Pith. sign in

REVIEW 3 major objections 5 minor 19 references

Learning from Similarity Proportion Loss for Classifying Skeletal Muscle Recovery Stages

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A similarity proportion loss trains an ordinal classifier of muscle recovery stages under weak supervision, beating large pre-trained and fine-tuned DINO baselines.

desk verdict A genuinely new ordinal LLP loss that deserves a serious referee, but the reported gain rests on five test WSIs and no ablation isolating the loss. read the letter →

arxiv 2505.04150 v2 pith:CQ6LTZXS submitted 2025-05-07 cs.CV cs.LG

Yu Yamaoka , Weng Ian Chan , Shigeto Seno , Soichiro Fukada , Hideo Matsuda This is my paper · ORCID
classification cs.CVcs.LG
keywords learningfromlabelproportionsordinalclassificationwholeslideimagesskeletalmuscleregenerationsimilarityproportionlossweaklysupervisedfeatureextractorfine-tuningcellmorphology
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

To classify cells in whole-slide images of regenerating skeletal muscle without per-cell labels, the paper proposes OSLSP, an LLP method that uses the date after cardiotoxin injection as weak supervision. The method derives a similarity proportion loss from pairs of bags, comparing the histogram of cosine similarities between instances in two bags with a ground-truth similarity distribution built from the two bags' class proportions and a hand-set linear ordering of the five recovery stages. Because the histogram is computed with Gaussian expansion, the loss is differentiable and can update the feature extractor, not just the classification head. On the authors' test set, OSLSP reaches 46.005% micro accuracy and F1 0.425, compared with 44.442%/0.377 for the pre-trained DINO model and 20.967%/0.217 for a fine-tuned DINO model. The main assumption is that the linear similarity between stages, $\mathrm{sim}(k,k') = 1 - |k-k'|/(K-1)$, correctly encodes the biological order and spacing of recovery.

What carries the argument

The central object is the similarity proportion loss $L_{\mathrm{SimProp}} = D_{\mathrm{KL}}(\hat{P} \parallel P)$, built by grouping same-date instances into bags, computing a differentiable histogram of pairwise cosine similarities between two bags using Gaussian expansion with $\sigma = 0.1$, and forming the ground-truth histogram as $P(\mathrm{sim}(k,k')) = p_k p'_{k'} + p_{k'} p'_k$ (with the diagonal case $k = k'$ using $p_k p'_k$) from the class proportions and the linear class similarity $\mathrm{sim}(k,k') = 1 - |k-k'|/(K-1)$. This loss is differentiable, so gradients reach the backbone, which standard LLP losses cannot do, and the same construction injects the ordinal prior into training.

What would settle it

Compute the class-pair confusion matrix on held-out WSIs for OSLSP and for the pre-trained DINO baseline; the paper's own Fig. 2 suggests OSLSP confuses intact myofibers and ghost fibers more often. If replacing the linear similarity with a squared or exponential spacing, or with a dissimilarity matrix estimated from a few expert-annotated regions, does not change test accuracy and F1, then the linear-ordering prior is not what drives the result.

Watch

Extended reading notes

Core claim

OSLSP's central claim is that label proportions can supervise both the classifier and the feature extractor in an ordinal setting. For two bags with class proportion vectors $\mathbf{p}_d$ and $\mathbf{p}_{d'}$, the ground-truth similarity distribution is formed by pairing classes and weighting each class-pair by the product of its proportions, with class similarity $\mathrm{sim}(k,k') = 1 - |k-k'|/(K-1)$ encoding the biological life cycle of myofibers (intact myofiber, ghost fiber, myoblast, myotube, recovered myofiber). The predicted distribution is a differentiable histogram of scaled cosine similarities between instance features from the two bags, and the KL divergence between the two distributions is the similarity proportion loss. Training with this loss plus a standard proportion loss for the head adapts a ViT-B/8 backbone to muscle morphology while respecting stage order. The paper reports that this combination beats a large pre-trained DINO feature extractor and a fine-tuned DINO model on accuracy, F1, and RMSE, while conceding that intact myofibers and ghost fibers are confused more often than in the pre-trained model because their pre-set similarity is high.

