Pith. sign in

REVIEW 3 major objections 5 minor 76 references

MRpro - open PyTorch-based MR reconstruction and processing package

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read MRpro presents a PyTorch-native framework that unifies classical reconstruction, learned reconstruction, and quantitative parameter estimation for MRI.

desk verdict A well-engineered PyTorch MR reconstruction framework that needs its reproducible notebook artifact completed before the claims fully land. read the letter →

arxiv 2507.23129 v1 pith:WIXX2Z43 submitted 2025-07-30 eess.IV cs.CVphysics.med-ph

Felix Frederik Zimmermann , Patrick Schuenke , Christoph S. Aigner , Bill A. Bernhardt , Mara Guastini , Johannes Hammacher , Noah Jaitner , Andreas Kofler
show 7 more authors
Leonid Lunin Stefan Martin Catarina Redshaw Kranich Jakob Schattenfroh David Schote Yanglei Wu Christoph Kolbitsch
This is my paper · ORCID
classification eess.IVcs.CVphysics.med-ph PACS 87.61.-c
keywords MRimagereconstructionPyTorchopen-sourcesoftwarenon-CartesiantrajectoriesNUFFTquantitativeMRImodel-baseddeeplearningmagneticresonancefingerprinting
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 introduces MRpro, an open-source image reconstruction package built on PyTorch, and argues that a single modular framework can cover classical variational reconstruction, model-based deep learning, and quantitative parameter estimation without forcing researchers to switch tools or rewrite low-level code. The reason this matters is that reproducible comparisons between reconstruction methods are currently hard: methods are often tied to specific sequences, raw-data formats, or C/C++ codebases. MRpro instead standardizes data handling around ISMRMRD, DICOM, and Pulseq, and makes every operator differentiable and GPU-capable. Its key enabling behavior is a Fourier operator that chooses FFT versus NUFFT and the transform dimensions automatically from the stored k-space trajectory, so a raw data file plus a Pulseq sequence file can be turned into a reconstruction with no additional inputs. If this design works as claimed, researchers can focus on algorithms rather than integration plumbing, and published reconstructions can be rerun from open data and executable notebooks.

What carries the argument

The load-bearing object is a unified Fourier operator whose concrete behavior—which axes get a fast Fourier transform, which get a non-uniform FFT, and what density compensation applies—is selected automatically from the trajectory stored in a KData container. Around it sits a LinearOperator algebra supporting composition, stacking, sums, Hermitian adjoints, and Jacobians, together with a library of proximable functionals whose proximal operators feed algorithms such as CG, FISTA/ISTA, and PDHG. Differentiable signal models, including extended phase graph simulation, and data-consistency layers extend the same algebra to quantitative imaging and deep learning, so that classical reconstruction, learned reconstruction, and parameter estimation share one code path.

What would settle it

Feed MRpro a diverse set of valid Pulseq trajectories, including variable-density spirals, rosettes, cones with off-center readouts, and alternating Cartesian and non-Cartesian segments, and check whether the automatically chosen Fourier operator and density compensation reconstruct without artifacts or require a manual override; any trajectory class needing manual correction would falsify the claim of fully automatic selection.

Watch

Extended reading notes

Core claim

The paper's central claim is that MRpro provides an extensible, PyTorch-native pipeline in which the same composable objects serve classical and learned MRI reconstruction as well as quantitative MRI. The demonstration is breadth: Cartesian SENSE reconstruction offline and online at the scanner, free-breathing motion-corrected radial imaging, spiral cardiac MR fingerprinting with extended-phase-graph dictionary matching, learned spatially adaptive TV regularization, MoDL, and end-to-end PINQI are all built from the same operator library and shipped as reproducible notebooks. The organizing insight is that trajectory metadata can itself determine the correct Fourier model, so that reconstructing data from an arbitrary acquisition does not require hand-coded per-sequence trajectory handling.

Load-bearing premise

The load-bearing premise is that the automatic choice of FFT versus non-uniform FFT, and of the dimensions to transform, is correct for every trajectory a user can supply, so that a raw data file plus a Pulseq sequence file always reconstructs without manual input.

