Toward a robust lesion detection model in breast DCE-MRI: adapting foundation models to high-risk women

Alex Nguyen; Andrew D. A. Maidment; Anne Marie McCarthy; Bruno Barufaldi; Carla R. Zeballos Torrez; Gabriel A.B. do Nascimento; Guilherme J. Cavalcante; James C. Gee; Telmo M. Silva Filho; Tha\'is G. do R\^ego

REVIEW 4 major objections 5 minor 18 references

Toward a robust lesion detection model in breast DCE-MRI: adapting foundation models to high-risk women

T0 review · 4 major / 5 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 A frozen medical-imaging transformer plus a Kolmogorov-Arnold classifier reaches AUC 0.80 for classifying benign versus malignant breast lesions on DCE-MRI in a high-risk screening cohort.

desk verdict Plausible incremental transfer-learning result, but the central comparison is under-documented and the slice-level labeling may inflate the AUC. read the letter →

arxiv 2509.02710 v1 pith:ZH4LR5QG submitted 2025-09-02 physics.med-ph cs.CVcs.LG

Gabriel A.B. do Nascimento , Vincent Dong , Guilherme J. Cavalcante , Alex Nguyen , Thaís G. do Rêgo , Yuri Malheiros , Telmo M. Silva Filho , Carla R. Zeballos Torrez

show 4 more authors

James C. Gee Anne Marie McCarthy Andrew D. A. Maidment Bruno Barufaldi

This is my paper · ORCID

classification physics.med-phcs.CVcs.LG

keywords breastDCE-MRIlesionclassificationMedicalSliceTransformerKolmogorov-ArnoldNetworksDINOv2attentionheatmapshigh-riskscreeningtransferlearning

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

This paper asks whether a frozen medical-imaging foundation model can be repurposed for breast lesion classification in a high-risk screening population by swapping its final classifier for a Kolmogorov-Arnold network. The authors report that the adapted pipeline reaches an AUC of 0.80 ± 0.02, a clear improvement over the original slice-wise transformer classifier, while attention heatmaps still point to lesion sites. If the result holds, it offers a lightweight route to adapt self-supervised MRI features to a new clinical task and population. The paper's central claim is the empirical performance gain itself, not a new theory.

What carries the argument

The load-bearing components are: the Medical Slice Transformer (MST), a slice-wise vision transformer pretrained with DINOv2 that converts each 256x256 axial subtraction slice into a feature embedding; and the Kolmogorov-Arnold Network (KAN), whose learnable B-spline activations replace fixed nonlinearities and are trained on the frozen embeddings. The attention mechanism provides the interpretability claim.

What would settle it

Take a held-out set of DCE-MRI studies with manually annotated lesion masks. Compare model AUC on slices that contain lesion versus slices that do not; if the classifier keeps high AUC on lesion-free slices, its decision is not lesion-detection. Additionally compute overlap of attention heatmaps with annotated masks; low Dice on true positives would contradict the localization claim.

Watch

Extended reading notes

Core claim

On a biopsy-confirmed cohort of 4,900 benign and 1,214 malignant breast DCE-MRI studies from high-risk screening, the paper's MST+KAN pipeline classifies each axial slice by extracting frozen DINOv2-based embeddings from the Medical Slice Transformer and training a two-hidden-layer KAN with adaptive B-spline activations. In five-fold cross-validation, it reports AUC=0.80±0.02, outperforming the standard MST classifier; attention heatmaps concentrate on lesions, including small lesions and dense tissue, even though the model is never given segmentation masks.

Load-bearing premise

Each 2D axial slice is labelled with the biopsy result of the whole breast, even though many slices contain no lesion, so high accuracy may come from whole-breast signals instead of the lesion.

Editorial extensions

If this is right

An AUC of 0.80 would support using frozen MST embeddings as general-purpose breast MRI features for new classification heads.
The KAN head could be retrained quickly on other cohorts or protocols without recomputing embeddings.
Attention heatmaps could serve as weakly supervised lesion localization for triage in BI-RADS 4 cases.
The pipeline addresses class imbalance via SMOTE and focal loss, making it more applicable to real screening data.