Load-bearing premise

The load-bearing premise is that the hand-set linear class similarity $\mathrm{sim}(k,k') = 1 - |k-k'|/(K-1)$ correctly encodes the biological order and spacing of the five recovery stages, and that the roughly estimated per-date class proportions are accurate enough to serve as ground truth in the supervision signal.

Editorial extensions

If this is right

  • LLP on whole-slide images no longer requires a frozen pre-trained feature extractor; the backbone can be adapted under weak proportion supervision when class order is meaningful.
  • The two-bag similarity distribution construction can be applied to any ordinal or structured label space, such as disease severity scores or developmental time points, when rough proportions per group are available.
  • Because the loss uses only pairwise cosine similarities, it avoids the pseudo-label noise that instance-level MIL methods such as IIB-MIL introduce, while still updating the backbone.
  • The reported RMSE improvement over the pre-trained DINO baseline (2.152 vs. 2.431) indicates that OSLSP makes fewer far-off ordinal errors, which matters for assessing muscle regeneration where confusing adjacent stages has different consequences than confusing distant ones.
  • The confusion between intact myofibers and ghost fibers, which the paper attributes to the high pre-set similarity between those classes, shows that the ordinal prior directly shapes the model's error profile rather than merely improving overall accuracy.

Reading between the lines

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

  • The linear similarity $\mathrm{sim}(k,k') = 1 - |k-k'|/(K-1)$ assumes equal spacing between consecutive stages; a testable extension is to replace it with an exponential or learned kernel, or with a dissimilarity matrix estimated from a few expert-annotated regions, and check whether accuracy and F1 improve.
  • Since the loss is agnostic to how bags are formed, it could be applied to other grouping variables such as subject, batch, or stain in weakly supervised pathology tasks where an ordinal progression is known, not just to days after injection.
  • The reported gain over the pre-trained DINO baseline is modest on a single five-WSI test set, so the more durable contribution may be the demonstration that backbone fine-tuning under LLP is feasible, rather than the exact accuracy difference over one dataset.
  • Combining the similarity proportion loss with a consistency regularizer similar in spirit to LLP-VAT could further stabilize the feature extractor, especially in low-data regimes where the rough manual proportions are noisy.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes OSLSP, an LLP method for classifying five ordinal skeletal muscle recovery stages from cell images extracted from WSIs. The method computes a differentiable similarity-proportion loss between pairs of bags, in which the ground-truth similarity distribution is obtained by combining a hand-set class order with per-date class proportions, and the predicted distribution is obtained from cosine similarities of instance features. OSLSP updates the last block of a pre-trained ViT-B/8 backbone together with the classifier, and the authors report that it achieves 46.005% micro accuracy and 0.425 F1 on five test WSIs, compared with 44.442% accuracy and 0.377 F1 for the pre-trained DINO baseline and 20.967% accuracy and 0.217 F1 for fine-tuned DINO.

Significance. If the reported results were statistically supported, the contribution would be useful for the medical-imaging LLP community: the paper proposes a principled, differentiable way to use ordinal class structure under label-proportion supervision, and the idea of updating a feature extractor through the similarity-proportion loss addresses a real limitation of previous LLP methods for WSIs. The paper deserves credit for anchoring the loss in a concrete biological ordering of recovery stages and for using Gaussian expansion to make the histogram-based loss differentiable. However, the current evidence base is too thin to support the headline claim that OSLSP outperforms DINO-based models: there are only five test WSIs, no error bars or significance tests, no ablation that removes the similarity-proportion term, and the fine-tuned baseline uses a different training recipe.