Editorial extensions

If this is right

  • A researcher with only an ISMRMRD raw file and a Pulseq sequence file can run standard reconstructions such as CG-SENSE and TV without writing trajectory handling or density compensation.
  • Classical variational methods and deep learning run in the same differentiable graph, which is what allows MoDL, PINQI, and unrolled PDHG to be assembled from the same operator library.
  • Quantitative imaging is part of the same stack: EPG-based dictionary matching produces T1 and T2 maps from spiral cardiac fingerprinting data, with the reported T1MES phantom agreement above R2 = 0.99 against spin-echo references.
  • Because every experiment ships as a notebook with its data on Zenodo, the paper's reconstructions can be re-run as executable examples rather than inspected only as figures.
  • With the OpenRecon container, the same code path performs both offline reconstruction and online reconstruction at the scanner console.

Reading between the lines

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

  • A natural next test is to use MRpro as a neutral benchmark harness: if raw data plus Pulseq becomes a common input, reconstruction methods from different groups could be compared on identical trajectories and identical data without reimplementation.
  • If the automatic trajectory-to-Fourier inference holds broadly, it opens the door to joint sequence-and-reconstruction optimization: a fully differentiable simulator could tune the Pulseq file and the reconstruction network together, an idea the paper touches on only through its compatibility with a differentiable simulator.
  • The authors list ESPIRiT and GRAPPA as not yet included; adding those autocalibrating parallel-imaging methods would let the framework cover most clinical reconstruction pipelines and is a concrete community contribution point.
  • A useful stress test beyond the demonstrated Cartesian, radial, and spiral cases would be to run the fully automatic path on unusual trajectories such as cones, rosettes, or variable-density spirals with off-center readouts.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper introduces MRpro, an open-source image reconstruction and processing package implemented in PyTorch. It describes the package's data structures for ISMRMRD, Pulseq, and DICOM data; a library of composable linear operators, proximable functionals, and optimizers; and PyTorch-based building blocks for deep-learning reconstruction, including data-consistency layers and backbone networks. The authors demonstrate the framework on Cartesian SENSE (offline and via Siemens OpenRecon), motion-corrected 3D radial reconstruction, learned spatially adaptive TV regularization, cardiac MR fingerprinting with EPG-based dictionary matching, MoDL, and PINQI. The paper claims that MRpro is a reproducible and extensible foundation for MR imaging research.

Significance. MRpro addresses a real gap in the MRI software ecosystem: a PyTorch-native, open-format-based reconstruction framework that supports both classical and learning-based methods. The manuscript gives concrete evidence of public code, a CI/CD pipeline with high unit-test coverage, executable notebooks, and Zenodo-hosted datasets. If the reproducibility gaps identified below are fixed, the package would be a useful community resource and the paper a valuable software contribution. The demonstrations are illustrative rather than comprehensive, but they cover a diverse set of reconstruction tasks, and the authors are careful to acknowledge missing features such as ESPIRiT and GRAPPA.

major comments (3)
  1. [Data Availability Statement] The Data Availability Statement explicitly notes that 'At time of submission some notebooks/features are not part of the main branch yet,' and the manuscript does not cite a release version or commit hash for the MRpro repository. Because the central claim of the paper (Abstract; Section 5) is reproducibility, a reader cannot currently execute the exact pipeline that produced the reported SSIM/nRMSE values. Please pin the code to a tagged release (or commit) that contains all notebooks and features, and provide a reproducibility manifest (code revision, dataset DOIs, and environment) so the results in Sections 3.1–3.6 are verifiable.
  2. [Section 2.4] The claim that Fourier operations (FFT vs NUFFT) and their dimensions are 'automatically selected' and that 'data obtained with arbitrary trajectories can be reconstructed from a MR raw data file and the corresponding Pulseq sequence file without any additional inputs' is a central feature, but the selection logic is not described in enough detail to assess its correctness for all common trajectory classes. The experiments cover Cartesian, radial, and spiral, but not, for example, 3D cones or arbitrary 2D readout orientations beyond those demonstrated. Please provide a formal description of the automatic selection algorithm and/or integration tests that cover a broader set of trajectories, or restrict the claim to the tested trajectory classes.
  3. [Sections 3.5 and 3.6] The statements that MoDL 'reproduc[es] the results of Aggarwal et al.' (Section 3.5) and that PINQI 'reproduces the reported performance' (Section 3.6) are supported by a single example each (one fastMRI test slice and one synthetic BrainWeb dataset). These single-instance quantitative results (SSIM 0.95 and 0.93, respectively) do not substantiate a claim of reproduction, which requires agreement across a set of test cases or a reported confidence interval. Please report results over the full test set (or a defined subset) with error bars, or soften the language to 'qualitatively consistent with' the cited works.
minor comments (5)
  1. [Section 2.2] The phrase 'lower semi-continuos' should be 'lower semi-continuous'.
  2. [Section 3.4] The phrase 'quantitive parameter maps' should be 'quantitative parameter maps'.
  3. [Equation (4)] The normalized dot product expression in Equation (4) is typeset poorly; please clarify the absolute value and the inner product notation so the dictionary matching criterion is unambiguous.
  4. [Section 2.5] The acronym 'W ASABI' has an extra space; it should read 'WASABI'.
  5. [Section 3.5] The reference to 'Aggarwar et al.' should be 'Aggarwal et al.' (matching reference 3).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: MRpro is a software framework whose demonstrations reimplement external published algorithms against external benchmarks; self-citations are not load-bearing.

