Pith. sign in

REVIEW 4 major objections 6 minor 59 references

IRS: Incremental Relationship-guided Segmentation for Digital Pathology

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 The paper claims that encoding anatomical relationships between previously learned and newly added classes in an incremental proposition matrix lets a kidney-pathology segmentation model add classes continually without forgetting old…

desk verdict A practically motivated continual segmentation paper whose central anatomy loss in Eq. (4) appears to invert the stated constraints; needs a major revision but deserves a referee's look. read the letter →

arxiv 2505.22855 v1 pith:AOKOEYQV submitted 2025-05-28 eess.IV cs.CV

Ruining Deng , Junchao Zhu , Juming Xiong , Can Cui , Tianyuan Yao , Junlin Guo , Siqi Lu , Marilyn Lionts
show 9 more authors
Mengmeng Yin Yu Wang Shilin Zhao Yucheng Tang Yihe Yang Paul Dennis Simonson Mert R. Sabuncu Haichun Yang Yuankai Huo
This is my paper
classification eess.IVcs.CV
keywords continuallearningsemanticsegmentationdigitalpathologywholeslideimagesclass-incrementalmixture-of-expertskidneyanatomicalrelationships
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 introduces IRS, a continual learning scheme for segmenting digital whole-slide kidney images. The authors' central claim is that explicitly modeling anatomical relationships—whether an old class is a superset, subset, or mutually exclusive with a newly introduced class—lets the model absorb new classes over time while retaining old-class knowledge, even when old-class labels are absent. They represent these relationships in an expandable 'incremental universal proposition matrix' and supervise a semi-supervised anatomy loss from it, alongside a prompt-driven mixture-of-experts backbone with class- and scale-specific tokens. On a 24-class kidney dataset spanning regions, functional units, cells, and out-of-distribution lesions, IRS reports higher Dice than prior continual segmentation baselines in 2-, 3-, and 4-step settings (e.g., 61.67 average vs. 57.52 for CL-LoRA in the 2-step setting). The work matters because clinical pathology data arrives over time with partial, multi-scale annotations, so a model that can add classes without full re-annotation or forgetting is directly actionable.

What carries the argument

The core object is the incremental universal proposition matrix $M^t \in \mathbb{R}^{m\times n}$, where $m$ is the number of old classes and $n$ the number of new classes at step $t$; each entry encodes whether the new class is a superset, subset, or mutually exclusive of the old class. From these entries the anatomy loss $L_{\text{anatomy}}$ in Eq. (4) constrains old-class predictions $P_i$ against new-class labels $Y_j$. The second machinery is the prompt-driven dynamic MoE backbone: three encoders (PrPSeg, Swin-UNETR, EfficientSAM) are fused via self-attention, and a learnable token bank supplies class-specific and scale-specific tokens ($5\times$, $10\times$, $20\times$, $40\times$) that condition both the encoder blocks and the dynamic head. A distillation loss then aligns old-class tokens, latent features, decoder features, and logits between steps. These components together implement the claimed spatial-temporal continual learning.

What would settle it

For any old/new class pair declared 'superset' in the matrix, measure the containment Dice between the predicted old-class region and the new-class label, that is, $\text{Dice}(P_i, P_i \cap Y_j)$, before and after training on a validation set of kidney WSIs. The paper's mechanism predicts this containment score should increase after the anatomy loss is applied; if it decreases or stays flat, the stated relationship constraint is not doing the claimed work. A complementary check is to flip the directions in Eq. (4), using $\text{DCE}(Y_j, P_i)$ for the superset case, and see whether performance, contrary to the mechanism, stays similar.

Watch

Extended reading notes

Core claim

The paper's central claim is that class-incremental segmentation in digital pathology can be guided by anatomical structure. The authors model relationships between every old class and every new class in a proposition matrix whose entries encode three relation types: superset, subset, and mutually exclusive. These entries generate an anatomy loss that constrains the old-class predictions produced from images labeled only for the new class, so the model receives spatial supervision for old classes without their ground-truth masks. Combined with a dynamic mixture-of-experts network whose three backbones (PrPSeg, Swin-UNETR, EfficientSAM) are fused by self-attention and steered by learned class and scale tokens, the authors report that the model retains old classes while learning new ones, and that it generalizes to out-of-distribution disease lesions. The strongest empirical statement is the 2-step average Dice of 61.67, above the best baseline CL-LoRA at 57.52, with similar advantages in the 3- and 4-step schedules.

Load-bearing premise

The load-bearing premise is that the anatomy-loss equation (4) actually pushes old-class predictions into the intended subset, superset, and mutual-exclusion relations with the new-class labels; if the sign or complement in these Dice terms is wrong, the relationship-guided constraint would distort rather than preserve old-class knowledge.

