Pith. sign in

REVIEW 4 major objections 3 minor 19 references

Vision Transformers for Kidney Stone Image Classification: A Comparative Study with CNNs

T0 review · 4 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a Vision Transformer pretrained on ImageNet-21k classifies kidney stone subtypes from endoscopic images more accurately than a ResNet50 CNN pretrained on ImageNet-1k, with the largest margin on the hardest section-patc

desk verdict Useful benchmark, but the ViT superiority claim is a pretraining-scale artifact until a ResNet50-21k baseline is added. read the letter →

arxiv 2508.13461 v2 pith:4HTGEZGX submitted 2025-08-19 cs.CV cs.LG

Ivan Reyes-Amezcua , Francisco Lopez-Tiro , Clement Larose , Andres Mendez-Vazquez , Gilberto Ochoa-Ruiz , Christian Daul This is my paper · ORCID
classification cs.CVcs.LG
keywords VisionTransformerkidneystoneclassificationendoscopicimagingCNNbaselineImageNet-21kpretrainingself-attentionmorpho-constitutionalanalysisexvivodatasets
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's central claim is that a standard Vision Transformer (ViT-base) pretrained on ImageNet-21k classifies kidney stone subtypes from endoscopic images more accurately than a ResNet50 CNN pretrained on ImageNet-1k. On the hardest subset—section patches from ureteroscope images—the reported numbers are 95.7% versus 64.5% accuracy, and on mixed-view CCD images 87.1% versus 78.4%. The authors attribute the advantage to self-attention's global patch dependencies and to tighter, more separable feature embeddings. If correct, automated visual stone recognition during ureteroscopy becomes a scalable route to personalized treatment and recurrence prevention.

What carries the argument

The central object is the ViT-base model with the standard patch-and-self-attention pipeline: the image is cut into fixed-size patches, each patch is linearly embedded, and multi-head self-attention re-weights every patch against every other patch before classification. It supplies global context in a single pass, which the paper says is what lets the model form compact, well-separated clusters under lighting and view variability. The ImageNet-21k pretraining is a second, equally load-bearing component: it is the version that yields the headline numbers, while the same ViT pretrained on ImageNet-1k or from scratch performs much worse.

What would settle it

Run the identical split and training protocol with a ResNet50 pretrained on ImageNet-21k (or both models initialized from the same 21k checkpoint) and compare on all six dataset-by-view subsets. If the 21k-pretrained CNN matches or exceeds the ViT, the architectural conclusion is falsified and the observed gap was pretraining data, not self-attention.

Watch

Extended reading notes

Core claim

The authors' positive claim is that transformer-based models, specifically ViT-base pretrained on ImageNet-21k, classify kidney stone subtypes from endoscopic images more accurately and with better feature structure than the ResNet50 CNN baseline. The support consists of an ablation across two datasets and three view types, embedding visualizations with silhouette scores, intra-class distance tables, confusion matrices, and a qualitative sample comparison. The authors attribute the results to self-attention's global patch-level context, which lets the model separate visually ambiguous stone subtypes that trip up convolutional features.

Load-bearing premise

The central comparison is not architecture-only: the ViT uses ImageNet-21k pretraining while the ResNet uses ImageNet-1k, and on one surface-view subset the ResNet already leads; the paper's conclusion assumes that pretraining scale is not doing most of the work.

Editorial extensions

If this is right

  • Automated stone subtype recognition could be built around a single pretrained ViT rather than task-specific CNN pipelines.
  • The large margins on hard Section patches suggest transformer features hold up where lighting and view artifacts break CNN embeddings.
  • Compact ViT embeddings, with lower intra-class distances and higher silhouette scores, imply better retrievability of visually similar stone subtypes.
  • If deployed, the system could give urologists an immediate visual read during ureteroscopy, reducing the need to extract fragments for infrared laboratory analysis.