full rationale

MRpro is a software-engineering contribution, not a derivation of new reconstruction theory, so the circularity patterns defined for 'X derives Y' do not directly apply. The paper's demonstrations are implementations of externally published methods evaluated on public or openly released data: MoDL (Ref. 3), learned regularization-parameter maps (Ref. 54), and PINQI (Ref. 34). The only author-group self-citation used as a numerical target is PINQI: Section 3.6 states 'our implementation of PINQI in MRpro reproduces the reported performance 34 with an SSIM of 0.93 and an nRMSE of 0.03.' This is a code-reproduction check against a peer-reviewed, externally citable result whose assumptions and training protocol are restated in the present paper; it is not a prediction derived from a fitted parameter, and it does not define the framework's capability in terms of itself. Likewise, Section 3.5 ('MoDL achieves the highest SSIM of 0.95, reproducing the results of Aggarwal et al. 3') is an external benchmark reproduction. No fitted parameter is renamed a prediction; no uniqueness theorem is imported; no ansatz is smuggled in via citation. Two limitations are present but are not circularity: the Data Availability Statement admits 'At time of submission some notebooks/features are not part of the main branch yet,' and no commit hash or release version is pinned, which weakens the reproducibility guarantee; and Section 2.4's claim of automatic FFT/NUFFT selection for arbitrary Pulseq trajectories is an unproven coverage claim. Both are verifiability and correctness risks, not reductions of outputs to inputs. Verdict: no significant circularity; score 0.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim of the paper is that MRpro is a versatile, reproducible framework. This claim rests on standard MR modeling assumptions rather than on fitted parameters. The free_parameters listed are experimental hyperparameters in the demonstrations; they do not affect the framework's general validity. No new physical entities are introduced.

free parameters (3)
  • TV regularization parameter (scalar lambda) = not reported (oracle line search)
    In Section 3.3, the best scalar lambda is found by line search using ground truth for comparison; this is an experimental hyperparameter, not part of the framework claim.
  • Number of MoDL iterations = 10
    In Section 2.7.5, MoDL uses 10 iterations matching the original paper; this is a fixed hyperparameter, not fitted to MRpro's claim.
  • Dictionary grid for cMRF (T1, T2 ranges) = T1 50 ms to 2 s, T2 6 ms to 200 ms
    In Section 2.7.4, the EPG dictionary grid is chosen by the authors; it affects the cMRF phantom maps but not the framework's general validity.
assumptions (4)
  • domain assumption The linear forward operator A = FourierOp * SensitivityOp (plus motion) accurately models the MR acquisition
    Used throughout Section 2.7 and Equation 1; standard in MR reconstruction.
  • domain assumption NUFFT provides a sufficiently accurate approximation of the non-Cartesian Fourier transform
    The unified Fourier operator relies on NUFFT for radial and spiral trajectories (Section 2.4).
  • domain assumption Extended Phase Graph simulation accurately models the cMRF signal evolution
    Used for dictionary generation in Section 2.7.4; standard in MR fingerprinting.
  • domain assumption PyTorch's autodiff and GPU support are sufficient for the implementation of differentiable optimization layers
    The framework builds entirely on PyTorch (Section 1).