Editorial extensions

If this is right

  • Old-class retention in later steps: the model keeps higher Dice on previously seen classes than baselines when new classes are added.
  • New-class learning speed: after only one epoch of supervised training per new class, the model achieves competitive new-class Dice, suggesting lightweight updates are sufficient.
  • Multi-scale coverage: one unified model segments regions, functional units, cells, and lesions across magnifications, so a single continual pipeline can cover the whole kidney panorama.
  • Data-order robustness: results hold under reversed orders D2→D1 and D2→D1→D3, indicating the method does not depend on a particular arrival sequence.
  • OOD extension: out-of-distribution lesions such as nodular sclerosis can be added as new classes without degrading earlier normal-anatomy classes.

Reading between the lines

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

  • The anatomy-loss formulation is generic: any domain with hierarchical part-of or exclusion relations, such as lung lobes→bronchioles→alveoli or tumor subregions, could reuse the same matrix and loss provided the relation directions are validated.
  • The single-epoch-per-class schedule suggests the relationship constraints carry most of the transfer signal; a factorial experiment varying the anatomy loss alone versus the MoE alone would quantify the interaction and identify which relation types drive the gain.
  • The proposition matrix is hand-defined from anatomy; a natural extension is to learn the relation entries from data, for instance from co-occurrence statistics or a small set of annotated pairs, so the method can scale to organs whose anatomy is less standardized.
  • The reported numbers compare against continual-learning baselines, but the Joint upper bound is not reached; the remaining gap likely comes from a combination of backbone capacity and the distillation schedule, an interaction the paper does not fully isolate.
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 IRS, a class-incremental semantic segmentation framework for kidney pathology. It combines a prompt-driven mixture-of-experts encoder/decoder with an incremental universal proposition matrix that encodes anatomical relations (superset, subset, mutually exclusive) between old and new classes. An anatomy loss plus feature/logit distillation losses are designed to retain old-class knowledge when only new-class labels are available. Experiments on a 24-class partially labeled kidney dataset across 2-, 3-, and 4-step continual settings report consistent improvements over existing continual segmentation baselines.

Significance. The problem is timely and the dataset/evaluation is substantial. The paper's empirical results are generally positive and cover several continual settings, and the code is publicly available. If the relationship-guided mechanism can be correctly specified, the idea of injecting anatomical priors into continual segmentation would be valuable. However, the central loss in Eq. (4) appears to implement the opposite of the stated constraints under the standard Dice loss, so the current manuscript cannot support its main mechanistic claim.

major comments (4)
  1. [§III-C, Eq. (4)] With DCE as the standard Dice loss, Eq. (4) does not implement the three stated constraints. For the superset case j ▷ i, DCE(1-Y_j, P_i) is minimized when P_i matches the complement of Y_j, which drives the old-class prediction outside the new-class region, the opposite of 'P_i should not exceed Y_j'. For the subset case j ◁ i, -DCE(Y_j, Y_j ∪ P_i) is minimized when the union Y_j ∪ P_i is as large as possible relative to Y_j, i.e., when P_i contributes outside Y_j; this penalizes coverage of Y_j, again contrary to the stated goal. Only the mutual-exclusion term DCE(Y_j, P_i) has the desired direction. Because Table IV attributes part of the performance gain to Lanatomy, the paper must either correct the signs and arguments in Eq. (4), define DCE with a non-standard formula and give it explicitly, or re-derive the ablation with a correctly implemented constraint.
  2. [§III-A] The matrix values that determine which relationship (superset, subset, or mutually exclusive) applies are only 'defined in [23]'; they are not given in this paper, nor is the mapping from relationship types to the symbols ▷, ◁, and ∥ made explicit. Since Eq. (4) and the entire anatomy-loss mechanism depend on these values, the manuscript should reproduce the relevant definitions in Sec. III-A or in an appendix, and state the exact convention used for the symbols.
  3. [§V-A, Tables II, IV, VI; Fig. 6] All reported p-values are p < 0.001 and no repeated runs, standard deviations, or error bars are provided. It is therefore unclear whether the observed margins (e.g., 61.67 vs. 57.52 in Table II) are stable across seeds or whether the p-values come from a per-class paired test. Please report the test procedure and variance estimates, or at least multiple-seed means and ranges, for the central comparisons.
  4. [§V, Model Complexity; §IV-B] The model-complexity paragraph states that all methods use the same 19,194,168-parameter backbone, but Sec. IV-B and Table V describe IRS as a MoE-3 network that integrates PrPSeg, Swin-UNETR, and EfficientSAM encoders. This makes it unclear whether the continual-learning baselines in Table II use a single one of these backbones while IRS uses all three, which would confound the comparison. Please specify the exact backbone configuration for each method and verify that the parameter count includes all three encoders.
minor comments (6)
  1. [§III-A] The manuscript says 'a incremental universal proposition matrix' and should say 'an incremental universal proposition matrix'; also, the notation m = M_t(i, j) reuses m both as the number of old classes and as a matrix entry, so a different symbol such as v = M_t(i, j) would be clearer.
  2. [§III-C] DCE is used in Eq. (4) before being defined; please define it with an explicit formula, including how soft predictions are binarized or thresholded.
  3. [§IV-B] The heading 'T raining Process' contains a typo and should read 'Training Process'.
  4. [Table IV] The checkmark layout is ambiguous; explicitly list which loss terms are active in each row. Also, 'Psuedo-labels' should be 'Pseudo-labels'.
  5. [Table V] The footnote says 'backones' and should say 'backbones'.
  6. [Fig. 6] The multiple panels and legends are hard to read; label each panel with the step order and method name to make the comparison interpretable.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the anatomy-guide matrix is prior clinical knowledge and the headline results are benchmarked against external baselines.

