pith. sign in

arxiv: 2605.22327 · v1 · pith:S4JKPU3Pnew · submitted 2026-05-21 · 💻 cs.CV · physics.med-ph

Robustness of breast lesion segmentation under MRI undersampling improves with k-space-aware deep learning

Lukas T. Rotkopf , Marco Schlimbach , Julius C. Holzschuh , Heinz-Peter Schlemmer , Jens Kleesiek , Moritz Rempe This is my paper

Pith reviewed 2026-05-22 06:51 UTC · model grok-4.3

classification 💻 cs.CV physics.med-ph
keywords breast lesion segmentationk-space aware deep learningMRI undersamplingrobustnessDCE-MRIhybrid modelimage segmentationaccelerated MRI
0
0 comments X

The pith

K-space-aware deep learning improves robustness of breast lesion segmentation under MRI undersampling and noise while matching standard methods at full sampling.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests whether breast lesion segmentation can be learned directly from acquired MRI k-space data rather than only from reconstructed images, and whether this improves performance when scans are accelerated or noisy. Using public DCE-MRI datasets with both real and synthetic k-space, the authors train hybrid k-space-to-image models alongside pure image-space baselines and evaluate them under increasing levels of undersampling and added complex Gaussian noise. At full sampling the approaches perform similarly, but as acceleration rises or noise is introduced in k-space the hybrid model retains higher patient-level Dice scores and degrades more slowly than the magnitude image-space baseline. The same pattern appears in a within-dataset synthetic control, and feature analysis indicates the k-space stage performs frequency filtering while the image stage handles lesion localization.

Core claim

A hybrid k-space-to-image 3D U-Net variant for breast lesion segmentation matches the patient-level Dice similarity coefficient of magnitude image-space baselines at full sampling but significantly outperforms them across moderate to high undersampling levels and when complex Gaussian noise is added directly to k-space, with the advantage reproduced in a within-dataset synthetic control and supported by analysis showing complementary roles for the k-space stage in frequency-domain filtering and the image stage in lesion localization.

What carries the argument

The hybrid k-space-to-image 3D U-Net variant that processes data first in the frequency domain before transitioning to image space, enabling complementary frequency filtering and lesion localization.

If this is right

  • The hybrid model retains substantially more segmentation accuracy than image-space baselines as acceleration increases.
  • The hybrid model degrades more slowly than image-space baselines when complex Gaussian noise is added to k-space.
  • Feature analysis shows the k-space stage concentrates on frequency-domain filtering while the image stage performs lesion localization.
  • The performance advantage is reproduced in a within-dataset synthetic control.
  • The hybrid and image-space models perform similarly at full sampling without added noise.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The hybrid architecture could be tested on segmentation tasks for other organs commonly imaged with accelerated MRI protocols.
  • Prospective validation on real accelerated clinical scans would be required to confirm whether the synthetic undersampling results hold in practice.
  • Combining the k-space stage with existing compressed-sensing or parallel-imaging reconstruction methods might produce even more robust end-to-end pipelines.
  • The observed complementary roles suggest hybrid k-space-image models may benefit other inverse imaging problems where raw sensor data preserves information lost in standard reconstructions.

Load-bearing premise

Synthetic undersampling and added complex Gaussian noise applied to retrospective public datasets accurately represent the characteristics of real prospectively accelerated clinical MRI acquisitions and their noise.

What would settle it

A direct comparison on prospectively undersampled breast DCE-MRI data acquired on clinical scanners showing that the hybrid model's Dice advantage over image-space baselines disappears or reverses.

Figures

Figures reproduced from arXiv: 2605.22327 by Heinz-Peter Schlemmer, Jens Kleesiek, Julius C. Holzschuh, Lukas T. Rotkopf, Marco Schlimbach, Moritz Rempe.