Reading between the lines

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

  • Editorial inference: the headline gap is as much about pretraining scale as about architecture. ViT-base pretrained on ImageNet-1k trails ResNet50-ImageNet-1k on five of the six subsets in Table 3, so the 21k pretraining is doing consequential work.
  • Editorial inference: the direct falsifying experiment is to train a ResNet50 with ImageNet-21k pretraining under the same protocol and compare; if the CNN matches or exceeds the ViT, the observed advantage is a pretraining-data effect rather than a self-attention effect.
  • Editorial inference: the paper's 'consistently outperformed' phrasing is stronger than the table shows—on Dataset A surface views the ViT-21k scores 76.8% versus ResNet50's 79.6%—so the claim depends on which subsets are highlighted.
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

4 major / 3 minor

Summary. The paper compares Vision Transformer (ViT) and CNN (ResNet) models for kidney stone subtype classification on two ex vivo image datasets, one from CCD cameras and one from flexible ureteroscopes. Across six view subsets, the authors report accuracy, F1, precision, and recall for ResNet50/ResNet152 and ViT-base/ViT-small under several pretraining conditions (ImageNet-1k, ImageNet-21k, from scratch), together with feature-embedding analyses based on t-SNE, silhouette scores, and intra-class distances. The central claim is that ViT architectures—particularly ViT-base pretrained on ImageNet-21k—provide superior classification performance and more discriminative embeddings than CNNs.

Significance. If the claim were established under controlled conditions, the result would be practically relevant for automated endoscopic kidney stone recognition, and the paper's combination of a multi-subset benchmark with an ablation over pretraining is a useful empirical contribution. I see no circularity: the headline numbers are measured outcomes on held-out test images. The main weakness is that the architecture comparison is confounded with pretraining data scale: the paper compares ViT pretrained on ImageNet-21k against ResNet50 pretrained on ImageNet-1k, and its own matched-pretraining rows largely contradict the superiority claim. The paper also reports headline abstract numbers that differ from the corresponding entries in Table 3. The strengths are the breadth of the benchmark and the inclusion of several quantitative embedding diagnostics, but the central comparative conclusion needs additional controlled experiments or a substantially more cautious framing.

major comments (4)
  1. [Table 3] The paper's central claim—that ViT architectures provide superior classification performance (Sec. 7)—rests on comparing ViT-base pretrained on ImageNet-21k with ResNet50 pretrained on ImageNet-1k. This confounds architecture with pretraining data scale. In the matched ImageNet-1k rows, ViT-base is worse than ResNet50 on five of six subsets (A MIX 0.624 vs 0.784; A SEC 0.657 vs 0.809; A SUR 0.492 vs 0.796; B MIX 0.769 vs 0.850; B SUR 0.704 vs 0.864), and better only on B SEC (0.739 vs 0.645). An ImageNet-21k-pretrained CNN baseline (e.g., ResNet50-21k) is needed before attributing the improvement to the transformer architecture; otherwise the results support a pretraining-scale effect, not an architecture effect.
  2. [Abstract vs. Table 3] The headline examples in the abstract do not match the reported table. The abstract states 87.1% accuracy for the CCD mixed-view subset and 95.2% accuracy/95.1% F1 for the endoscopic section subset, but Table 3 reports 0.890 for A MIX (ViT-base/21k) and 0.957/0.958 for B SEC. If these figures come from a different configuration or table, that configuration must be cited; as printed, the key numbers are not reproducible from the paper's main table.
  3. [Abstract and Sec. 7] Even under the authors' chosen (unmatched) comparison, the statement that ViT 'consistently outperformed' ResNet50 is contradicted by the A SUR row: ResNet50/ImageNet-1k achieves 0.796 accuracy vs 0.768 for ViT-base/ImageNet-21k. The conclusions should be qualified to name the subsets and pretraining conditions for which ViT wins, and should not claim general superiority.
  4. [Sec. 6.2, Fig. 5, Table 4] The embedding comparison ('same experimental conditions') is also affected by the pretraining confound. If the ViT features come from the ImageNet-21k model and the CNN features from the ImageNet-1k model, the higher silhouette scores and lower intra-class distances may reflect pretraining scale rather than architecture. Please specify which rows of Table 3 are used, and either repeat the embedding analysis with matched pretraining or discuss the confound explicitly.
minor comments (3)
  1. [Sec. 1.2] The statement that ViTs 'require less data for training because their self-attention mechanism efficiently captures global contextual relationships' is not supported and appears contradicted by Table 3: from-scratch ViT-base attains 0.209--0.505 across subsets, whereas from-scratch ResNet50 attains 0.583--0.773. Please remove or substantiate this claim.
  2. [Table 3] To assess whether the reported differences are stable, please add confidence intervals, repeated-run variance, or at least the number of test images per subset. The current point estimates, especially for small subsets, do not by themselves establish 'consistent' superiority.
  3. [Table 3 header] The table header appears garbled (e.g., 'Subv敲獩潮'). Please ensure the final PDF typesets the header correctly.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the ViT-vs-CNN claim is an empirical benchmark, with a non-circular pretraining confound.