Reading between the lines

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

Because slices are labeled by whole-breast biopsy outcome, the reported AUC may partly reflect breast-level confounds rather than lesion detection; testing on lesion-free slices would separate these.
The improvement is attributed to KAN, but the paper does not compare against an MLP trained on identical embeddings; that comparison would isolate the KAN contribution.
If attention localization is accurate, the same embeddings could support lesion segmentation without additional supervision.
The method's generality across scanner types is asserted from attention maps; a multi-site external validation would turn that assertion into measurable generalization.

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.

Desk Editor's Note

read the letter

Quick take: this is a modest transfer-learning paper with a real empirical result, but you cannot yet verify the main claim because the baseline AUC is never reported and the slice-to-patient label mapping is underspecified.

The new bit is narrow: taking the MST embeddings from Müller-Franzes et al. and feeding them into a KAN classifier on their UPHS high-risk cohort. That combination is not in the literature as far as I can tell, so on that narrow point the result is new. The AUC of 0.80 on 6,114 biopsy-confirmed studies is plausible and could mean something clinically if it holds up. What the paper does well: it follows a sensible transfer-learning recipe, uses group-stratified 5-fold CV, addresses imbalance with focal loss and Borderline-SMOTE, and it is honest about several limitations (small cohort, slice-wise loses 3D context, expert-label subjectivity).

The problems are in the evidence for the headline claim. First, the baseline MST AUC is never given numerically. The text says 'clear improvement' and Figure 2 shows two ROC curves, but without the baseline number the comparison is not assessable. Second, there is no statistical test, so the ±0.02 is just fold spread. Third, the label question is potentially load-bearing. Each 2D axial slice is treated as independent while the label is the biopsy outcome for the breast. The text says final classification is at patient level, but no aggregation step is described. If slices are labeled benign/malignant by whole-breast outcome, most slices in a malignant study contain no lesion, so the model can learn non-lesion correlates like background parenchymal enhancement or density. That could inflate AUC and make the attention heatmaps misleading. The paper admits in Figure 1B that some attention is wrongly assigned, which is consistent with this worry. Absent lesion-level annotations or a slice-level localization evaluation, the 'lesion detection' framing is not established.

None of this makes the paper worthless. The KAN-on-frozen-embeddings trick is cheap and easy to reproduce if the embeddings are available, and the dataset size is a plus. But as written, it supports a 'we tried this and it seems promising' claim, not 'clear improvement.'

Who is it for: people working on breast MRI CAD or foundation-model transfer for medical imaging might want to read it. It deserves a serious referee because the question is clinically relevant and the cohort is substantial. I would not cite it yet; I would wait for a version that reports the baseline, describes the patient-level aggregation, and ideally external validation.

Referee Report

4 major / 5 minor

Summary. The manuscript proposes a transfer-learning pipeline for benign/malignant classification of breast DCE-MRI in a high-risk screening cohort. It uses the Medical Slice Transformer (MST), pretrained with DINOv2 on breast MRI, to extract per-slice embeddings from subtraction images. These embeddings are then used to train a Kolmogorov-Arnold Network (KAN) with two hidden layers (128 and 64 neurons) and B-spline activations. Training uses 5-fold stratified cross-validation with Borderline-SMOTE and focal loss, and the authors report AUC = 0.80 ± 0.02 for the MST+KAN pipeline, claiming it outperforms the MST baseline while preserving interpretability through attention heatmaps. The paper includes ROC curves and representative attention maps, but no code or data are provided.

Significance. If the comparative result were properly supported, the contribution would be useful: it demonstrates that frozen embeddings from a medical foundation model can be combined with a compact KAN classifier to handle class imbalance in a challenging high-risk population, and it proposes an attention-based visualization for clinician-facing interpretability. The approach is practical and reproducible in principle. However, the current manuscript does not substantiate the central comparative claim because the baseline MST AUC is never stated numerically, and the slice-level labeling with breast-level labels makes the 'lesion detection' interpretation precarious. Strengths include the explicit preprocessing pipeline, use of a public foundation model, and stratified cross-validation. Weaknesses include the absence of statistical tests, external validation, and code/data availability, as well as the unresolved slice-to-patient aggregation.

major comments (4)

