Pith. sign in

REVIEW 4 major objections 6 minor 22 references

Transformer-Based Explainable Deep Learning for Breast Cancer Detection in Mammography: The MammoFormer Framework

T0 review · 4 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read MammoFormer claims transformer models can match or beat CNNs on mammograms when each architecture is paired with the right image enhancement, and packages the pairing with explainability into a clinical screening framework.

desk verdict A useful architecture–enhancement matrix undermined by an undocumented evaluation protocol and an overclaim about clinical deployment. read the letter →

arxiv 2508.06137 v1 pith:ECGZS2K7 submitted 2025-08-08 eess.IV cs.CV

classification eess.IVcs.CV
keywords BreastCancerDeepLearningMammographyExplainableAI(XAI)VisionTransformersSwinTransformerCBIS-DDSMFeatureEnhancement
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 transformer models do not intrinsically underperform convolutional networks on mammography; they only appear to when they are fed the wrong image representation. On the authors' evaluation set, the same Vision Transformer moves from 54.3% to 99.0% accuracy depending on preprocessing, Swin Transformer moves from 51.7% to 96.3%, and HOG is the most reliable enhancement overall at 98.4% average accuracy, nearly matching the 99.9% of CNN-based models. MammoFormer is the packaging of this finding: a tiered clinical framework with ResNet-50 on original images as the primary screen, ViT and Swin as global-context validators on contrast- and gradient-enhanced images, weighted voting between them, and five XAI methods for human review. The reason to care is that transformer global-context modeling plus explainability would remove two barriers—long-range dependency failure and black-box opacity—that currently block AI mammography from clinical adoption.

What carries the argument

The load-bearing mechanism is architecture–enhancement pairing: each model family is matched to the preprocessing that exposes the features its mechanism can use. HOG (histogram of oriented gradients) emphasizes edge and gradient structure, which suits Swin's shifted-window, patch-based attention and gives the reported 13.0% improvement; AHE (adaptive histogram equalization) normalizes dense-tissue contrast, which suits ViT's global patch comparison and restores it to 98.3%. Around that pairing, MammoFormer wraps a weighted-voting ensemble—ResNet-50 on original images as the high-accuracy primary tier, ViT and Swin as complementary contextual validators—and five XAI attributions (Integrated

What would settle it

Re-run the evaluation on CBIS-DDSM with a strict patient-exclusive split (no images of the same participant in both training and test) and with every synthetic image removed from the test set, then compare ViT+AHE and Swin+HOG against ResNet-50. If the transformer accuracies drop below the reported 98.3% and 96.3% while CNN accuracy stays near 99.9%, or if the ordering inverts, the paper's central claim that suitable enhancement makes transformers match CNNs is falsified. The simpler pilot: report accuracy on the real-image-only test subset and compare it with the headline numbers.

Watch

Extended reading notes

Core claim

On its own terms, MammoFormer claims to close the CNN-transformer gap in mammography by systematic architecture-specific enhancement. It evaluates seven architectures (custom CNN, ResNet-50, ViT, Swin, DenseTransformer, ConvMixer, ConvNeXt) under four preprocessings (original, negative, AHE, HOG) on a balanced CBIS-DDSM subset enriched with synthetic images. The finding is that pure transformers are highly preprocessing-sensitive—ViT ranges from 54.3% to 99.0%, Swin from 51.7% to 96.3%—while CNN-based models stay at 99.9% across settings. Matching enhancement to architecture gives ViT 98.3% with AHE and Swin 96.3% with HOG, a 13.0% improvement; HOG is the most reliable overall at 98.4% avera

Load-bearing premise

The load-bearing premise is that the balanced, synthetically augmented mammogram set used for evaluation—including the 'artificial testing samples' the authors created and without a documented patient-level train/test split—represents real screening conditions closely enough that the reported 96–99.9% accuracies reflect genuine model ability rather than the way the test set was assembled.