full rationale

The derivation chain is not circular. The proposition matrix values are anatomical priors (superset/subset/mutual-exclusion relations among kidney structures) taken from the authors' earlier PrPSeg paper [23] and from clinical anatomy; they are not fitted to the Dice scores reported in Tables II–VI. The headline claim (61.67 vs. 57.52 for CL-LoRA in the 2-step setting) is an empirical comparison against external baselines, and the ablation 'IRS w/o Anatomy' (61.12 average) provides a within-paper control for the anatomy loss. The self-citations to [23] and [42] supply architectural and relational definitions, not the evidence that the method works. The concern raised about Eq. (4)—that with the standard Dice loss, DCE(1-Y_j,P_i) and -DCE(Y_j,Y_j∪P_i) appear to invert the stated superset/subset semantics—is a correctness/internal-consistency issue, not a circularity: the loss is not equivalent to the input labels or to the reported metric by construction. No fitted parameter is renamed as a prediction, and no uniqueness theorem or unverified self-citation is invoked to force the conclusion.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The central claim does not introduce new physical entities. It relies on an anatomical relationship matrix inherited from a self-citation, a potentially mis-specified loss term, and a dataset split that does not support true OOD generalization. The main free parameter is the unspecified weighting of the auxiliary losses.

free parameters (1)
  • Loss weights for L_anatomy and L_semi relative to supervised loss = Not reported
    The total loss is described as an aggregate of supervised loss and anatomy-aware semi-supervised losses, but no weighting scheme or values are given. The final performance likely depends on these weights.
assumptions (3)
  • domain assumption The anatomical relationship matrix from PrPSeg [23] accurately captures the spatial relationships among all 24 classes in this dataset.
    The matrix values are taken from the authors' prior work [23], not re-derived or validated on this dataset; the anatomy loss and all relationship-guided supervision depend on it.
  • ad hoc to paper DCE in Eq. (4) denotes the standard Dice loss and the signs in Eq. (4) are as intended.
    As printed, Eq. (4) appears to invert the stated subset/superset/mutex rules if DCE is the standard Dice loss (see analysis). Either the loss function is mis-specified or DCE is not the standard Dice loss; no definition is given beyond 'DCE denotes the Dice Loss.'
  • domain assumption The 'OOD' test set is representative of out-of-distribution domains.
    All test patches come from the same four institutional sources split at patient level; no external or cross-domain test set is used, so 'OOD disease lesions' refers to unseen classes, not unseen distributions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of IRS: Incremental Relationship-guided Segmentation for Digital Pathology." pith.science (2026). https://pith.science/paper/AOKOEYQV

@misc{pith2026250522855,
  author       = {Pith},
  title        = {Pith review of: IRS: Incremental Relationship-guided Segmentation for Digital Pathology},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AOKOEYQV}},
  note         = {Machine review of arXiv:2505.22855}
}
read the original abstract

Continual learning is rapidly emerging as a key focus in computer vision, aiming to develop AI systems capable of continuous improvement, thereby enhancing their value and practicality in diverse real-world applications. In healthcare, continual learning holds great promise for continuously acquired digital pathology data, which is collected in hospitals on a daily basis. However, panoramic segmentation on digital whole slide images (WSIs) presents significant challenges, as it is often infeasible to obtain comprehensive annotations for all potential objects, spanning from coarse structures (e.g., regions and unit objects) to fine structures (e.g., cells). This results in temporally and partially annotated data, posing a major challenge in developing a holistic segmentation framework. Moreover, an ideal segmentation model should incorporate new phenotypes, unseen diseases, and diverse populations, making this task even more complex. In this paper, we introduce a novel and unified Incremental Relationship-guided Segmentation (IRS) learning scheme to address temporally acquired, partially annotated data while maintaining out-of-distribution (OOD) continual learning capacity in digital pathology. The key innovation of IRS lies in its ability to realize a new spatial-temporal OOD continual learning paradigm by mathematically modeling anatomical relationships between existing and newly introduced classes through a simple incremental universal proposition matrix. Experimental results demonstrate that the IRS method effectively handles the multi-scale nature of pathological segmentation, enabling precise kidney segmentation across various structures (regions, units, and cells) as well as OOD disease lesions at multiple magnifications. This capability significantly enhances domain generalization, making IRS a robust approach for real-world digital pathology applications.