[§3, Fig. 2] The baseline MST AUC is never reported numerically. The only quantitative result is AUC = 0.80 ± 0.02 for the proposed pipeline, so the claims of 'clear improvement' and 'outperforms' cannot be evaluated. Please report the baseline AUC, per-fold AUCs, confidence intervals, and a paired statistical comparison (e.g., DeLong test on patient-level scores, or bootstrap across folds).
[§2.3–2.4] The model is trained on per-slice inputs labeled by breast-level biopsy outcome, and no aggregation from slice-level to patient-level predictions is described, despite §2.3 stating that final classification is at the patient level. Since no lesion segmentation is used, most slices in malignant studies likely contain no lesion but are labeled malignant. This permits the classifier to exploit non-lesion confounds (e.g., background parenchymal enhancement, breast density, biopsy-related artifacts) and would inflate the reported AUC. The attention heatmaps would then not reflect lesion detection. The authors should either train/evaluate with lesion-level annotations or provide a slice-level localization evaluation and explicitly describe how per-slice predictions are aggregated to patient level.
[§2.4, §3] The '±0.02' is reported without stating how the uncertainty was computed, and the text uses 'significant performance improvement' without any statistical test. Additionally, 'non-overlapping groups' is not defined, so it is unclear whether the same patient can appear in both training and validation folds (e.g., through multiple studies, laterality, or repeated exams). Address this grouping explicitly, report per-fold scores, and provide confidence intervals or tests. Without these, the robustness/generalizability claims are not supported.
[§3, Fig. 1] The interpretability argument rests on attention heatmaps, but the paper's own Figure 1B states that 'some attention is incorrectly assigned,' and there is no quantitative localization evaluation (e.g., Dice score against lesion masks, or distance to annotated lesion centers). The heatmaps are anecdotal. To support the claim that attention identifies lesion-relevant regions, the authors need a quantitative evaluation or should temper the interpretability claim.

minor comments (5)

[§1, References] Reference [10] is cited for the Medical Slice Transformer, but the MST model used is from Müller-Franzes et al. [13]. Please correct the citation in the introduction.
[§2.1] Typo: 'We excluded studies malignant studies where cancer laterality was not provided' should read 'We excluded malignant studies where cancer laterality was not provided.'
[§2.1, §2.4] The text uses 'studies' and 'patients' interchangeably; clarify whether 4,900 benign and 1,214 malignant entries are unique patients or imaging studies, and whether a patient may contribute both a benign and a malignant study.
[§2.4] The phrase 'we randomly sample a volume from the cohort of benign patients such that each sample consists of one breast' is unclear. Specify the unit of analysis and how left/right breast volumes are constructed and split across folds.
[§3, Fig. 2] Figure 2 would benefit from adding AUC values to the legend or panel labels, and from clearly distinguishing the two ROC curves; the current text does not give the baseline AUC.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the reported AUC is an empirical held-out comparison, not a quantity that reduces to its inputs by construction.

full rationale

Walked the derivation chain: (1) A pretrained external MST (DINOv2-based, Müller-Franzes et al.) is used to extract per-slice embeddings; the weights are not refit to the target labels. (2) A KAN classifier is trained on those embeddings with 5-fold stratified cross-validation; the reported AUC=0.80±0.02 is measured on held-out folds. (3) The comparison to the MST baseline is an empirical ROC comparison; no equation in the paper defines the KAN AUC as equal to the fitted spline parameters or to the MST baseline by construction. (4) The only overlapping reference among the authors is N4 bias-field correction [Tustison et al., includes Gee], which is a standard external preprocessing algorithm and is not load-bearing for the central claim. The paper has substantive weaknesses—slice-level labels with no lesion segmentation, no explicit slice-to-patient aggregation described in Section 2.4, an unreported numeric baseline AUC, and the caption of Figure 1 admits 'some attention is incorrectly assigned'—but these are correctness/generalization risks, not circular reasoning. No fitted parameter is renamed as a prediction, no self-citation is invoked as a uniqueness proof, and no ansatz is smuggled in via citation. Hence no circular step can be exhibited with the required quote-and-reduction evidence.