major comments (3)
  1. [Section 4, Table 2] The evaluation rests on only five test WSIs (one per day), and the reported metrics are point estimates without confidence intervals, error bars, or significance tests. The accuracy gap between OSLSP (46.005%) and pre-trained DINO (44.442%) is 1.56 percentage points; with only five independent test slides, this gap is well within plausible sampling variability. Please report per-slide scores, bootstrap confidence intervals, or a paired significance test, and state the number of cells per slide and per class.
  2. [Section 3.2 and Table 2] There is no ablation that isolates the contribution of L_SimProp. The only condition that updates a feature extractor is a fully fine-tuned DINO using a different training recipe (the official DINO pipeline), and it collapses to 20.967% accuracy. To attribute the gain to the ordinal similarity-proportion loss, add at least a last-block fine-tuning condition with Lprop only (removing L_SimProp) and an OSLSP variant trained with a scrambled or reversed class order. Without these controls, the 1.56-point gain over the pre-trained baseline could be caused by the fine-tuning procedure, data augmentation, or the trained head rather than by the proposed loss.
  3. [Section 3.1, Eqs. (5)-(6)] The ground-truth similarity distribution P is determined by the hand-set linear similarity sim(k,k') = 1 - |k-k'|/(K-1) and by rough manual class proportions p_d; both are treated as fixed prior knowledge. If the linear spacing misrepresents the true morphological distances, or the rough proportions are noisy, the loss will systematically drive the feature extractor toward an incorrect target. Since the RMSE evaluation metric uses the same linear class order, the ordinal evaluation shares that assumption. Please provide a sensitivity analysis (e.g., vary the similarity matrix, sigma, and the proportion estimates) or validate the ordinal prior against an independent annotation.
minor comments (5)
  1. [Section 3.1, Eq. (4)] The text says 'all instance pairs' but the summation in Eq. (4) runs over n = 1,...,N, which suggests only N paired instances rather than all N^2 pairs; please clarify which pairing is used, as this changes the effective training distribution.
  2. [Section 3.1] The phrase 'inter-atomic distances' appears to be a typo for 'instance similarities' or 'pairwise distances'.
  3. [Section 4] The paper does not state the number of instances (cells) per WSI or per class; reporting these numbers is needed to interpret the macro precision, recall, and F1 scores, which can be very volatile with only five test slides.
  4. [Section 4, Table 2 and Section 5] A confusion matrix would help quantify the acknowledged confusion between intact MF and ghost fiber and would make the ordinal error structure visible; the current RMSE alone does not reveal which class pairs are confused.
  5. [Abstract and Section 5] The claim that the model 'outperforms' large-scale pre-trained and fine-tuning models should be tempered to 'on the reported dataset' unless statistical support is added.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: OSLSP is trained against a target built from manual proportions and a hand-set ordinal similarity, but the headline accuracy/F1 comparison is evaluated on held-out expert annotations and is not forced by construction.

full rationale

The paper's derivation chain is self-contained in the relevant sense. The ground-truth similarity PDF P in Eq. 6 is constructed from the per-date class proportions p_d (obtained by rough manual annotation) and the hand-set class similarity sim(k,k')=1-|k-k'|/(K-1) in Eq. 5, and the predicted PDF P_hat is computed from cosine similarities of instance features via Eq. 4; the KL loss in Eq. 1 then updates the feature extractor and head. This is a legitimate supervised objective, not a quantity that is its own input. The central performance claim (46.005% accuracy and F1 0.425 vs. 44.442%/0.377 for pre-trained DINO and 20.967%/0.217 for fine-tuned DINO in Table 2) is evaluated against held-out expert annotations, so it is not a re-statement of the training proportions or of the similarity matrix. The RMSE metric uses the same ordinal class ordering as the training prior, but RMSE is applied to model predictions versus expert labels and is a standard ordinal error measure; this is consistency of the task definition, not circularity. The paper's self-citations are limited to standard tool citations (DINO, Cellpose) and are not load-bearing, and no uniqueness theorem or ansatz is smuggled in via self-citation. Weaknesses in the experimental comparison (no ablation removing L_SimProp, small test set, no error bars) are correctness/evidence concerns, not circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a hand-constructed ordinal similarity prior and roughly annotated proportions, not on a new physical entity. Since OSLSP is a training objective, the ledger contains free parameters and assumptions rather than invented entities.