Figure 1
Figure 1. Figure 1: Study overview. Two large-scale breast MRI datasets were used: fastMRI breast, providing [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the model architectures compared in this study. FFT = fast Fourier [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Segmentation performance and robustness analyses. (A) MAMA-MIA dataset with [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Representative segmentation outputs at 1 [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Feature visualization of the hybrid k-space-to-image model’s k-space stage on fully sampled [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
read the original abstract

Purpose: To assess whether breast lesion segmentation can be learned directly from acquired MRI k-space, and whether doing so improves robustness when data are accelerated or noisy. Materials and Methods: This retrospective study used public breast dynamic contrast-enhanced MRI (DCE-MRI) datasets with acquired and synthetic k-space, together with a within-dataset synthetic control. We compared four 3D U-Net variants: a hybrid k-space-to-image model, a native k-space model, and magnitude and complex image-space baselines. Models were evaluated under increasing undersampling and added complex Gaussian k-space noise. The primary outcome was patient-level Dice similarity coefficient under cross-validation, with the hybrid model prespecified as the main comparison against the magnitude image-space baseline. Results: At full sampling, the hybrid and image-space models performed similarly. As acceleration increased, the hybrid model retained substantially more segmentation accuracy and significantly outperformed the magnitude image-space baseline across moderate to high undersampling levels. The same pattern was observed when noise was added directly to k-space: the hybrid model degraded more slowly, whereas the image-space baseline failed under heavier noise. This advantage was reproduced in the within-dataset synthetic control. Feature analysis suggested that the k-space stage and image-space stage played complementary roles, with frequency-domain filtering concentrated before image-domain lesion localization. Conclusion: K-space-aware deep learning improves the robustness of breast lesion segmentation under MRI undersampling and k-space noise, while matching image-space methods at full sampling.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper evaluates whether breast lesion segmentation from DCE-MRI can be performed directly in k-space or via hybrid k-space-to-image models using 3D U-Net variants. It claims that a hybrid model matches image-space baselines at full sampling but retains substantially higher patient-level Dice scores under retrospective undersampling and added complex Gaussian k-space noise, with the advantage reproduced in a within-dataset synthetic control. The primary prespecified comparison is the hybrid model versus the magnitude image-space baseline, evaluated via cross-validation on public datasets.

Significance. If the central empirical findings hold, the work provides evidence that incorporating k-space processing stages can improve robustness of segmentation models to common MRI acquisition degradations without sacrificing performance at full sampling. Strengths include the use of public datasets, cross-validation, a prespecified primary endpoint, and consistent patterns across acceleration factors and noise levels. However, the significance is tempered by reliance on synthetic retrospective degradations whose fidelity to prospective clinical scans remains unproven.

major comments (2)
  1. [§3] §3 (Methods, simulation pipeline): The central robustness claim depends on retrospective undersampling masks plus i.i.d. complex Gaussian noise added to k-space from public DCE-MRI data being representative of prospective accelerated acquisitions. Real scans introduce trajectory-dependent aliasing, coil-sensitivity effects, B0/B1 inhomogeneities, and non-i.i.d. noise components that are not reproduced; without additional validation (e.g., on prospectively undersampled data or more realistic noise models), the observed Dice advantage may be an artifact of the simulation rather than an intrinsic property of k-space-aware architectures.
  2. [Results] Results (primary outcome reporting): While the abstract states that the hybrid model 'significantly outperformed' the magnitude baseline across moderate-to-high undersampling, exact effect sizes, confidence intervals, p-values, and the precise statistical test used for the prespecified comparison are not detailed. This information is load-bearing for interpreting whether the retained accuracy constitutes a clinically meaningful improvement.
minor comments (2)
  1. [Abstract / Results] The abstract mentions 'feature analysis suggested that the k-space stage and image-space stage played complementary roles' but does not specify the analysis method (e.g., activation visualization, ablation, or frequency-domain filtering quantification). Adding a brief methods paragraph or supplementary figure would improve reproducibility.
  2. [Methods] Notation for the four U-Net variants (hybrid, native k-space, magnitude image-space, complex image-space) should be defined consistently with a table or equation early in the methods to avoid ambiguity when comparing performance curves.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive and detailed review. The comments have prompted us to improve the statistical reporting and to more explicitly discuss the limitations of our simulation approach. We respond to each major comment below.

read point-by-point responses

  1. Referee: [§3] §3 (Methods, simulation pipeline): The central robustness claim depends on retrospective undersampling masks plus i.i.d. complex Gaussian noise added to k-space from public DCE-MRI data being representative of prospective accelerated acquisitions. Real scans introduce trajectory-dependent aliasing, coil-sensitivity effects, B0/B1 inhomogeneities, and non-i.i.d. noise components that are not reproduced; without additional validation (e.g., on prospectively undersampled data or more realistic noise models), the observed Dice advantage may be an artifact of the simulation rather than an intrinsic property of k-space-aware architectures.