Figures

Figures reproduced from arXiv: 2505.22855 by the authors.

Figure 1
Figure 1. Illustration of continual learning challenges in renal pathology segmentation, depicting: (1) temporal data collection; (2) multi-scale comprehensive objects ranging from regions to cells; (3) partial an￾notations from different annotators; and (4) progression from normal anatomical structures to out-of-distribution (OOD) pathological lesions. wide range of real-world applications, particularly in clinical workflows… view at source ↗
Figure 2
Figure 2. This figure illustrates the transformation of complex clinical anatomical relationships within the kidney into a continual learning paradigm. (a) The kidney’s anatomy showcases the spatial relationships among large-scale objects, including regions, functional units, cells, and lesions. (b) Previous continual learning approaches merely add new classes without establishing connections between new and old classes. (c) … view at source ↗
Figure 3
Figure 3. This figure illustrates the design of the incremental universal proposition matrix as new classes are added in different settings. The proposed matrix is easily adaptable to large-scale objects within the continual learning paradigm. the context of kidney pathology ( [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: This figure illustrates the architecture of our proposed prompt￾driven dynamic MoE network. The model integrates three different architectures using a self-attention mechanism and a dynamic MoE head to enhance segmentation capabilities for large-scale pathology segment…
Figure 5
Figure 5. Figure 5: This figure showcases the key innovation of knowledge distil￾lation in our proposed method. In subsequent steps, the model learns from new classes supervised by new labels, while old class tokens are utilized on new images to predict old classes. The new model distills…
Figure 6
Figure 6. Figure 6: This figure shows the results of 3-step continual learning with 2 orders. Dice similarity coefficient scores (%) are reported. The proposed method achieved superior performance in panoramic renal pathology segmentation across different continual learning settings. TABL…
Figure 7
Figure 7. Figure 7: This figure shows the backbone capability on 24-class segmen￾tation at step 2 for continual learning. The labels for all 24 classes are provided during the training. Dice similarity coefficient scores (%) are reported. C. 4-Step Continual Segmentation Performance Table…
Figure 8
Figure 8. Figure 8: This figure presents the qualitative results of different continual learning approaches on all classes. The proposed method demonstrates superior knowledge distillation for old classes while achieving better learning on new classes. results obtained using different dat…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

59 extracted references · 44 canonical work pages

  1. [23]

    Prpseg: Universal proposition learning for panoramic renal pathology segmentation,

    R. Deng, Q. Liu, C. Cui, T. Yao, J. Yue, J. Xiong, L. Yu, Y . Wu, M. Yin, Y . Wanget al., “Prpseg: Universal proposition learning for panoramic renal pathology segmentation,” inProceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, 2024, pp. 11 736– 11 746

    work page 2024

  2. [42]

    Omni-seg: A scale-aware dynamic 10 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. XX, NO. XX, XXXX 2025 network for renal pathological image segmentation,

    R. Deng, Q. Liu, C. Cui, T. Yao, J. Long, Z. Asad, R. M. Womick, Z. Zhu, A. B. Fogo, S. Zhaoet al., “Omni-seg: A scale-aware dynamic 10 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. XX, NO. XX, XXXX 2025 network for renal pathological image segmentation,”IEEE Transactions on Biomedical Engineering, 2023

    work page 2025

  3. [1]

    Three types of incremental learning,

    G. M. Van de Ven, T. Tuytelaars, and A. S. Tolias, “Three types of incremental learning,”Nature Machine Intelligence, vol. 4, no. 12, pp. 1185–1197, 2022

    2022

  4. [2]

    Ella: An efficient lifelong learning algorithm,

    P. Ruvolo and E. Eaton, “Ella: An efficient lifelong learning algorithm,” inInternational conference on machine learning. PMLR, 2013, pp. 507–515

    work page 2013

  5. [3]

    Alleviating catastrophic forgetting using context-dependent gating and synaptic stabilization,