free parameters (4)
  • Linear ordinal class-similarity coefficients
    Eq. 5 sets sim(k,k')=1-|k-k'|/(K-1), fixing all pairwise class similarities by hand; the ground truth PDF in Eq. 6 is built from these values, so the ordinal target is imposed rather than learned from data.
  • Gaussian expansion width sigma = 0.1
    Eq. 4 uses sigma=0.1 to make the histogram differentiable; no search or justification is given for this value.
  • Number of histogram bins b
    The similarity PDF in Eq. 1 and Eq. 4 depends on the number of bins b, but the paper never specifies b.
  • Per-date class proportions p_d
    Section 3.1 says rough manual class annotation on the training data yields proportions p_d; these serve as the ground truth for both Lprop and the similarity PDF, and any noise in them propagates into the training target.
assumptions (5)
  • domain assumption The five cell morphology stages have a strict linear order with equal ordinal spacing.
    Section 3.1 and Fig. 1(a) motivate Eq. 5; the linear similarity formula treats adjacent stages as equally distant, which may not hold biologically.
  • domain assumption Roughly annotated per-date class proportions are accurate enough to act as weak supervision.
    Section 3.1 states rough manual annotation yields p_d; these proportions define both losses, and their accuracy is never quantified.
  • domain assumption Cosine similarity between feature vectors reflects cell-class similarity.
    Eq. 2 and Eq. 3 assume that learned feature cosine similarity can approximate class similarity; no evidence is given that this holds in feature space.
  • domain assumption DINO ViT-B/8 features provide a suitable initialization for muscle cell images.
    Section 4 uses the DINO ViT-B/8 backbone; the paper argues medical features differ from natural images and therefore fine-tunes, but it still relies on the pretrained initialization.
  • ad hoc to paper The indicator-function histogram can be replaced by Gaussian expansion with sigma=0.1 without materially changing the training signal.
    Eq. 4 replaces Eq. 3 with a hand-chosen sigma=0.1; this is an algorithmic choice introduced for differentiability.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning from Similarity Proportion Loss for Classifying Skeletal Muscle Recovery Stages." pith.science (2026). https://pith.science/paper/CQ6LTZXS

@misc{pith2026250504150,
  author       = {Pith},
  title        = {Pith review of: Learning from Similarity Proportion Loss for Classifying Skeletal Muscle Recovery Stages},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CQ6LTZXS}},
  note         = {Machine review of arXiv:2505.04150}
}
read the original abstract

Evaluating the regeneration process of damaged muscle tissue is a fundamental analysis in muscle research to measure experimental effect sizes and uncover mechanisms behind muscle weakness due to aging and disease. The conventional approach to assessing muscle tissue regeneration involves whole-slide imaging and expert visual inspection of the recovery stages based on the morphological information of cells and fibers. There is a need to replace these tasks with automated methods incorporating machine learning techniques to ensure a quantitative and objective analysis. Given the limited availability of fully labeled data, a possible approach is Learning from Label Proportions (LLP), a weakly supervised learning method using class label proportions. However, current LLP methods have two limitations: (1) they cannot adapt the feature extractor for muscle tissues, and (2) they treat the classes representing recovery stages and cell morphological changes as nominal, resulting in the loss of ordinal information. To address these issues, we propose Ordinal Scale Learning from Similarity Proportion (OSLSP), which uses a similarity proportion loss derived from two bag combinations. OSLSP can update the feature extractor by using class proportion attention to the ordinal scale of the class. Our model with OSLSP outperforms large-scale pre-trained and fine-tuning models in classification tasks of skeletal muscle recovery stages.

Figures

Figures reproduced from arXiv: 2505.04150 by the authors.