    Authors: We agree that retrospective undersampling combined with i.i.d. complex Gaussian noise does not fully reproduce the artifacts present in prospective accelerated acquisitions, including trajectory-specific aliasing, coil-sensitivity maps, B0/B1 inhomogeneities, and non-i.i.d. noise. Because the study relies on public fully sampled retrospective datasets, prospectively undersampled data were not available. We have added a dedicated limitations paragraph in the Discussion that acknowledges these gaps and recommends prospective validation in future work. At the same time, the advantage of the hybrid model remained consistent across multiple acceleration factors and noise amplitudes and was reproduced in the within-dataset synthetic control, suggesting that the architectural difference confers some robustness even under the controlled degradations we could simulate. revision: partial

  2. Referee: [Results] Results (primary outcome reporting): While the abstract states that the hybrid model 'significantly outperformed' the magnitude baseline across moderate-to-high undersampling, exact effect sizes, confidence intervals, p-values, and the precise statistical test used for the prespecified comparison are not detailed. This information is load-bearing for interpreting whether the retained accuracy constitutes a clinically meaningful improvement.

    Authors: We appreciate this observation. In the revised manuscript we have expanded the Results section to report, for the prespecified hybrid-versus-magnitude comparison at each acceleration factor: mean patient-level Dice scores with standard deviations, 95 % confidence intervals, p-values from paired t-tests (or Wilcoxon signed-rank tests where normality assumptions were not met), and Cohen’s d effect sizes. These quantities are now presented both in the text and in an updated supplementary table so that readers can directly evaluate the magnitude and statistical significance of the observed differences. revision: yes

standing simulated objections not resolved
  • Prospective validation on real accelerated clinical acquisitions is not possible with the public retrospective datasets used in this study.

Circularity Check

0 steps flagged

No circularity: empirical ML comparison on held-out data

full rationale

The manuscript describes a retrospective empirical study that trains and evaluates four 3D U-Net variants (hybrid k-space-to-image, native k-space, magnitude and complex image-space baselines) on public breast DCE-MRI datasets using cross-validation. Performance is measured by patient-level Dice under synthetic undersampling and added complex Gaussian noise, with direct numerical comparisons reported. No equations, derivations, or predictions are presented that reduce to fitted parameters by construction, and no load-bearing self-citations or uniqueness theorems are invoked to justify the central claims. The results rest on standard held-out evaluation rather than self-referential definitions or ansatz smuggling.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard deep learning training assumptions and the representativeness of synthetic perturbations; no new physical entities are postulated.

free parameters (2)
  • U-Net architecture and training hyperparameters
    Depth, learning rate, batch size, and augmentation choices are optimized during model training to achieve segmentation performance.
  • Undersampling acceleration factors and noise levels
    Specific levels of data removal and Gaussian noise variance are selected to simulate clinical acceleration scenarios.
axioms (1)
  • domain assumption Synthetic k-space undersampling and noise addition faithfully model real accelerated MRI data characteristics
    The evaluation relies on retrospective data with simulated acceleration rather than prospectively acquired accelerated scans.

pith-pipeline@v0.9.0 · 5821 in / 1352 out tokens · 52141 ms · 2026-05-22T06:51:33.615268+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

28 extracted references · 28 canonical work pages

  1. [1]

    Abbreviated Breast Magnetic Resonance Imaging (MRI): First Postcontrast Subtracted Images and Maximum-Intensity Projection—A Novel Approach to Breast Cancer Screening With MRI,

    C. K. Kuhl, S. Schrading, K. Strobel, H. H. Schild, R.-D. Hilgers, and H. B. Bieling, “Abbreviated Breast Magnetic Resonance Imaging (MRI): First Postcontrast Subtracted Images and Maximum-Intensity Projection—A Novel Approach to Breast Cancer Screening With MRI,”Journal of Clinical Oncology, vol. 32, pp. 2304–2310, Aug. 2014

    work page 2014

  2. [2]

    Generalized autocalibrating partially parallel acquisitions (GRAPPA),

    M. A. Griswold, P. M. Jakob, R. M. Heidemann, M. Nittka, V. Jellus, J. Wang, B. Kiefer, and A. Haase, “Generalized autocalibrating partially parallel acquisitions (GRAPPA),”Magnetic Resonance in Medicine, vol. 47, pp. 1202–1210, June 2002. 10

    work page 2002

  3. [3]

