arning for Weakly Supervised Histopathology Image Segmentation,” in Proceed- ings of Medical Image Computing and Computer Assisted Inter vention, 2022, pp. 160–170

    2022

  12. [18]

    BoNuS: Boundary Mining for Nuclei Segmen tation with Partial Point Labels,

    Y . Lin et al., “BoNuS: Boundary Mining for Nuclei Segmen tation with Partial Point Labels,” IEEE Transactions on Medical Imaging , pp. 1–11, 2024

    2024

  13. [20]

    Nuclei Segmentation in Histopatho logy Images Using Structure-Preserving Color Normalization Based Ens emble Deep Learning Frameworks,

    M. R. Prusty et al., “Nuclei Segmentation in Histopatho logy Images Using Structure-Preserving Color Normalization Based Ens emble Deep Learning Frameworks,” Computers, Materials & Continua , vol. 77, no. 3, 2023

    2023

  14. [21]

    Optimized Detection and Segmentatio n of Nuclei in Gastric Cancer Images Using Stain Normalization and Blurre d Artifact Removal,

    O. Martos et al., “Optimized Detection and Segmentatio n of Nuclei in Gastric Cancer Images Using Stain Normalization and Blurre d Artifact Removal,” Pathology - Research and Practice, vol. 248, p. 154694, 2023

    2023

  15. [22]

    Improving Generalization Capabilit y of Deep Learning-based Nuclei Instance Segmentation by Non-deter ministic Train Time and Deterministic Test Time Stain Normalization ,

    A. Mahbod et al., “Improving Generalization Capabilit y of Deep Learning-based Nuclei Instance Segmentation by Non-deter ministic Train Time and Deterministic Test Time Stain Normalization ,” Com- putational and Structural Biotechnology Journal , vol. 23, pp. 669–678, 2024

    2024

  16. [24]

    RA-UNet: A Hybrid Deep Attention-Aware N etwork to Extract Liver and Tumor in CT Scans,

    Q. Jin et al., “RA-UNet: A Hybrid Deep Attention-Aware N etwork to Extract Liver and Tumor in CT Scans,” Frontiers in Bioengineering and Biotechnology, vol. 8, p. 605132, 2020

    2020

  17. [25]

    Image Quality Assessment: From Error Vi sibility to Structural Similarity,

    W. Zhou et al., “Image Quality Assessment: From Error Vi sibility to Structural Similarity,” IEEE Transactions on Image Processing , vol. 13, no. 4, pp. 600–612, 2004

    2004

  18. [26]

    Algorithm and Benchmark Dataset fo r Stain Separation in Histology Images,

    M. T. McCann et al., “Algorithm and Benchmark Dataset fo r Stain Separation in Histology Images,” in Proceedings of IEEE International Conference on Image Processing , 2014, pp. 3953–3957

    2014

  19. [27]

    A Nonlinear Mapping Approach to Stain Normal- ization in Digital Histopathology Images Using Image-Spec ific Color Deconvolution,

    A. M. Khan et al., “A Nonlinear Mapping Approach to Stain Normal- ization in Digital Histopathology Images Using Image-Spec ific Color Deconvolution,” IEEE Transactions on Biomedical Engineering , vol. 61, no. 6, pp. 1729–1738, 2014

    2014

  20. [28]

    Deep Adversarial Training for Multi -Organ Nuclei Segmentation in Histopathology Images,

    F. Mahmood et al., “Deep Adversarial Training for Multi -Organ Nuclei Segmentation in Histopathology Images,” IEEE Transactions on Medical Imaging, vol. 39, no. 11, pp. 3257–3267, 2020

    2020

  21. [29]

    Weakly Supervised Nuclei Segme ntation Via Instance Learning,

    W. Liu, Q. He, and X. He, “Weakly Supervised Nuclei Segme ntation Via Instance Learning,” in Proceedings of IEEE International Symposium on Biomedical Imaging , 2022, pp. 1–5

    2022

  22. [30]

    Evaluation and Benchmark for Biol ogical Image Segmentation,

    E. D. Gelasca et al., “Evaluation and Benchmark for Biol ogical Image Segmentation,” in Proceedings of IEEE International Conference on Image Processing, 2008, pp. 1816–1819

    2008

  23. [31]

    A Dataset and a Technique for Generaliz ed Nuclear Segmentation for Computational Pathology,

    N. Kumar et al., “A Dataset and a Technique for Generaliz ed Nuclear Segmentation for Computational Pathology,” IEEE Transactions on Medical Imaging , vol. 36, no. 7, pp. 1550–1560, 2017

    2017

  24. [32]

    New V ariants of A Method of MRI Scale St andard- ization,

    L. G. Nyul et al., “New V ariants of A Method of MRI Scale St andard- ization,” IEEE Transactions on Medical Imaging , vol. 19, no. 2, pp. 143–150, 2000

    2000

Pith tools

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