full rationale

The paper's central claim is an empirical benchmark result: it reports test-set accuracy, F1, precision, and recall for ViT and ResNet variants trained on two ex vivo datasets. No parameter is fitted to a subset of the test data and then renamed as a prediction; the headline numbers in Table 3 are measured on held-out images and are not constructed from the inputs by any equation. The datasets and some baselines come from the authors' prior work (refs 6, 13, 14, 15), but self-supplied data are not circular because the test metrics are external outcomes of trained models. The closest concern is the comparison ViT-21k vs ResNet50-1k, which confounds architecture with pretraining scale; at matched ImageNet-1k pretraining, ViT loses on five of six subsets, and A SUR contradicts the abstract's 'consistently outperformed.' That is an internal-validity/external-validity problem, not a derivation that reduces to its own inputs. No uniqueness theorem, ansatz-by-citation, or renamed known result appears. The ablation study actually provides the counter-evidence internally, so the claim is not forced by construction.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The paper's central comparison leans on standard transfer-learning assumptions and on the group's own prior datasets and baselines. No exotic free parameters or invented entities appear; the main ledger issue is that a hand made asymmetry (ViT gets ImageNet-21k, CNNs do not) does much of the work that the prose attributes to the architecture.

free parameters (2)
  • Training hyperparameters (learning rate, epochs, batch size, augmentation)
    Section 4 (training setup) is absent from the extract, so the hand chosen values that determine the reported accuracies are unverifiable from this text. Any imbalance in training budget between the ViT and ResNet runs could explain part of the gap.
  • ImageNet-21k pretraining applied only to ViT
    The decision to give ViT a 21k-class pretrained initialization while CNN baselines use ImageNet-1k or none is a hand made design choice that conflates architecture with pretraining scale (Table 3). It does much of the work that the prose attributes to the architecture.
assumptions (3)
  • domain assumption ImageNet-pretrained features transfer to ex vivo endoscopic kidney stone images
    Invoked in the transfer learning setup described in Sections 3 and 4; the entire comparison assumes ImageNet-1k/21k pretraining helps or at least does not hurt the stone classifiers. The from-scratch rows in Table 3 provide a partial internal check, but transfer validity is assumed.
  • domain assumption Ground-truth stone subtypes (Daudon classification) in Datasets A and B are correct
    Labels follow the MCA taxonomy (refs 2, 3, 4); label fidelity is assumed, not verified in this paper, and errors in the reference standard would propagate to all accuracy numbers.
  • domain assumption Train/test patches are drawn IID with no leakage between overlapping regions of the same stone image
    Patch extraction (Section 2.1) amplifies the sample count; if patches from the same stone appear in both train and test splits, accuracies are inflated. The split procedure is not visible in the provided extract.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Vision Transformers for Kidney Stone Image Classification: A Comparative Study with CNNs." pith.science (2026). https://pith.science/paper/4HTGEZGX