Assumptions & free parameters 6 free parameters · 3 assumptions · 0 invented entities

The central claim rests on pretrained embeddings (external inputs) plus a set of hand-chosen training hyperparameters and a strong slice-label assumption. No new physical entities or mathematical axioms are introduced.

free parameters (6)

KAN hidden layer sizes = 128 and 64
Architecture choice reported in Section 2.4; no ablation or sensitivity analysis is given.
Learning rate = 1e-6
AdamW learning rate reported in Section 2.4; chosen by hand without justification.
Batch size = 64
Training batch size reported in Section 2.4; chosen by hand.
Early stopping patience and max epochs = patience 5, epochs 500
Reported in Section 2.4; not justified or ablated.
Focal loss hyperparameters (gamma/alpha) = not reported
Focal loss is used in Section 2.4 but gamma and class weights are not specified.
Borderline-SMOTE sampling ratio/neighbor settings = not reported
Used within each fold but the sampling ratio and neighbor parameters are missing.

assumptions (3)

domain assumption Each axial slice inherits the breast/patient-level biopsy label.
Section 2.3 states each slice is an independent input while the final label is patient/breast-level. Slices without any lesion are therefore labeled by the whole-breast outcome, which can inject label noise.
domain assumption Pretrained MST/DINOv2 embeddings transfer from DUKE and ImageNet to UPHS scanners.
Section 2.4 freezes the encoder and uses it as a feature extractor. Scanner and protocol differences between DUKE and UPHS could shift the feature distribution and degrade performance.
domain assumption The preprocessing sequence preserves information needed for lesion classification.
Section 2.2 applies resampling, N4 bias correction, PLHE, subtraction, and cropping to 256x256x32. If these steps remove or distort small lesions, the classifier cannot recover the lost signal.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Toward a robust lesion detection model in breast DCE-MRI: adapting foundation models to high-risk women." pith.science (2026). https://pith.science/paper/ZH4LR5QG

@misc{pith2026250902710,
  author       = {Pith},
  title        = {Pith review of: Toward a robust lesion detection model in breast DCE-MRI: adapting foundation models to high-risk women},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZH4LR5QG}},
  note         = {Machine review of arXiv:2509.02710}
}

read the original abstract

Accurate breast MRI lesion detection is critical for early cancer diagnosis, especially in high-risk populations. We present a classification pipeline that adapts a pretrained foundation model, the Medical Slice Transformer (MST), for breast lesion classification using dynamic contrast-enhanced MRI (DCE-MRI). Leveraging DINOv2-based self-supervised pretraining, MST generates robust per-slice feature embeddings, which are then used to train a Kolmogorov--Arnold Network (KAN) classifier. The KAN provides a flexible and interpretable alternative to conventional convolutional networks by enabling localized nonlinear transformations via adaptive B-spline activations. This enhances the model's ability to differentiate benign from malignant lesions in imbalanced and heterogeneous clinical datasets. Experimental results demonstrate that the MST+KAN pipeline outperforms the baseline MST classifier, achieving AUC = 0.80 \pm 0.02 while preserving interpretability through attention-based heatmaps. Our findings highlight the effectiveness of combining foundation model embeddings with advanced classification strategies for building robust and generalizable breast MRI analysis tools.

Figures

Figures reproduced from arXiv: 2509.02710 by the authors.

**Figure 2.** ROC-AUC curves of the KAN and MST classifiers. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

18 extracted references · 16 canonical work pages

[1]

Rapid increase in breast magnetic resonance imaging use: Trends from 2000 to 2011,

Stout, N. K., Nekhlyudov, L., Li, L., Malin, E. S., Ross-Degnan, D., Buist, D. S. M., Rosenberg, M. A., Alfisher, M., and Fletcher, S. W., “Rapid increase in breast magnetic resonance imaging use: Trends from 2000 to 2011,” JAMA Internal Medicine 174, 114–121 (01 2014)

work page 2000
[2]

Patterns of breast magnetic resonance imaging use in community practice,

Wernli, K. J., DeMartini, W. B., Ichikawa, L., Lehman, C. D., Onega, T., Kerlikowske, K., Henderson, L. M., Geller, B. M., Hofmann, M., Yankaskas, B. C., and for the Breast Cancer Surveillance Consortium, “Patterns of breast magnetic resonance imaging use in community practice,” JAMA Internal Medicine 174, 125–132 (01 2014)

work page 2014
[3]

Recent trends in screening breast MRI,

Lee, M. V., Aharon, S., Kim, K., Sunn Konstantinoff, K., Appleton, C. M., Stwalley, D., and Olsen, M. A., “Recent trends in screening breast MRI,” J Breast Imaging 4, 39–47 (Dec. 2021)

work page 2021
[4]