    N. Y . Masse, G. D. Grant, and D. J. Freedman, “Alleviating catastrophic forgetting using context-dependent gating and synaptic stabilization,” Proceedings of the National Academy of Sciences, vol. 115, no. 44, pp. E10 467–E10 475, 2018

    work page 2018

  6. [4]

    Model zoo: A growing

    R. Ramesh and P. Chaudhari, “Model zoo: A growing” brain” that learns continually,”arXiv preprint arXiv:2106.03027, 2021

    arXiv 2021

  7. [5]

    Classic: Continual and con- trastive learning of aspect sentiment classification tasks,

    Z. Ke, B. Liu, H. Xu, and L. Shu, “Classic: Continual and con- trastive learning of aspect sentiment classification tasks,”arXiv preprint arXiv:2112.02714, 2021

    arXiv 2021

  8. [6]

    An efficient domain-incremental learning approach to drive in all weather condi- tions,

    M. J. Mirza, M. Masana, H. Possegger, and H. Bischof, “An efficient domain-incremental learning approach to drive in all weather condi- tions,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 3001–3011

    work page 2022

Show all 59 references
  1. [7]

    Continual learning with hypernetworks,

    J. V on Oswald, C. Henning, B. F. Grewe, and J. Sacramento, “Continual learning with hypernetworks,”arXiv preprint arXiv:1906.00695, 2019

    1906 arXiv

  2. [8]

    Continual learning with deep generative replay,

    H. Shin, J. K. Lee, J. Kim, and J. Kim, “Continual learning with deep generative replay,”Advances in neural information processing systems, vol. 30, 2017

    2017

  3. [9]

    Brain-inspired replay for continual learning with artificial neural networks,

    G. M. Van de Ven, H. T. Siegelmann, and A. S. Tolias, “Brain-inspired replay for continual learning with artificial neural networks,”Nature communications, vol. 11, no. 1, p. 4069, 2020

    2020

  4. [10]

    A systematic collection of medical image datasets for deep learning,

    J. Li, G. Zhu, C. Hua, M. Feng, B. Bennamoun, P. Li, X. Lu, J. Song, P. Shen, X. Xuet al., “A systematic collection of medical image datasets for deep learning,”ACM Computing Surveys, vol. 56, no. 5, pp. 1–51, 2023

    2023

  5. [11]

    A systematic approach of data collection and analysis in medical imaging research,

    K. Manjunath, C. Rajaram, G. Hegde, A. Kulkarni, R. Kurady, and K. Manuel, “A systematic approach of data collection and analysis in medical imaging research,”Asian Pacific Journal of Cancer Prevention: APJCP, vol. 22, no. 2, p. 537, 2021

    2021

  6. [12]

    Continual learning in medical devices: Fda’s action plan and beyond,

    K. N. V okinger, S. Feuerriegel, and A. S. Kesselheim, “Continual learning in medical devices: Fda’s action plan and beyond,”The Lancet Digital Health, vol. 3, no. 6, pp. e337–e338, 2021

    2021

  7. [13]

    Passive data collection and use in healthcare: A systematic review of ethical issues,

    N. A. Maher, J. T. Senders, A. F. Hulsbergen, N. Lamba, M. Parker, J.-P. Onnela, A. L. Bredenoord, T. R. Smith, and M. L. Broekman, “Passive data collection and use in healthcare: A systematic review of ethical issues,”International journal of medical informatics, vol. 129, pp...

    2019

  8. [14]

    Data collection theory in healthcare research: the minimum dataset in quanti- tative studies,

    C. S. Kwok, E.-A. Muntean, C. D. Mallen, and J. A. Borovac, “Data collection theory in healthcare research: the minimum dataset in quanti- tative studies,”Clinics and practice, vol. 12, no. 6, pp. 832–844, 2022

    2022

  9. [15]

    Towards flexible mobile data collection in healthcare,

    J. Schobel, R. Pryss, M. Schickler, and M. Reichert, “Towards flexible mobile data collection in healthcare,” in2016 IEEE 29th International Symposium on Computer-Based Medical Systems (CBMS). IEEE, 2016, pp. 181–182

    2016

  10. [16]

    Digital pathology and computational image analysis in nephropathol- ogy,

    L. Barisoni, K. J. Lafata, S. M. Hewitt, A. Madabhushi, and U. G. Balis, “Digital pathology and computational image analysis in nephropathol- ogy,”Nature Reviews Nephrology, vol. 16, no. 11, pp. 669–685, 2020

    2020

  11. [17]

    Dynamic memory to alleviate catastrophic forgetting in continual learning with medical imaging,

    M. Perkonigg, J. Hofmanninger, C. J. Herold, J. A. Brink, O. Pianykh, H. Prosch, and G. Langs, “Dynamic memory to alleviate catastrophic forgetting in continual learning with medical imaging,”Nature commu- nications, vol. 12, no. 1, p. 5678, 2021