Figure 1
Figure 1. Overview of our OSLSP. (a) Similarity of class and morphological changes of cells over time. (b) A pipeline to obtain instance features and class inference results for each cell from WSI. (c) We computed similarity proportion loss using KL divergence between ground truth similarity distribution derived by combining two bags’ propor￾tions and predicted one derived by cosine similarity of each instance in two bags. sh… view at source ↗
Figure 2
Figure 2. Classification results of WSIs for each day. Blue: ghost fiber, Red: intact my￾ofiber, Pink: recovered myofiber, Orange: myotube, Yellow: myoblast. In the manual expert annotations, unannotated white areas indicate uncertain regions. Unlike traditional LLP that treats classes as nominal scales, OSLSP considers the ordinal scale nature of class similarity, given that intact MF and ghost fibers are in similar states i… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

19 extracted references · 7 canonical work pages

  1. [1]

    LLP-Bench: A Large Scale Tabular Benchmark for Learning from Label Proportions

    Brahmbhatt, A., Pokala, M., Saket, R., Raghuveer, A.: Llp-bench: A large scale tabular benchmark for learning from label proportions. arXiv preprint arXiv:2310.10096 (2023). https://doi.org/10.48550/arXiv.2310.10096

    work page Pith review arXiv doi:10.48550/arxiv.2310.10096 2023

  2. [2]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Caron, M., Touvron, H., Misra, I., Jégou, H., Mairal, J., Bojanowski, P., Joulin, A.: Emerging properties in self-supervised vision transformers. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 9650–9660 (2021)

    2021

  3. [3]

    In: International Conference on Computer Vision (ICCV)

    Chan, L., Hosseini, M.S., Rowsell, C., Plataniotis, K.N., Damaskinos, S.: Histoseg- net: Semantic segmentation of histological tissue type in whole slide images. In: International Conference on Computer Vision (ICCV). pp. 10661–10670 (2019). https://doi.org/10.1109/ICCV.2019.01076

    arXiv 2019

  4. [4]

    Skeletal Muscle9 (2019)

    Desgeorges, T., Liot, S., Lyon, S., Bouvière, J., Kemmel, A., Trignol, A., Rousseau, D., Chapuis, B., Gondin, J., Mounier, R., Chazaud, B., Juban, G.: Open-CSAM, a new tool for semi-automated analysis of myofiber cross-sectional area in regener- ating adult skeletal muscle. Skeletal Muscle9 (2019). https://doi.org/10.1186/ s13395-018-0186-6

    work page 2019

  5. [5]

    IEEE Transactions on Image Processing26(6), 2825–2838 (2017)

    Gao, B.B., Xing, C., Xie, C.W., Wu, J., Geng, X.: Deep label distribution learning with label ambiguity. IEEE Transactions on Image Processing26(6), 2825–2838 (2017). https://doi.org/10.1109/TIP.2017.2689998

    arXiv 2017

  6. [6]

    Statistical Methods in Medical Research27 (05 2016)

    Hernández-González, J., Inza, I., Crisol-Ortíz, L., Guembe, M., Iñarra, M., Lozano, J.: Fitting the data from embryo implantation prediction: Learning from label proportions. Statistical Methods in Medical Research27 (05 2016). https://doi. org/10.1177/0962280216651098

    work page doi:10.1177/0962280216651098 2016

  7. [7]

    Kotzias, D., Denil, M., de Freitas, N., Smyth, P.: From group to individual labels usingdeepfeatures.In:ProceedingsoftheACMSIGKDDInternationalConference on Knowledge Discovery and Data Mining. p. 597–606 (2015).https://doi.org/ 10.1145/2783258.2783380

    arXiv 2015

  8. [8]

    In: Medical Image Computing and Computer Assisted Intervention (MICCAI) (2024)

    Matsuo, S., Suehiro, D., Uchida, S., Ito, H., Terada, K., Yoshizawa, A., Bise, R.: Learning from partial label proportions for whole slide image segmentation. In: Medical Image Computing and Computer Assisted Intervention (MICCAI) (2024). https://doi.org/10.48550/arXiv.2405.09041

    arXiv doi:10.48550/arxiv.2405.09041 2024

Show all 19 references
  1. [9]

    In: Proceedings of International Conference on Machine Learning

    Nair, V., Hinton, G.E.: Rectified linear units improve restricted boltzmann ma- chines. In: Proceedings of International Conference on Machine Learning. p. 807–814 (2010) 10 Yu Yamaoka, Weng Ian Chan et al

    2010

  2. [10]

    In: Proceedings of the 25th International Conference on Machine Learning

    Quadrianto, N., Smola, A.J., Caetano, T.S., Le, Q.V.: Estimating labels from la- bel proportions. In: Proceedings of the 25th International Conference on Machine Learning. p. 776–783. ICML ’08, Association for Computing Machinery, New York, NY, USA (2008).https://doi.org/10.11...