how reviews work

0 comments
Cite this review

Pith. "Pith review of MRpro - open PyTorch-based MR reconstruction and processing package." pith.science (2026). https://pith.science/paper/WIXX2Z43

@misc{pith2026250723129,
  author       = {Pith},
  title        = {Pith review of: MRpro - open PyTorch-based MR reconstruction and processing package},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WIXX2Z43}},
  note         = {Machine review of arXiv:2507.23129}
}
read the original abstract

We introduce MRpro, an open-source image reconstruction package built upon PyTorch and open data formats. The framework comprises three main areas. First, it provides unified data structures for the consistent manipulation of MR datasets and their associated metadata (e.g., k-space trajectories). Second, it offers a library of composable operators, proximable functionals, and optimization algorithms, including a unified Fourier operator for all common trajectories and an extended phase graph simulation for quantitative MR. These components are used to create ready-to-use implementations of key reconstruction algorithms. Third, for deep learning, MRpro includes essential building blocks such as data consistency layers, differentiable optimization layers, and state-of-the-art backbone networks and integrates public datasets to facilitate reproducibility. MRpro is developed as a collaborative project supported by automated quality control. We demonstrate the versatility of MRpro across multiple applications, including Cartesian, radial, and spiral acquisitions; motion-corrected reconstruction; cardiac MR fingerprinting; learned spatially adaptive regularization weights; model-based learned image reconstruction and quantitative parameter estimation. MRpro offers an extensible framework for MR image reconstruction. With reproducibility and maintainability at its core, it facilitates collaborative development and provides a foundation for future MR imaging research.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

76 extracted references · 71 canonical work pages

  1. [1]

    SENSE: sensitivity encoding for fast MRI

    Pruessmann Klaas P., Weiger Markus, Scheidegger Markus B., Boesiger Peter. SENSE: sensitivity encoding for fast MRI. Magnetic Resonance Imaging. 1999;42:952– 962

    work page 1999

  2. [2]

    Generalized autocalibrating partially parallel acquisitions (GRAPPA)

    Griswold Mark A., Jakob Peter M., Heidemann Robin M., et al. Generalized autocalibrating partially parallel acquisitions (GRAPPA). Magnetic Resonance in Medicine. 2002;47:1202–1210

    work page 2002

  3. [3]

    MoDL: Model-based deep learning architecture for inverse problems

    Aggarwal Hemant K., Mani Merry P., Jacob Math- ews. MoDL: Model-based deep learning architecture for inverse problems. IEEE Transactions on Medical Imag- ing. 2018;38(2):394–405

    work page 2018

  4. [4]

    Image Reconstruction: From Sparsity to Data-Adaptive Methods and Machine Learning

    Ravishankar Saiprasad, Ye Jong Chul, Fessler Jeffrey A.. Image Reconstruction: From Sparsity to Data-Adaptive Methods and Machine Learning. Proceedings of the IEEE. 2020;108:86–109

    work page 2020

  5. [5]

    End-to-end variational networks for accelerated MRI reconstruction

    Sriram Anuroop, Zbontar Jure, Murrell Tullie, et al. End-to-end variational networks for accelerated MRI reconstruction. In: International conference on medi- cal image computing and computer-assisted intervention 2020:64–73

    work page 2020

  6. [6]

    Robust compressed sensing mri with deep generative priors.Advances in neu- ral information processing systems

    Jalal Ajil, Arvinte Marius, Daras Giannis, Price Eric, Dimakis Alexandros G., Tamir Jon. Robust compressed sensing mri with deep generative priors.Advances in neu- ral information processing systems. 2021;34:14938–14954

    work page 2021

  7. [7]

    AI-Based Reconstruction for Fast MRI-A System- atic Review and Meta-Analysis

    Chen Yutong, Schonlieb Carola Bibiane, Lio Pietro, et al. AI-Based Reconstruction for Fast MRI-A System- atic Review and Meta-Analysis. Proceedings of the IEEE. 2022;110:224–245

    work page 2022

  8. [8]

    ISMRM Raw data format: A proposed standard for MRI raw datasets

    Inati Souheil J., Naegele Joseph D., Zwart Nicholas R., et al. ISMRM Raw data format: A proposed standard for MRI raw datasets. Magnetic Resonance in Medicine. 2017;77:411–421

    work page 2017

Show all 76 references
  1. [9]