    2021

  12. [18]

    Continual learning for abdominal multi-organ and tumor segmentation,

    Y . Zhang, X. Li, H. Chen, A. L. Yuille, Y . Liu, and Z. Zhou, “Continual learning for abdominal multi-organ and tumor segmentation,” inInter- national conference on medical image computing and computer-assisted intervention. Springer, 2023, pp. 35–45

    2023

  13. [19]

    Low- rank mixture-of-experts for continual medical image segmentation,

    Q. Chen, L. Zhu, H. He, X. Zhang, S. Zeng, Q. Ren, and Y . Lu, “Low- rank mixture-of-experts for continual medical image segmentation,” in International Conference on Medical Image Computing and Computer- Assisted Intervention. Springer, 2024, pp. 382–392

    2024

  14. [20]

    Lifelong nnu-net: a framework for standardized medical continual learning,

    C. Gonz ´alez, A. Ranem, D. Pinto dos Santos, A. Othman, and A. Mukhopadhyay, “Lifelong nnu-net: a framework for standardized medical continual learning,”Scientific Reports, vol. 13, no. 1, p. 9381, 2023

    2023

  15. [21]

    Deep hierarchical semantic segmentation,

    L. Li, T. Zhou, W. Wang, J. Li, and Y . Yang, “Deep hierarchical semantic segmentation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 1246–1257

    2022

  16. [22]

    Unsupervised hierarchical semantic segmentation with multiview cosegmentation and clustering transformers,

    T.-W. Ke, J.-J. Hwang, Y . Guo, X. Wang, and S. X. Yu, “Unsupervised hierarchical semantic segmentation with multiview cosegmentation and clustering transformers,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 2571–2581

    2022

  17. [24]

    Artificial intelligence in renal pathology: Current status and future,

    C. Feng and F. Liu, “Artificial intelligence in renal pathology: Current status and future,”Bosnian Journal of Basic Medical Sciences, 2022

    2022

  18. [25]

    Evaluating tubulointerstitial compartments in renal biopsy specimens using a deep learning-based approach for classifying normal and abnormal tubules,

    S. Hara, E. Haneda, M. Kawakami, K. Morita, R. Nishioka, T. Zoshima, M. Kometani, T. Yoneda, M. Kawano, S. Karashimaet al., “Evaluating tubulointerstitial compartments in renal biopsy specimens using a deep learning-based approach for classifying normal and abnormal tubules,” ...

    2022

  19. [26]

    Cnn cascades for segmenting whole slide images of the kidney. arxiv 2017,

    M. Gadermayr, A. Dombrowski, B. Klinkhammer, P. Boor, and D. Mer- hof, “Cnn cascades for segmenting whole slide images of the kidney. arxiv 2017,”arXiv preprint arXiv:1708.00251

    2017 arXiv

  20. [27]

    Glomerulus classification and detection based on convolutional neural networks,

    J. Gallego, A. Pedraza, S. Lopez, G. Steiner, L. Gonzalez, A. Laurinavi- cius, and G. Bueno, “Glomerulus classification and detection based on convolutional neural networks,”Journal of Imaging, vol. 4, no. 1, p. 20, 2018

    2018

  21. [28]

    Glomerulosclerosis identification in whole slide images using semantic segmentation,

    G. Bueno, M. M. Fernandez-Carrobles, L. Gonzalez-Lopez, and O. Deniz, “Glomerulosclerosis identification in whole slide images using semantic segmentation,”Computer methods and programs in biomedicine, vol. 184, p. 105273, 2020

    2020

  22. [29]

    An integrated iterative annotation technique for easing neural network training in medical image analysis,

    B. Lutnick, B. Ginley, D. Govind, S. D. McGarry, P. S. LaViolette, R. Yacoub, S. Jain, J. E. Tomaszewski, K.-Y . Jen, and P. Sarder, “An integrated iterative annotation technique for easing neural network training in medical image analysis,”Nature machine intelligence, vol. 1,...