    2008

  3. [11]

    In: Medical Image Computing and Computer Assisted Intervention (MICCAI)

    Ren, Q., Zhao, Y., He, B., Wu, B., Mai, S., Xu, F., Huang, Y., He, Y., Huang, J., Yao, J.: Iib-mil: Integrated instance-level and bag-level multiple instances learn- ing with label disambiguation for pathological image analysis. In: Medical Image Computing and Computer Assiste...

    2023 doi

  4. [12]

    Psychological Review65(6), 386 (1958)

    Rosenblatt, F.: The perceptron: a probabilistic model for information storage and organization in the brain. Psychological Review65(6), 386 (1958). https://doi. org/10.1037/h0042519

    1958 doi

  5. [13]

    Nature communica- tions 8(1), 13890 (2017).https://doi.org/10.1038/ncomms13890

    Schütt, K.T., Arbabzadah, F., Chmiela, S., Müller, K.R., Tkatchenko, A.: Quantum-chemical insights from deep tensor neural networks. Nature communica- tions 8(1), 13890 (2017).https://doi.org/10.1038/ncomms13890

    2017 doi

  6. [14]

    In: International Conference on Computer Vision (ICCV)

    Shao, Z., Wang, Y., Chen, Y., Bian, H., Liu, S., Wang, H., Zhang, Y.: Lnpl-mil: Learning from noisy pseudo labels for promoting multiple instance learning in whole slide image. In: International Conference on Computer Vision (ICCV). pp. 21438–21438 (2023). https://doi.org/10.1...

    2023

  7. [15]

    In: Advances in Neural Information Processing Systems (2021)

    Shao, Z., Bian, H., Chen, Y., Wang, Y., Zhang, J., Ji, X., yongbing, z.: Trans- mil: Transformer based correlated multiple instance learning for whole slide im- age classification. In: Advances in Neural Information Processing Systems (2021). https://doi.org/10.48550/arXiv.2106.00908

    2021 doi arXiv

  8. [16]

    Nature Methods18(1), 100–106 (2021)

    Stringer, C., Wang, T., Michaelos, M., Pachitariu, M.: Cellpose: a generalist algo- rithm for cellular segmentation. Nature Methods18(1), 100–106 (2021). https: //doi.org/10.1038/s41592-020-01018-x

    2021 doi

  9. [17]

    Scientific Reports 9(1), 16 (2019)

    Tanaka, Y., Kita, S., Nishizawa, H., Fukuda, S., Fujishima, Y., Obata, Y., Nagao, H., Masuda, S., Nakamura, Y., Shimizu, Y., Ryohei, M., Tomoaki, N., Tohru, F., Barbara, R., So-Ichiro, F., Norikazu, M., Iichiro, S.: Adiponectin promotes muscle regeneration through binding to t...

    2019 doi

  10. [18]

    In: Asian Conference on Machine Learning

    Tsai, K.H., Lin, H.T.: Learning from label proportions with consistency regular- ization. In: Asian Conference on Machine Learning. pp. 513–528. PMLR (2020). https://doi.org/10.48550/arXiv.1910.13188

    2020 doi arXiv

  11. [19]

    IEEE Journal of Biomedical and Health Informatics 25(9), 3288–3299 (2021).https://doi.org/10.1109/jbhi.2021.3071276

    Ye, Q., Zhang, Q., Tian, Y., Zhou, T., Ge, H., Wu, J., Lu, N., Bai, X., Liang, T., Li, J.: Method of tumor pathological micronecrosis quantification via deep learning from label fuzzy proportions. IEEE Journal of Biomedical and Health Informatics 25(9), 3288–3299 (2021).https:...

    2021

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.