Editorial extensions

If this is right

  • A screening workflow can put ResNet-50 on original images first (99.9% accuracy), with ViT and Swin validators triggered only for ambiguous cases—this is the deployment MammoFormer describes.
  • Transformer models on mammograms no longer need to be written off: on the paper's data, preprocessing alone moves ViT from 54.3% to 99.0% and Swin from 51.7% to 96.3%.
  • HOG is the most generally reliable preprocessing (98.4% average accuracy), so gradient-based input representations should be a candidate for any transformer-based medical image pipeline.
  • Multi-perspective explainability, with Occlusion best aligned to radiologist workflow, is integral rather than optional to the framework's clinical claims.
  • The design rule that emerges is architecture-specific enhancement pairing rather than applying one preprocessing to all models.

Reading between the lines

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

  • Beyond the paper, the reported 44.7-point range in ViT accuracy across preprocessings implies that any published transformer-vs-CNN comparison on medical images is only interpretable when the enhancement protocol is fixed and reported; preprocessing effects can swamp architectural differences.
  • Because the paper notes it 'developed artificial testing samples' and that one participant can carry multiple patient IDs, a patient-exclusive split with synthetic images kept out of the test set is the natural next experiment before the clinical claims are taken at face value—this is the paper's own external-validation call made concrete.
  • The architecture–enhancement matching principle could extend to ultrasound and MRI as the authors propose, but the specific pairings (HOG for Swin, AHE for ViT) would need to be re-derived per modality rather than transferred unchanged.
  • If the numbers survive external validation, the tiered ensemble suggests a cost-conscious clinical workflow: high-confidence CNN reads pass through, and transformer+XAI resources are spent only on disagreement or ambiguity—an allocation testable for sensitivity, specificity, and radiologist workload.
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 / 6 minor

Summary. The paper proposes 'MammoFormer,' a framework for binary benign/malignant classification of mammograms that combines seven CNN, transformer, and hybrid architectures with four image preprocessing enhancements (original, negative, AHE, HOG) and five XAI techniques. The central claim is that architecture-specific feature enhancement enables transformer models to match or exceed CNN accuracy (ViT with AHE reaching 98.3%, Swin with HOG gaining 13.0%), and that the resulting weighted-voting ensemble is clinically deployable. Evaluation uses CBIS-DDSM images balanced to 1,000 images per class, with Table 4 reporting accuracies/precision/recall/F1 for each architecture-enhancement combination.

Significance. If the reported results were valid, the paper would offer a useful systematic comparison of architecture-enhancement pairings and a multi-perspective XAI workflow for mammography, and it would provide evidence that transformers can approach CNN performance with appropriate preprocessing. The explicit comparison of seven architectures under four preprocessing settings and the inclusion of five XAI methods are strengths, as is the internal consistency of the headline numbers in Table 4 with the abstract. However, the significance is conditional on the evaluation protocol, which is not documented in a way that supports the central claims; the stated use of artificial testing samples and the absence of a patient-level split make the reported accuracies insufficient to support conclusions about real mammography or clinical deployability.