    Gadgetron: an open source framework for medical image reconstruction

    Hansen Michael Schacht, Sørensen Thomas Sangild. Gadgetron: an open source framework for medical image reconstruction. Magnetic Resonance in Medicine. 2013;69(6):1768–1776

    2013

  2. [10]

    Christian M., Uecker Martin

    Blumenthal Moritz, Luo Guanxiong, Schilling Martin, Holme H. Christian M., Uecker Martin. Deep, deep learning with BART. Magnetic Resonance in Medicine. 2023;89(2):678–693

    2023

  3. [11]

    PyTorch: an imperative style, high-performance deep learning library

    Paszke Adam, Gross Sam, Massa Francisco, et al. PyTorch: an imperative style, high-performance deep learning library. Red Hook, NY, USA: Curran Associates Inc. 2019

    2019

  4. [12]

    Com- piling machine learning programs via high-level tracing

    Frostig Roy, Johnson Matthew James, Leary Chris. Com- piling machine learning programs via high-level tracing. Systems for Machine Learning. 2018;4(9)

    2018

  5. [13]

    Pulseq: A Rapid and Hardware-Independent Pulse Sequence Pro- totyping Framework

    Layton Kelvin J., Kroboth Stefan, Jia Feng, et al. Pulseq: A Rapid and Hardware-Independent Pulse Sequence Pro- totyping Framework. Magnetic Resonance in Medicine. 2017;77(4):1544–1552

    2017

  6. [14]

    Advances in sensitivity encoding with arbitrary k-space trajectories

    Pruessmann Klaas P., Weiger Markus, B¨ ornert Peter, Boesiger Peter. Advances in sensitivity encoding with arbitrary k-space trajectories. Magnetic Resonance in Medicine. 2001;46(4):638–651

    2001

  7. [15]

    Parallel MR Image Reconstruction Using Augmented Lagrangian Methods

    Ramani Sathish, Fessler Jeffrey A.. Parallel MR Image Reconstruction Using Augmented Lagrangian Methods. IEEE Transactions on Medical Imaging. 2011;30(3):694– 706

    2011

  8. [16]

    An iterative thresholding algorithm for linear inverse problems with a sparsity constraint

    Daubechies Ingrid, Defrise Michel, De Mol Christine. An iterative thresholding algorithm for linear inverse problems with a sparsity constraint. Communications on Pure and Applied Mathematics: A Journal Issued by the Courant Institute of Mathematical Sciences. 2004;57(11):1413–1457

    2004

  9. [17]

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

    Lustig Michael, Donoho David, Pauly John M.. Sparse MRI: The application of compressed sensing for rapid MR imaging. Magnetic Resonance Imaging. 2007;58:1182– 1195

    2007

  10. [18]

    A first-order primal- dual algorithm for convex problems with applications to imaging

    Chambolle Antonin, Pock Thomas. A first-order primal- dual algorithm for convex problems with applications to imaging. Journal of mathematical imaging and vision. 2011;40:120–145

    2011

  11. [19]

    Fast reduc- tion of undersampling artifacts in radial MR angiography with 3D total variation on graphics hardware

    Knoll Florian, Unger Markus, Diwoky Clemens, Clason Christian, Pock Thomas, Stollberger Rudolf. Fast reduc- tion of undersampling artifacts in radial MR angiography with 3D total variation on graphics hardware. Magnetic Resonance Materials in Physics, Biology and Medicine. 201...

    2010

  12. [20]

    A Fast Optimal Method for Coil Sensitivity Estimation and Adaptive Coil Combination for Complex Images

    Inati Souheil J., Hansen Michael S., Kellman Peter. A Fast Optimal Method for Coil Sensitivity Estimation and Adaptive Coil Combination for Complex Images. In: Proceedings of the 22nd Annual Meeting of ISMRM 2014

    2014

  13. [21]

    Adaptive reconstruction of phased array MR imagery

    Walsh David O., Gmitro Arthur F., Marcellin Michael W.. Adaptive reconstruction of phased array MR imagery. Magnetic Resonance in Medicine. 2000;43(5):682–690

    2000

  14. [22]