    Sparse MRI: The application of compressed sensing for rapid MR imaging,

    M. Lustig, D. Donoho, and J. M. Pauly, “Sparse MRI: The application of compressed sensing for rapid MR imaging,”Magnetic Resonance in Medicine, vol. 58, pp. 1182–1195, Dec. 2007

    work page 2007

  4. [4]

    Learning a variational network for reconstruction of accelerated MRI data,

    K. Hammernik, T. Klatzer, E. Kobler, M. P. Recht, D. K. Sodickson, T. Pock, and F. Knoll, “Learning a variational network for reconstruction of accelerated MRI data,”Magnetic Resonance in Medicine, vol. 79, pp. 3055–3071, June 2018

    work page 2018

  5. [5]

    fastMRI: An Open Dataset and Benchmarks for Accelerated MRI,

    J. Zbontar, F. Knoll, A. Sriram, T. Murrell, Z. Huang, M. J. Muckley, A. Defazio, R. Stern, P. Johnson, M. Bruno, M. Parente, K. J. Geras, J. Katsnelson, H. Chandarana, Z. Zhang, M. Drozdzal, A. Romero, M. Rabbat, P. Vincent, N. Yakubova, J. Pinkerton, D. Wang, E. Owens, C. L. Zitnick, M. P. Recht, D. K. Sodickson, and Y. W. Lui, “fastMRI: An Open Dataset...

    work page

  6. [6]

    Radiologist-Level Performance by Using Deep Learning for Segmentation of Breast Cancers on MRI Scans,

    L. Hirsch, Y. Huang, S. Luo, C. Rossi Saccarelli, R. Lo Gullo, I. Daimiel Naranjo, A. G. V. Bitencourt, N. Onishi, E. S. Ko, D. Leithner, D. Avendano, S. Eskreis-Winkler, M. Hughes, D. F. Martinez, K. Pinker, K. Juluru, A. E. El-Rowmeim, P. Elnajjar, E. A. Morris, H. A. Makse, L. C. Parra, and E. J. Sutton, “Radiologist-Level Performance by Using Deep Lea...

    work page 2022

  7. [7]

    Diagnostic performance of susceptibility-weighted magnetic resonance imaging for the detection of calcifications: A systematic review and meta-analysis,

    L. C. Adams, K. K. Bressem, S. M. Böker, Y.-N. Y. Bender, D. Nörenberg, B. Hamm, and M. R. Makowski, “Diagnostic performance of susceptibility-weighted magnetic resonance imaging for the detection of calcifications: A systematic review and meta-analysis,”Scientific Reports, vol. 7, p. 15506, 2017

    work page 2017

  8. [8]

    Brain Segmentation from k-Space with End-to-End Recurrent Attention Network,

    Q. Huang, X. Chen, D. Metaxas, and M. S. Nadar, “Brain Segmentation from k-Space with End-to-End Recurrent Attention Network,” inMedical Image Computing and Computer Assisted Intervention – MICCAI 2019(D. Shen, T. Liu, T. M. Peters, L. H. Staib, C. Essert, S. Zhou, P.-T. Yap, and A. Khan, eds.), vol. 11766, pp. 275–283, Cham: Springer International Publis...

    work page 2019

  9. [9]

    Direct MRI Segmentation from k-Space Data by Iterative Potts Minimization,

    L. Kiefer, S. Petra, M. Storath, and A. Weinmann, “Direct MRI Segmentation from k-Space Data by Iterative Potts Minimization,” inScale Space and Variational Methods in Computer Vision(J. Lellmann, M. Burger, and J. Modersitzki, eds.), vol. 11603, pp. 406–418, Cham: Springer International Publishing,

    work page

  10. [11]

    k-strip: A novel segmentation algorithm in k-space for the application of skull stripping,

    M. Rempe, F. Mentzel, K. L. Pomykala, J. Haubold, F. Nensa, K. Kroeninger, J. Egger, and J. Kleesiek, “k-strip: A novel segmentation algorithm in k-space for the application of skull stripping,”Computer Methods and Programs in Biomedicine, vol. 243, p. 107912, Jan. 2024

    work page 2024

  11. [12]