major comments (4)
  1. [§3.2 Dataset and §4 Results] The evaluation set is not demonstrably a real mammography test set. Section 3.2 states that 'Initial image technical issues prompted our team to develop artificial testing samples during preprocessing' and that the final dataset was balanced to 1,000 images per class via augmentation and synthetic generation. No train/test split or patient-level grouping is documented; CBIS-DDSM has only 1,566 participants but 6,775 studies, so random image-level partitions can leak views from the same patient into both training and test. If synthetic or augmented images appear in the test set, the 99.9% CNN and 96.3–99.0% transformer accuracies in Table 4 do not measure performance on real mammograms. This directly undermines the abstract's central claim that transformers with suitable enhancements equal or beat CNNs on mammography.
  2. [§4 Result Discussion / Table 4] The text claims 'Table 4 also reports standard deviations, which remain below 0.3% for CNN-based and hybrid models across all enhancements,' but Table 4 contains no standard deviations, confidence intervals, or number of repeated runs; it reports only point estimates. This unquantified robustness claim is load-bearing for recommending specific model–enhancement pairs, especially where differences are small (e.g., ViT AHE 98.3 vs. HOG 99.0; CNN/ResNet at 99.9). The paper also does not state how many independent training runs were performed.
  3. [§3.9 Ethics and Clinical Integration vs. Abstract] The abstract and §1 claim MammoFormer is 'a clinically deployable ensemble system,' but §3.9 states that 'external validation on diverse datasets before any clinical deployment' is required, and §5 lists clinical validation as future work. This is an internal contradiction. The current evaluation on a single curated dataset with artificial samples cannot support the deployability claim, and the manuscript itself concedes the missing external validation.
  4. [§3.2 Synthetic image realism] The paper asserts that synthetic mammogram images 'accurately mimic' the distinguishing characteristics of benign and malignant presentations, but no fidelity metrics, radiologist evaluation, or distributional comparison are provided. If synthetic images are present in the training set and possibly the test set, the burden is on the authors to show that they are representative of real mammograms; otherwise, the reported accuracies may reflect the synthetic distribution rather than clinical data.
minor comments (6)
  1. [Abstract vs. §5 Conclusion] The abstract highlights ViT with AHE at 98.3%, while the conclusion highlights ViT with HOG at 99.0%. Both values appear in Table 4, but the paper should state which pairing is recommended and why, to avoid apparent inconsistency.
  2. [§4 Result Discussion] The phrase 'a 83.3% accuracy increase for Swin Transformer' is confusing; it likely means the absolute accuracy with HOG is 83.3%? Actually Table 4 shows Swin HOG 96.3% and the increase over original is 13.0 percentage points. The sentence should be corrected (e.g., '96.3% accuracy, a 13.0-point improvement over original').
  3. [§2.1 and §2.2] Sections 2.1 and 2.2 contain nearly identical paragraphs on HATNet, Abimouloud et al., and TEBLS, with duplicated text. One should be removed or condensed.
  4. [References] References [11] and [16] are the same Lee et al. paper (Transformer-based Deep Neural Network for Breast Cancer Classification on Digital Breast Tomosynthesis Images), and reference [4] has an incomplete author list ('Hussain, et al.' with 'S. Hussain' in text). Please deduplicate and complete the entries.
  5. [§3.7, Eq. (4)] The Integrated Gradients equation is malformed in the rendering; the integral and gradient notation need to be typeset correctly.
  6. [§4 Aggregated results] The aggregated average accuracies (95.6% for original, 92.2% for negative, 92.0% for AHE) are not reproducible from Table 4 if all seven models are averaged; the paper should state which subset of models is included in each average, or the values should be recalculated.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation; only minor non-load-bearing self-citations.

full rationale

Walked the claimed derivation chain. MammoFormer's central claims—that transformer models can reach CNN-comparable accuracy with suitable preprocessing and that a weighted ensemble plus XAI provides a deployable workflow—are supported by the paper's own Table 4 measurements on CBIS-DDSM, not by a mathematical derivation from the definitions. No parameter is fitted to a target and then reported as a prediction; no uniqueness theorem or ansatz is imported from the authors' prior work. The self-citations (refs 18-21) appear in Section 2.4 and the Literature Gap as contextual parallels, e.g., "The research team achieved a text-guided synthesis accuracy level of 93% using Vision Transformers as reported in [20], which approaches MammoFormer 98.3% performance level"; they are not load-bearing and deleting them would not alter any reported result. Section 3.2's statement that "Initial image technical issues prompted our team to develop artificial testing samples during preprocessing" and Section 3.9's admission that "external validation on diverse datasets before any clinical deployment" is needed are internal-validity limitations for the clinical deployability claim, but they are not circularity: the benchmark numbers are not equivalent to their inputs by construction. The score is 2 rather than 0 only because of several minor self-citations in the related-work narrative; none of them carries a derivation step.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The paper has no mathematical derivation; its central claim rests on an empirical benchmark. The free parameters are hand-chosen hyperparameters and unspecified preprocessing settings. The axioms are the unstated assumptions about data quality, patient-level splitting, and the clinical meaning of accuracy. Because code and data are absent, none of these assumptions can be checked.