Trends and clinical implications of preoperative breast MRI in medicare beneficiaries with breast cancer,

Killelea, B. K., Long, J. B., Chagpar, A. B., Ma, X., Soulos, P. R., Ross, J. S., and Gross, C. P., “Trends and clinical implications of preoperative breast MRI in medicare beneficiaries with breast cancer,” Breast Cancer Res Treat 141, 155–163 (Aug. 2013)

work page 2013
[5]

Utility of bi-rads assessment category 4 subdivisions for screening breast mri,

Strigel, R. M., Burnside, E. S., Elezaby, M., and et al., “Utility of bi-rads assessment category 4 subdivisions for screening breast mri,” AJR Am J Roentgenol 208(6), 1392–1399 (2017)

work page 2017
[6]

Bi-rads 3, 4, and 5 lesions: Value of us in management—follow-up and outcome,

Raza, S., Chikarmane, S. A., Neilsen, S. S., Zorn, L. M., and Birdwell, R. L., “Bi-rads 3, 4, and 5 lesions: Value of us in management—follow-up and outcome,” Radiology 248(3), 773–781 (2008)

work page 2008
[7]

Analysis of utilization patterns and associated costs of the breast imaging and diagnostic procedures after screening mammography,

Vlahiotis, A., Griffin, B., Stavros, M. D. F., and Margolis, J., “Analysis of utilization patterns and associated costs of the breast imaging and diagnostic procedures after screening mammography,”ClinicoEconomics and Outcomes Research 10, 157–167 (2018)

work page 2018
[8]

External validation of deep learning algorithms for radiologic diagnosis: A systematic review,

Yu, A. C., Mohajer, B., and Eng, J., “External validation of deep learning algorithms for radiologic diagnosis: A systematic review,” Radiology: Artificial Intelligence 4, e210064 (May 2022)

work page 2022

Show all 18 references

[9]

Foundation models in radiology: What, how, why, and why not,

Wang, X. et al., “Foundation models in radiology: What, how, why, and why not,” Radiology 308, e240597 (2025)

2025
[10]

M3t: Three-dimensional medical image classifier using multi-plane and multi-slice transformer,

Jang, J., Hwang, D., and et al., “M3t: Three-dimensional medical image classifier using multi-plane and multi-slice transformer,” in [ Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition (CVPR) ], IEEE, New Orleans, LA, USA (June 2022)

2022
[11]

N4ITK: Improved N3 bias correction,

Tustison, N. J., Avants, B. B., Cook, P. A., Zheng, Y., Egan, A., Yushkevich, P. A., and Gee, J. C., “N4ITK: Improved N3 bias correction,” IEEE Transactions on Medical Imaging 29, 1310–1320 (June 2010)

2010
[12]

Adjustable contrast enhancement using fast piecewise linear histogram equalization,

Javadi, S., Dahl, M., and Pettersson, M. I., “Adjustable contrast enhancement using fast piecewise linear histogram equalization,” in [ Proceedings of the 2020 3rd International Conference on Image and Graphics Processing], ICIGP ’20 , 57–61, Association for Computing Machiner...

2020
[13]

Medical slice transformer for improved diagnosis and explainability on 3d medical images with dinov2,

M¨ uller-Franzes, G., Khader, F., Siepmann, R., and et al., “Medical slice transformer for improved diagnosis and explainability on 3d medical images with dinov2,” Scientific Reports 15, 23979 (2025). Received 11 February 2025; Accepted 25 June 2025; Published 04 July 2025

2025
[14]

Dinov2: Learning robust visual features without supervision,

Oquab, M., Darcet, T., Moutakanni, T., Vo, H. V., Szafraniec, M., Khalidov, V., Fernandez, P., Haziza, D., Massa, F., El-Nouby, A., Howes, R., Huang, P.-Y., Xu, H., Sharma, V., Li, S.-W., Galuba, W., Rabbat, M., Assran, M., Ballas, N., Synnaeve, G., Misra, I., Jegou, H., Maira...

2023
[15]

Dynamic contrast-enhanced magnetic resonance images of breast cancer patients with tumor locations

Saha, A., Harowicz, M. R., Grimm, L. J., Weng, J., Cain, E. H., Kim, C. E., Ghate, S. V., Walsh, R., and Mazurowski, M. A., “Dynamic contrast-enhanced magnetic resonance images of breast cancer patients with tumor locations.” https://doi.org/10.7937/TCIA.e3sv-re93 (2021). The ...