    2019

  23. [30]

    Iterative learning to make the most of unlabeled and quickly obtained labeled data in histology,

    L. Gupta, B. M. Klinkhammer, P. Boor, D. Merhof, and M. Gadermayr, “Iterative learning to make the most of unlabeled and quickly obtained labeled data in histology,” inInternational Conference on Medical Imaging with Deep Learning–Full Paper Track, 2018

    2018

  24. [31]

    Segmentation of glomeruli within trichrome images using deep learning,

    S. Kannan, L. A. Morgan, B. Liang, M. G. Cheung, C. Q. Lin, D. Mun, R. G. Nader, M. E. Belghasem, J. M. Henderson, J. M. Francis et al., “Segmentation of glomeruli within trichrome images using deep learning,”Kidney international reports, vol. 4, no. 7, pp. 955–962, 2019

    2019

  25. [32]

    Automatic nucleus segmentation with mask-rcnn,

    J. W. Johnson, “Automatic nucleus segmentation with mask-rcnn,” in Science and Information Conference. Springer, 2019, pp. 399–407

    2019

  26. [33]

    Evaluating transformer-based seman- tic segmentation networks for pathological image segmentation,

    C. Nguyen, Z. Asad, and Y . Huo, “Evaluating transformer-based seman- tic segmentation networks for pathological image segmentation,”arXiv preprint arXiv:2108.11993, 2021

    2021 arXiv

  27. [34]

    Instance-based vision transformer for subtyping of papillary renal cell carcinoma in histopathological image,

    Z. Gao, B. Hong, X. Zhang, Y . Li, C. Jia, J. Wu, C. Wang, D. Meng, and C. Li, “Instance-based vision transformer for subtyping of papillary renal cell carcinoma in histopathological image,” inMedical Image Computing and Computer Assisted Intervention–MICCAI 2021: 24th Interna...

    2021

  28. [35]

    Self reinforcing multi-class transformer for kidney glomerular basement membrane segmentation,

    S. Yan, X. Huang, W. Lian, and C. Song, “Self reinforcing multi-class transformer for kidney glomerular basement membrane segmentation,” IEEE Access, 2023

    2023

  29. [36]

    Automatic evaluation of histological prognostic factors using two consecutive convolutional neural networks on kidney samples,

    E. Marechal, A. Jaugey, G. Tarris, M. Paindavoine, J. Seibel, L. Martin, M. F. de la Vega, T. Crepin, D. Ducloux, G. Zanettaet al., “Automatic evaluation of histological prognostic factors using two consecutive convolutional neural networks on kidney samples,”Clinical Journal ...

    2022

  30. [37]

    Multi-structure segmentation from partially labeled datasets. application to body compo- sition measurements on ct scans,

    G. Gonz ´alez, G. R. Washko, and R. San Jos ´e Est ´epar, “Multi-structure segmentation from partially labeled datasets. application to body compo- sition measurements on ct scans,” inImage Analysis for Moving Organ, Breast, and Thoracic Images. Springer, 2018, pp. 215–224

    2018

  31. [38]

    Multi-organ segmentation over partially labeled datasets with multi-scale feature abstraction,

    X. Fang and P. Yan, “Multi-organ segmentation over partially labeled datasets with multi-scale feature abstraction,”IEEE Transactions on Medical Imaging, vol. 39, no. 11, pp. 3619–3629, 2020

    2020

  32. [39]

    Deep learning–based segmentation and quantification in experimental kidney histopathology,

    N. Bouteldja, B. M. Klinkhammer, R. D. B ¨ulow, P. Droste, S. W. Otten, S. F. von Stillfried, J. Moellmann, S. M. Sheehan, R. Korstanje, S. Menzelet al., “Deep learning–based segmentation and quantification in experimental kidney histopathology,”Journal of the American Society...

    2021

  33. [40]

    Med3d: Transfer learning for 3d medical image analysis,

    S. Chen, K. Ma, and Y . Zheng, “Med3d: Transfer learning for 3d medical image analysis,”arXiv preprint arXiv:1904.00625, 2019

    1904 arXiv

  34. [41]

    Dodnet: Learning to segment multi-organ and tumors from multiple partially labeled datasets,

    J. Zhang, Y . Xie, Y . Xia, and C. Shen, “Dodnet: Learning to segment multi-organ and tumors from multiple partially labeled datasets,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 1195–1204

    2021

  35. [43]

    Segmentation of tumour regions for tubule formation assessment on breast cancer histopathology images,

    T. Y . Jing, N. Mustafa, H. Yazid, and K. S. A. Rahman, “Segmentation of tumour regions for tubule formation assessment on breast cancer histopathology images,” inProceedings of the 11th International Con- ference on Robotics, Vision, Signal Processing and Power Applications. ...

    2022

  36. [44]

    Adversarial learning with data selection for cross-domain histopathological breast cancer segmentation,

    Z. Lin, J. Li, Q. Yao, H. Shen, and L. Wan, “Adversarial learning with data selection for cross-domain histopathological breast cancer segmentation,”Multimedia Tools and Applications, vol. 81, no. 4, pp. 5989–6008, 2022

    2022

  37. [45]

    Muscle: Multi-task self-supervised continual learning to pre-train deep models for x-ray images of multiple body parts,

    W. Liao, H. Xiong, Q. Wang, Y . Mo, X. Li, Y . Liu, Z. Chen, S. Huang, and D. Dou, “Muscle: Multi-task self-supervised continual learning to pre-train deep models for x-ray images of multiple body parts,” in International Conference on Medical Image Computing and Computer- Ass...