free parameters (8)
  • Initial learning rate = 0.001
    Chosen by hand in Section 3.5, Equation 3; no sensitivity analysis.
  • Learning rate decay factor and step size = gamma = 0.1, step = 7 epochs
    Set in Section 3.5 without justification or ablation.
  • Number of training epochs = 10
    Set in Section 3.5; very short for transformer fine-tuning and no convergence curves are shown.
  • Input image resolution = 224 x 224 RGB
    Normalization choice in Section 3.3; no multi-resolution comparison.
  • Class-balanced dataset size = 1,000 images per class
    Composed via augmentation and synthetic generation in Section 3.2; the balance choice affects all reported metrics.
  • HOG parameters = not specified
    Orientations, cell size, and block normalization are not given in Section 3.3, so HOG results are not reproducible.
  • AHE parameters = not specified
    Clip limit and kernel size are not given in Section 3.3.
  • Ensemble weighted voting thresholds = not specified
    Section 3.6 describes dynamic weighted voting and a divergence threshold but gives no values or procedure.
assumptions (5)
  • domain assumption CBIS-DDSM ground truth labels are accurate for benign/malignant classification.
    The paper uses these labels as ground truth in Section 3.2 without verifying them against pathology or radiology review.
  • ad hoc to paper Synthetic mammogram images accurately mimic the distribution of real mammograms.
    Synthetic images are introduced in Sections 3.2 and 3.3 to address data scarcity, but no fidelity or clinical validity check is provided.
  • domain assumption The train/test split separates images at the patient level and excludes synthetic images from the test set.
    No split is described; the paper mentions artificial testing samples, creating a risk of leakage from synthetic data or same-patient images.
  • domain assumption Pretrained model weights provide a valid starting point for mammography transfer learning.
    All architectures are used as pretrained models in Section 3.4 without evidence that pretraining on ImageNet helps mammography.
  • domain assumption Accuracy on a balanced test set approximates clinical utility.
    Section 3.8 equates high accuracy with diagnostic value; no prevalence, recall at fixed specificity, or clinical workflow metrics are reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Transformer-Based Explainable Deep Learning for Breast Cancer Detection in Mammography: The MammoFormer Framework." pith.science (2026). https://pith.science/paper/ECGZS2K7

@misc{pith2026250806137,
  author       = {Pith},
  title        = {Pith review of: Transformer-Based Explainable Deep Learning for Breast Cancer Detection in Mammography: The MammoFormer Framework},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ECGZS2K7}},
  note         = {Machine review of arXiv:2508.06137}
}
read the original abstract

Breast cancer detection through mammography interpretation remains difficult because of the minimal nature of abnormalities that experts need to identify alongside the variable interpretations between readers. The potential of CNNs for medical image analysis faces two limitations: they fail to process both local information and wide contextual data adequately, and do not provide explainable AI (XAI) operations that doctors need to accept them in clinics. The researcher developed the MammoFormer framework, which unites transformer-based architecture with multi-feature enhancement components and XAI functionalities within one framework. Seven different architectures consisting of CNNs, Vision Transformer, Swin Transformer, and ConvNext were tested alongside four enhancement techniques, including original images, negative transformation, adaptive histogram equalization, and histogram of oriented gradients. The MammoFormer framework addresses critical clinical adoption barriers of AI mammography systems through: (1) systematic optimization of transformer architectures via architecture-specific feature enhancement, achieving up to 13% performance improvement, (2) comprehensive explainable AI integration providing multi-perspective diagnostic interpretability, and (3) a clinically deployable ensemble system combining CNN reliability with transformer global context modeling. The combination of transformer models with suitable feature enhancements enables them to achieve equal or better results than CNN approaches. ViT achieves 98.3% accuracy alongside AHE while Swin Transformer gains a 13.0% advantage through HOG enhancements

Figures

Figures reproduced from arXiv: 2508.06137 by the authors.