@misc{pith2026250813461,
  author       = {Pith},
  title        = {Pith review of: Vision Transformers for Kidney Stone Image Classification: A Comparative Study with CNNs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4HTGEZGX}},
  note         = {Machine review of arXiv:2508.13461}
}
read the original abstract

Kidney stone classification from endoscopic images is critical for personalized treatment and recurrence prevention. While convolutional neural networks (CNNs) have shown promise in this task, their limited ability to capture long-range dependencies can hinder performance under variable imaging conditions. This study presents a comparative analysis between Vision Transformers (ViTs) and CNN-based models, evaluating their performance on two ex vivo datasets comprising CCD camera and flexible ureteroscope images. The ViT-base model pretrained on ImageNet-21k consistently outperformed a ResNet50 baseline across multiple imaging conditions. For instance, in the most visually complex subset (Section patches from endoscopic images), the ViT model achieved 95.2% accuracy and 95.1% F1-score, compared to 64.5% and 59.3% with ResNet50. In the mixed-view subset from CCD-camera images, ViT reached 87.1% accuracy versus 78.4% with CNN. These improvements extend across precision and recall as well. The results demonstrate that ViT-based architectures provide superior classification performance and offer a scalable alternative to conventional CNNs for kidney stone image analysis.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

19 extracted references · 17 canonical work pages

  1. [1]

    npj Digital Medicine5(1), 184 (2022) 3

    Ali, S.: Where do we stand in ai for endoscopic image analysis? deciphering gaps and future directions. npj Digital Medicine5(1), 184 (2022) 3

    work page 2022

  2. [2]

    European Urology Focus 7(1), 13–21 (2021) 2, 4, 5, 8 Kidney Stone Recognition with ViT 17

    Corrales, M., Doizi, S., Barghouthy, Y., Traxer, O., Daudon, M.: Classification of stones according to michel daudon: a narrative review. European Urology Focus 7(1), 13–21 (2021) 2, 4, 5, 8 Kidney Stone Recognition with ViT 17

    work page 2021

  3. [3]

    Comptes Rendus Chimie 19(11-12), 1470–1491 (2016) 2

    Daudon, M., Dessombz, A., Frochot, V., Letavernier, E., Haymann, J.P., Jungers, P., Bazin, D.: Comprehensive morpho-constitutional analysis of urinary stones im- proves etiological diagnosis and therapeutic strategy of nephrolithiasis. Comptes Rendus Chimie 19(11-12), 1470–1491 (2016) 2

    work page 2016

  4. [4]

    Nephron Physiology98(2), p31–p36 (2004) 1

    Daudon,M.,Jungers,P.:Clinicalvalueofcrystalluriaandquantitativemorphocon- stitutional analysis of urinary calculi. Nephron Physiology98(2), p31–p36 (2004) 1

    work page 2004

  5. [5]

    arXiv preprint arXiv:2010.11929 (2020) 6, 7, 9

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., et al.: An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929 (2020) 6, 7, 9

    arXiv 2010

  6. [6]

    BJU international130(6), 786–798 (2022) 4, 5, 8

    El Beze, J., Mazeaud, C., Daul, C., Ochoa-Ruiz, G., Daudon, M., Eschwège, P., Hubert, J.: Evaluation and understanding of automated urinary stone recognition methods. BJU international130(6), 786–798 (2022) 4, 5, 8

    work page 2022

  7. [7]

    survey results practice committee of the afu lithiasis completed in 2011

    Estrade, V., Bensalah, K., Bringer, J., Chabannes, E., Carpentier, X., Conort, P., Denis, E., Dore, B., Gautier, J., Hadjadj, H., et al.: Place of the flexible ureterorenoscopy first choice for the treatment of kidney stones. survey results practice committee of the afu lithiasis completed in 2011. Progres en Urologie: Journal de L’association Francaise D...

    work page 2011

  8. [8]