    Cardiac MR Segmentation from Undersampled k-space Using Deep Latent Representation Learning,

    J. Schlemper, O. Oktay, W. Bai, D. C. Castro, J. Duan, C. Qin, J. V. Hajnal, and D. Rueckert, “Cardiac MR Segmentation from Undersampled k-space Using Deep Latent Representation Learning,” inMedical Image Computing and Computer Assisted Intervention – MICCAI 2018(A. F. Frangi, J. A. Schnabel, C. Davatzikos, C. Alberola-López, and G. Fichtinger, eds.), vol...

    work page 2018

  12. [13]

    Direct Cardiac Segmentation from Undersampled K-Space using Transformers,

    Y. Zhang, N. Stolt-Ansó, J. Pan, W. Huang, K. Hammernik, and D. Rueckert, “Direct Cardiac Segmentation from Undersampled K-Space using Transformers,” in2024 IEEE International Symposium on Biomedical Imaging (ISBI), (Athens, Greece), pp. 1–4, IEEE, May 2024

    work page 2024

  13. [14]

    Classification, Regression and Segmentation Directly from K-Space in Cardiac MRI,

    R. Li, J. Pan, Y. Zhu, J. Ni, and D. Rueckert, “Classification, Regression and Segmentation Directly from K-Space in Cardiac MRI,” inMachine Learning in Medical Imaging(X. Xu, Z. Cui, I. Rekik, X. Ouyang, and K. Sun, eds.), vol. 15241, pp. 31–41, Cham: Springer Nature Switzerland, 2025. Series Title: Lecture Notes in Computer Science

    work page 2025

  14. [15]

    K2S Challenge: From Undersampled K-Space to Automatic Segmentation,

    A. A. Tolpadi, U. Bharadwaj, K. T. Gao, R. Bhattacharjee, F. G. Gassert, J. Luitjens, P. Giesler, J. N. Morshuis, P. Fischer, M. Hein, C. F. Baumgartner, A. Razumov, D. Dylov, Q. V. Lohuizen, S. J. Fransen, X. Zhang, R. Tibrewala, H. L. De Moura, K. Liu, M. V. W. Zibetti, R. Regatte, S. Majumdar, and 11 V. Pedoia, “K2S Challenge: From Undersampled K-Space...

    work page 2023

  15. [16]

    SKM-TEA: A Dataset for Accelerated MRI Reconstruction with Dense Image Labels for Quantitative Clinical Evaluation,

    A. D. Desai, A. M. Schmidt, E. B. Rubin, C. M. Sandino, M. S. Black, V. Mazzoli, K. J. Stevens, R. Boutin, C. Ré, G. E. Gold, B. A. Hargreaves, and A. S. Chaudhari, “SKM-TEA: A Dataset for Accelerated MRI Reconstruction with Dense Image Labels for Quantitative Clinical Evaluation,” 2022. Version Number: 1

    work page 2022

  16. [17]

    Domain Influence in MRI Medical Image Segmentation: Spatial Versus k-Space Inputs,

    E. Gösche, R. Eghbali, F. Knoll, and A. M. Rauschecker, “Domain Influence in MRI Medical Image Segmentation: Spatial Versus k-Space Inputs,” inMachine Learning in Medical Imaging(X. Xu, Z. Cui, I. Rekik, X. Ouyang, and K. Sun, eds.), vol. 15241, pp. 310–319, Cham: Springer Nature Switzerland,

    work page

  17. [18]

    Series Title: Lecture Notes in Computer Science

    work page

  18. [19]

    Encoding and reconstruction in parallel MRI,

    K. P. Pruessmann, “Encoding and reconstruction in parallel MRI,”NMR in Biomedicine, vol. 19, pp. 288–299, May 2006

    work page 2006

  19. [20]

    Parallel MR imaging,

    A. Deshmane, V. Gulani, M. A. Griswold, and N. Seiberlich, “Parallel MR imaging,”Journal of Magnetic Resonance Imaging, vol. 36, pp. 55–72, July 2012

    work page 2012

  20. [21]

    Breast MRI: State of the Art,

    R. M. Mann, N. Cho, and L. Moy, “Breast MRI: State of the Art,”Radiology, vol. 292, pp. 520–536, Sept. 2019

    work page 2019

  21. [22]