    2022

  38. [46]

    Universeg: Universal medical image segmentation,

    V . I. Butoi, J. J. G. Ortiz, T. Ma, M. R. Sabuncu, J. Guttag, and A. V . Dalca, “Universeg: Universal medical image segmentation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 21 438–21 451

    2023

  39. [47]

    Continual self-supervised learning: Towards universal multi-modal medical data representation learning,

    Y . Ye, Y . Xie, J. Zhang, Z. Chen, Q. Wu, and Y . Xia, “Continual self-supervised learning: Towards universal multi-modal medical data representation learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 11 114–11 124

    2024

  40. [48]

    Continual segment: Towards a single, unified and non- forgetting continual segmentation model of 143 whole-body organs in ct scans,

    Z. Ji, D. Guo, P. Wang, K. Yan, L. Lu, M. Xu, Q. Wang, J. Ge, M. Gao, X. Yeet al., “Continual segment: Towards a single, unified and non- forgetting continual segmentation model of 143 whole-body organs in ct scans,” inProceedings of the IEEE/CVF International Conference on Co...

    2023

  41. [49]

    Swin unetr: Swin transformers for semantic segmentation of brain tumors in mri images,

    A. Hatamizadeh, V . Nath, Y . Tang, D. Yang, H. R. Roth, and D. Xu, “Swin unetr: Swin transformers for semantic segmentation of brain tumors in mri images,” inInternational MICCAI Brainlesion Workshop. Springer, 2021, pp. 272–284

    2021

  42. [50]

    Efficientsam: Leveraged masked image pretraining for efficient segment anything,

    Y . Xiong, B. Varadarajan, L. Wu, X. Xiang, F. Xiao, C. Zhu, X. Dai, D. Wang, F. Sun, F. Iandolaet al., “Efficientsam: Leveraged masked image pretraining for efficient segment anything,”arXiv preprint arXiv:2312.00863, 2023

    2023 arXiv

  43. [51]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gellyet al., “An image is worth 16x16 words: Transformers for image recognition at scale,”arXiv preprint arXiv:2010.11929, 2020

    2010 arXiv

  44. [52]

    Plop: Learning without forgetting for continual semantic segmentation,

    A. Douillard, Y . Chen, A. Dapogny, and M. Cord, “Plop: Learning without forgetting for continual semantic segmentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 4040–4050

    2021

  45. [53]

    Modeling the background for incremental learning in semantic segmentation,

    F. Cermelli, M. Mancini, S. R. Bulo, E. Ricci, and B. Caputo, “Modeling the background for incremental learning in semantic segmentation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 9233–9242

    2020

  46. [54]

    Incrementer: Transformer for class-incremental semantic segmentation with knowl- edge distillation focusing on old class,

    C. Shang, H. Li, F. Meng, Q. Wu, H. Qiu, and L. Wang, “Incrementer: Transformer for class-incremental semantic segmentation with knowl- edge distillation focusing on old class,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 7214–7224

    2023

  47. [55]

    Comformer: Continual learn- ing in semantic and panoptic segmentation,

    F. Cermelli, M. Cord, and A. Douillard, “Comformer: Continual learn- ing in semantic and panoptic segmentation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 3010–3020

    2023

  48. [56]

    Incremental learning techniques for semantic segmentation,

    U. Michieli and P. Zanuttigh, “Incremental learning techniques for semantic segmentation,” inProceedings of the IEEE/CVF international conference on computer vision workshops, 2019, pp. 0–0

    2019

  49. [57]

    Class similarity weighted knowledge distillation for continual semantic seg- mentation,

    M. H. Phan, S. L. Phung, L. Tran-Thanh, A. Bouzerdoumet al., “Class similarity weighted knowledge distillation for continual semantic seg- mentation,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 16 866–16 875

    2022

  50. [58]

    Towards a general-purpose foundation model for computational pathology,

    R. J. Chen, T. Ding, M. Y . Lu, D. F. Williamson, G. Jaume, B. Chen, A. Zhang, D. Shao, A. H. Song, M. Shabanet al., “Towards a general-purpose foundation model for computational pathology,”Nature Medicine, 2024

    2024

  51. [59]

    A whole-slide foundation model for digital pathology from real-world data,

    H. Xu, N. Usuyama, J. Bagga, S. Zhang, R. Rao, T. Naumann, C. Wong, Z. Gero, J. Gonz ´alez, Y . Guet al., “A whole-slide foundation model for digital pathology from real-world data,”Nature, pp. 1–8, 2024

    2024

Pith tools

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