    PROGRES EN UROLO- GIE 27(2), F26–F35 (2017) 1, 3

    Estrade, V., Daudon, M., Traxer, O., Meria, P.: Why should urologist recognize uri- nary stone and how? the basis of endoscopic recognition. PROGRES EN UROLO- GIE 27(2), F26–F35 (2017) 1, 3

    work page 2017

Show all 19 references
  1. [9]

    BJU international128(3), 319–330 (2021) 3

    Estrade, V., Denis de Senneville, B., Meria, P., Almeras, C., Bladou, F., Bernhard, J.C., Robert, G., Traxer, O., Daudon, M.: Toward improved endoscopic examina- tion of urinary stones: a concordance study between endoscopic digital pictures vs microscopy. BJU international128...

    2021

  2. [10]

    arXiv preprint arXiv:2408.05952 (2024) 3

    Habib, G., Singh, D., Malik, I.A., Lall, B.: Optimizing vision transformers with data-free knowledge transfer. arXiv preprint arXiv:2408.05952 (2024) 3

    2024 arXiv

  3. [11]

    Scientific Reports 14(1), 12567 (2024) 3

    Halder, A., Gharami, S., Sadhu, P., Singh, P.K., Woźniak, M., Ijaz, M.F.: Imple- menting vision transformer for classifying 2d biomedical images. Scientific Reports 14(1), 12567 (2024) 3

    2024

  4. [12]

    Khan, A.H., Imran, S., Talati, J., Jafri, L.: Fourier transform infrared spectroscopy foranalysisofkidneystones.InvestigativeandClinicalUrology 59(1),32–37(2018) 2

    2018

  5. [13]

    In: 2021 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC)

    Lopez, F., Varelo, A., Hinojosa, O., Mendez, M., Trinh, D.H., ElBeze, Y., Hubert, J., Estrade, V., Gonzalez, M., Ochoa, G., et al.: Assessing deep learning methods for the identification of kidney stones in endoscopic images. In: 2021 43rd Annual International Conference of th...

    2021

  6. [14]

    IEEE Access12, 10736–10759 (2024) 3, 6

    Lopez-Tiro, F., Estrade, V., Hubert, J., Flores-Araiza, D., Gonzalez-Mendoza, M., Ochoa, G., Daul, C.: On the in vivo recognition of kidney stones using machine learning. IEEE Access12, 10736–10759 (2024) 3, 6

    2024

  7. [15]

    Lopez-Tiro, F., Flores-Araiza, D., Betancur-Rengifo, J.P., Reyes-Amezcua, I., Hu- bert, J., Ochoa-Ruiz, G., Daul, C.: Boosting kidney stone identification in endo- scopicimagesusingtwo-steptransferlearning.In:Mexicaninternationalconference on artificial intelligence. pp. 131–1...

    2023

  8. [16]

    arXiv preprint arXiv:1711.05101 (2017) 7 18 I

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2017) 7 18 I. Reyes et al

    2017 arXiv

  9. [17]

    Journal of Endourology35(6), 931– 936 (2021) 3

    Taratkin, M., Laukhtina, E., Singla, N., Tarasov, A., Alekseeva, T., Enikeev, M., Enikeev, D.: How lasers ablate stones: in vitro study of laser lithotripsy (ho: Yag and tm-fiber lasers) in different environments. Journal of Endourology35(6), 931– 936 (2021) 3

    2021

  10. [18]

    BioMedical Engineering OnLine23(1), 27 (2024) 3

    Yin, Y., Tang, Z., Weng, H.: Application of visual transformer in renal image analysis. BioMedical Engineering OnLine23(1), 27 (2024) 3

    2024

  11. [19]

    Journal of clinical medicine research 11(3), 208 (2019) 1

    Zeng, J., Wang, S., Zhong, L., Huang, Z., Zeng, Y., Zheng, D., Zou, W., Lai, H.: A retrospective study of kidney stone recurrence in adults. Journal of clinical medicine research 11(3), 208 (2019) 1

    2019

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.