Figure 4
Figure 4. MammoFormer performance analysis: (a) Model accuracy across enhancement techniques showing variable performance of transformer models compared to stable CNN-based models; (b) Average accuracy by enhancement type with HOG achieving highest overall performance at 98.4%; (c) Comparison between transformer-based and CNN architectures showing transformers approach CNN performance with appropriate enhancements; (d) Signif… view at source ↗
Figure 6
Figure 6. Comparison of XAI methods on mammogram analysis: Integrated Gradients (top left) highlights sparse point-like features; Occlu￾sion analysis (top right) reveals stronger attribution in lower regions indicating areas that significantly affect prediction when blocked; Sa￾liency maps (bottom left) identify high-gradient features influencing classification; DeepLIFT (bottom right) shows similar sparse attribution pattern… view at source ↗
Figure 7
Figure 7. reveals consistent attribution patterns across all XAI methods focused on a central region of potential ab￾normality. Occlusion analysis provides the strongest signal with bright yellow-orange highlighting, indicating this area's critical importance to model prediction. Integrated Gradients, Saliency, and DeepLIFT all converge on the same central feature, demonstrating robust consensus across different ex￾planation … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 16 canonical work pages

  1. [1]

    Mehta, et al., ―End-to-End diagnosis of breast biopsy images with transformers,‖ Medical Image Analysis, vol

    S. Mehta, et al., ―End-to-End diagnosis of breast biopsy images with transformers,‖ Medical Image Analysis, vol. 79, 2022. https://doi.org/10.1016/j.media.2022.102466

  2. [2]

    Ayana, et al., ―Vision -Transformer-Based Transfer Lear n- ing for Mammogram Classification,‖ Diagnostics (Basel), vol

    G. Ayana, et al., ―Vision -Transformer-Based Transfer Lear n- ing for Mammogram Classification,‖ Diagnostics (Basel), vol. 13, no. 2, 2023. https://doi.org/10.3390/diagnostics13020178

  3. [3]

    A. A. Jeny, et al., ―Hybrid transformer -based model for mammogram classification by integrating prior and current images,‖ Medical Physics, 2025. https://doi.org/10.1002/mp.17650

  4. [4]

    S. Hussain, et al., ―Performance Evaluation of Deep Learning and Transformer Models Using Multimodal Data for Breast Cancer Classification,‖ Cancer Prevention, Detection, and I n- tervention, Caption 2024, vol. 15199, pp. 59-69, 2025. https://doi.org/10.48550/arXiv.2410.10146

  5. [5]

    Iqbal and M

    A. Iqbal and M. Sharif, ―BTS -ST: Swin transformer network for segmentation and classification of multimodality breast cancer images,‖ Knowledge - Based Systems, vol. 267, 2023. https://doi.org/10.1016/j.knosys.2023.110393

  6. [6]

    Y. Shen, J. P., F. Yeung, E. Goldberg, L. Heacock, F. Shamout, K. J. Geras, ―Leveraging Transformers to Improve Breast Cancer Classification and Risk Assessment with Multi -modal and Longitudinal Data,‖ arxiv.org, 2023. https://doi.org/10.48550/arXiv.2311.03217

  7. [7]

    X. X. Chen, et al., ―Transformers Improve Breast Cancer Diagnosis from Unregistered Multi -View Mammograms,‖ Diagnostics, vol. 12, no. 7, 2022. https://doi.org/10.3390/diagnostics12071549

  8. [8]

    E. Z. Dalah, et al., ―Screening Mammography Diagnostic Reference Level System According to Compressed Breast Thickness: Dubai Health,‖ Journal of Imaging, vol. 10, no. 8,

Show all 22 references
  1. [9]

    Sarker, et al., ―MV -Swin-T: MAMMOGRAM CLASS I- FICATION WITH MULTI-VIEW SWIN TRANSFORMER,‖ IEEE International Symposium on Biomedical Imaging, ISBI 2024, 2024

    S. Sarker, et al., ―MV -Swin-T: MAMMOGRAM CLASS I- FICATION WITH MULTI-VIEW SWIN TRANSFORMER,‖ IEEE International Symposium on Biomedical Imaging, ISBI 2024, 2024. https://doi.org/isbi56570.2024.10635578

  2. [10]