    FastMRI Breast: A Publicly Available Radial k-Space Dataset of Breast Dynamic Contrast-enhanced MRI,

    E. Solomon, P. M. Johnson, Z. Tan, R. Tibrewala, Y. W. Lui, F. Knoll, L. Moy, S. G. Kim, and L. Heacock, “FastMRI Breast: A Publicly Available Radial k-Space Dataset of Breast Dynamic Contrast-enhanced MRI,”Radiology: Artificial Intelligence, vol. 7, p. e240345, Jan. 2025

    work page 2025

  22. [23]

    A large-scale multicenter breast cancer DCE-MRI benchmark dataset with expert segmentations,

    L. Garrucho, K. Kushibar, C.-A. Reidel, S. Joshi, R. Osuala, A. Tsirikoglou, M. Bobowicz, J. Del Riego, A. Catanese, K. Gwoździewicz, M.-L. Cosaka, P. M. Abo-Elhoda, S. W. Tantawy, S. S. Sakrana, N. O. Shawky-Abdelfatah, A. M. A. Salem, A. Kozana, E. Divjak, G. Ivanac, K. Nikiforaki, M. E. Klontzas, R. García-Dosdá, M. Gulsun-Akpinar, O. Lafcı, R. Mann, C...

    work page 2025

  23. [24]

    Checklist for Artificial Intelligence in Medical Imaging (CLAIM): A Guide for Authors and Reviewers,

    J. Mongan, L. Moy, and C. E. Kahn, “Checklist for Artificial Intelligence in Medical Imaging (CLAIM): A Guide for Authors and Reviewers,”Radiol. Artif. Intell., vol. 2, p. e200029, Mar. 2020

    work page 2020

  24. [25]

    fastMRI: A Publicly Available Raw k-Space and DICOM Dataset of Knee Images for Accelerated MR Image Reconstruction Using Machine Learning,

    F. Knoll, J. Zbontar, A. Sriram, M. J. Muckley, M. Bruno, A. Defazio, M. Parente, K. J. Geras, J. Katsnelson, H. Chandarana, Z. Zhang, M. Drozdzalv, A. Romero, M. Rabbat, P. Vincent, J. Pinkerton, D. Wang, N. Yakubova, E. Owens, C. L. Zitnick, M. P. Recht, D. K. Sodickson, and Y. W. Lui, “fastMRI: A Publicly Available Raw k-Space and DICOM Dataset of Knee...

    work page 2020

  25. [26]

    ESPIRiT–an eigenvalue approach to autocalibrating parallel MRI: where SENSE meets GRAPPA,

    M. Uecker, P. Lai, M. J. Murphy, P. Virtue, M. Elad, J. M. Pauly, S. S. Vasanawala, and M. Lustig, “ESPIRiT–an eigenvalue approach to autocalibrating parallel MRI: where SENSE meets GRAPPA,” Magn. Reson. Med., vol. 71, pp. 990–1001, Mar. 2014

    work page 2014

  26. [27]

    MONAI: Medical open network for AI,

    MONAI Consortium, “MONAI: Medical open network for AI,” July 2022

    work page 2022

  27. [28]

    A novel automatic segmentation method directly based on magnetic resonance imaging K-space data for auxiliary diagnosis of glioma,

    Y. Li, Y. Qi, Z. Hu, K. Zhang, S. Jia, L. Zhang, W. Xu, S. Shen, Y. X. J. Wáng, Z. Li, D. Liang, X. Liu, H. Zheng, G. Cheng, and N. Zhang, “A novel automatic segmentation method directly based on magnetic resonance imaging K-space data for auxiliary diagnosis of glioma,”Quant. Imaging Med. Surg., vol. 14, pp. 2008–2020, Feb. 2024

    work page 2008

  28. [29]

    Implicit data crimes: Machine learning bias arising from misuse of public data,

    E. Shimron, J. I. Tamir, K. Wang, and M. Lustig, “Implicit data crimes: Machine learning bias arising from misuse of public data,”Proceedings of the National Academy of Sciences, vol. 119, p. e2117203119, Mar. 2022. 12 Supplement S1. Dataset preprocessing fastMRI breast The fastMRI breast dataset provides raw multi-coil radial k-space from contrast-enhanc...

    work page 2022