2021 doi
[16]

Kan: Kolmogorov-arnold networks,

Liu, Z., Wang, Y., Vaidya, S., Ruehle, F., Halverson, J., Soljaˇ ci´ c, M., Hou, T. Y., and Tegmark, M., “Kan: Kolmogorov-arnold networks,” arXiv preprint arXiv:2404.19756 (2024)

2024 arXiv
[17]

Borderline-smote: a new over-sampling method in imbalanced data sets learning,

Han, H., Wang, W.-Y., and Mao, B.-H., “Borderline-smote: a new over-sampling method in imbalanced data sets learning,” in [ Proceedings of the 2005 International Conference on Advances in Intelligent Computing - Volume Part I ], ICIC’05, 878–887, Springer-Verlag, Berlin, Heide...

2005
[18]

Focal loss for dense object detection,

Lin, T.-Y., Goyal, P., Girshick, R., He, K., and Doll´ ar, P., “Focal loss for dense object detection,” (2018)

2018

Pith tools

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

[1] [1]

Rapid increase in breast magnetic resonance imaging use: Trends from 2000 to 2011,

Stout, N. K., Nekhlyudov, L., Li, L., Malin, E. S., Ross-Degnan, D., Buist, D. S. M., Rosenberg, M. A., Alfisher, M., and Fletcher, S. W., “Rapid increase in breast magnetic resonance imaging use: Trends from 2000 to 2011,” JAMA Internal Medicine 174, 114–121 (01 2014)

work page 2000

[2] [2]

Patterns of breast magnetic resonance imaging use in community practice,

Wernli, K. J., DeMartini, W. B., Ichikawa, L., Lehman, C. D., Onega, T., Kerlikowske, K., Henderson, L. M., Geller, B. M., Hofmann, M., Yankaskas, B. C., and for the Breast Cancer Surveillance Consortium, “Patterns of breast magnetic resonance imaging use in community practice,” JAMA Internal Medicine 174, 125–132 (01 2014)

work page 2014

[3] [3]

Recent trends in screening breast MRI,

Lee, M. V., Aharon, S., Kim, K., Sunn Konstantinoff, K., Appleton, C. M., Stwalley, D., and Olsen, M. A., “Recent trends in screening breast MRI,” J Breast Imaging 4, 39–47 (Dec. 2021)

work page 2021

[4] [4]

Trends and clinical implications of preoperative breast MRI in medicare beneficiaries with breast cancer,

Killelea, B. K., Long, J. B., Chagpar, A. B., Ma, X., Soulos, P. R., Ross, J. S., and Gross, C. P., “Trends and clinical implications of preoperative breast MRI in medicare beneficiaries with breast cancer,” Breast Cancer Res Treat 141, 155–163 (Aug. 2013)

work page 2013

[5] [5]

Utility of bi-rads assessment category 4 subdivisions for screening breast mri,

Strigel, R. M., Burnside, E. S., Elezaby, M., and et al., “Utility of bi-rads assessment category 4 subdivisions for screening breast mri,” AJR Am J Roentgenol 208(6), 1392–1399 (2017)

work page 2017

[6] [6]

Bi-rads 3, 4, and 5 lesions: Value of us in management—follow-up and outcome,

Raza, S., Chikarmane, S. A., Neilsen, S. S., Zorn, L. M., and Birdwell, R. L., “Bi-rads 3, 4, and 5 lesions: Value of us in management—follow-up and outcome,” Radiology 248(3), 773–781 (2008)

work page 2008

[7] [7]

Analysis of utilization patterns and associated costs of the breast imaging and diagnostic procedures after screening mammography,

Vlahiotis, A., Griffin, B., Stavros, M. D. F., and Margolis, J., “Analysis of utilization patterns and associated costs of the breast imaging and diagnostic procedures after screening mammography,”ClinicoEconomics and Outcomes Research 10, 157–167 (2018)

work page 2018

[8] [8]

External validation of deep learning algorithms for radiologic diagnosis: A systematic review,

Yu, A. C., Mohajer, B., and Eng, J., “External validation of deep learning algorithms for radiologic diagnosis: A systematic review,” Radiology: Artificial Intelligence 4, e210064 (May 2022)

work page 2022