    Kassis, et al., ―Detection of breast cancer in digital breast tomosynthesis with vision transformers,‖ Scientific Reports, vol

    I. Kassis, et al., ―Detection of breast cancer in digital breast tomosynthesis with vision transformers,‖ Scientific Reports, vol. 14, no. 1, 2024. https://doi.org/10.1038/s41598-024-72707-2

  3. [12]

    M. L. Abimouloud, K. B., M. Elleuch, O. Aiadi, and M. Kherallah, ―Vision transformer -convolution for breast cancer classification using mammography images: A comparative study,‖ International Journal of Hybrid Intelligent Systems, vol. 20, no. 2, pp. 67-83, 2024. https://doi....

  4. [13]

    H. N. Wang, et al., ―Transformer-Based Explainable Model for Breast Cancer Lesion Segmentation,‖ Applied Sciences-Basel, vol. 15, no. 3, 2025. https://doi.org/10.3390/app15031295

  5. [14]

    Adeniran, T

    O. Adeniran, T. Blessing, T. E. Ajibola, O. O. Ejiga Peter, A. O. Ajala, M. M. Rahman, and F. Khalifa, ―Explainable MRI-Based Ensemble Learnable Architecture for Alzheimer’s Disease Detection,‖ Algorithms, vol. 18, no. 3, pp. 163, 2025. https://doi.org/10.3390/a18030163

  6. [15]

    Wang, et al., ―S emi-supervised vision transformer with adaptive token sampling for breast cancer classification,‖ Frontiers in Pharmacology, vol

    W. Wang, et al., ―S emi-supervised vision transformer with adaptive token sampling for breast cancer classification,‖ Frontiers in Pharmacology, vol. 13, 2022. https://doi.org/10.3389/fphar.2022.929755

  7. [16]

    W. S. Lee, et al., ―Transformer -based Deep Neural Network for Breast Canc er Classification on Digital Breast Tomosy n- thesis Images,‖ Radiology-Artificial Intelligence, vol. 5, no. 3,

  8. [17]

    Basaad, et al., ―A BERT -GNN Approach for Metastatic Breast Cancer Prediction Using Histopathology Repo rts,‖ Diagnostics, vol

    A. Basaad, et al., ―A BERT -GNN Approach for Metastatic Breast Cancer Prediction Using Histopathology Repo rts,‖ Diagnostics, vol. 14, no. 13, 2024. https://doi.org/10.3390/diagnostics14131365

  9. [18]

    O. O. Ejiga Peter, M. M. Rahman, and F. Khalifa, ―Advancing AI-Powered Medical Image Synthesis: Insights from MedVQA-GI Challenge Using CLIP, Fine - Tuned Stable Diffusion, and Dream-Booth + LoRA,‖ 2024. https://doi.org/0.48550/arXiv.2502.20667

  10. [19]

    O. O. Ejiga Peter, ―Advancing Colonoscopy Analysis Through Text-to-Image Synthesis Using Generative AI for Intelligent Data Augmentation, Image Classification, and Segmentation,‖ ProQuest, 2024

  11. [20]

    O. O. Ejiga Peter, O. T. Adeniran, J. A. MacGregor, F. Khalifa, and M. M. Rahman, ―Text -Guided Synthesis in Medical Mu l- timedia Retrieval: A Framework for Enhanced Colonoscopy Image Classification and Segmentation,‖ Algorithms, vol. 18, no. 3, pp. 155, 2025. https://doi.org...

  12. [21]

    O. O. Ejiga Peter, O. Akingbola, C. Amalahu, O. Adeniran, F. Khalifa, and M. M. Rahman, ―Synthetic data -driven mu l- ti-architecture framework for automated polyp segmentation through integrated detection and mask generation,‖ Proc. SPIE, vol. 13410, 2025. https://doi.org/10....

  13. [2023]

    https://doi.org/10.1148/ryai.220159 American Journal of Computer Science and Technology http://www.sciencepg.com/journal/ajcst 137

  14. [2024]

    https://doi.org/10.3390/jimaging10080188

Pith tools

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