    A software channel compression technique for faster reconstruction with many channels

    Huang Feng, Vijayakumar Sathya, Li Yu, Hertel Sarah, Duensing George R.. A software channel compression technique for faster reconstruction with many channels. Magnetic resonance imaging. 2008;26(1):133–141

    2008

  15. [23]

    U-net: Convolutional networks for biomedical image segmen- tation

    Ronneberger Olaf, Fischer Philipp, Brox Thomas. U-net: Convolutional networks for biomedical image segmen- tation. Medical image computing and computer-assisted intervention (MICCAI). 2015;part III 18:234–241

    2015

  16. [24]

    Atten- tion U-Net: Learning Where to Look for the Pancreas

    Oktay Ozan, Schlemper Jo, Le Folgoc Loic, et al. Atten- tion U-Net: Learning Where to Look for the Pancreas. In: Medical Imaging with Deep Learning

  17. [25]

    Restormer: Efficient transformer for high-resolution image restoration

    Zamir Syed Waqas, Arora Aditya, Khan Salman, Hayat Munawar, Khan Fahad Shahbaz, Yang Ming-Hsuan. Restormer: Efficient transformer for high-resolution image restoration. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition 2022:5728–5739

    2022

  18. [26]

    Uformer: A general u-shaped transformer for image restoration

    Wang Zhendong, Cun Xiaodong, Bao Jianmin, Zhou Wengang, Liu Jianzhuang, Li Houqiang. Uformer: A general u-shaped transformer for image restoration. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition 2022:17683–17693. 12 Zimmermann et al

    2022

  19. [27]

    Swinir: Image restoration using swin transformer

    Liang Jingyun, Cao Jiezhang, Sun Guolei, Zhang Kai, Van Gool Luc, Timofte Radu. Swinir: Image restoration using swin transformer. In:Proceedings of the IEEE/CVF international conference on computer vision 2021:1833– 1844

    2021

  20. [28]

    Deep Com- pression Autoencoder for Efficient High-Resolution Diffu- sion Models

    Chen Junyu, Cai Han, Chen Junsong, et al. Deep Com- pression Autoencoder for Efficient High-Resolution Diffu- sion Models. In: The Thirteenth International Conference on Learning Representations 2024

    2024

  21. [29]

    Scalable high-resolution pixel-space image synthesis with hourglass diffusion transformers

    Crowson Katherine, Baumann Stefan Andreas, Birch Alex, Abraham Tanishq Mathew, Kaplan Daniel Z., Ship- pole Enrico. Scalable high-resolution pixel-space image synthesis with hourglass diffusion transformers. In: Pro- ceedings of the 41st International Conference on Machine Lea...

    2024

  22. [30]

    FiLM: Visual reasoning with a general conditioning layer

    Perez Ethan, Strub Florian, De Vries Harm, Dumoulin Vincent, Courville Aaron. FiLM: Visual reasoning with a general conditioning layer. 32nd AAAI Conference on Artificial Intelligence, AAAI 2018. 2018;:3942–3951

    2018

  23. [31]

    A Deep Cascade of Con- volutional Neural Networks for Dynamic MR Image Reconstruction

    Schlemper Jo, Caballero Jose, Hajnal Joseph V., Price Anthony N., Rueckert Daniel. A Deep Cascade of Con- volutional Neural Networks for Dynamic MR Image Reconstruction. IEEE Transactions on Medical Imaging. 2018;37(2):491–503

    2018

  24. [32]

    Shanno D. F.. Conditioning of Quasi-Newton Methods for Function Minimization.Mathematics of Computation. 1970;24(111):647–656

    1970

  25. [33]

    Amos Brandon, Kolter J. Zico. Optnet: Differentiable optimization as a layer in neural networks. In: Interna- tional conference on machine learning 2017:136–145

    2017

  26. [34]

    PINQI: An End-to-End Physics- Informed Approach to Learned Quantitative MRI Recon- struction

    Zimmermann Felix F., Kolbitsch Christoph, Schuenke Patrick, Kofler Andreas. PINQI: An End-to-End Physics- Informed Approach to Learned Quantitative MRI Recon- struction. IEEE Transactions on Computational Imag- ing. 2024;10:628–639

    2024

  27. [35]

    fastMRI: An open dataset and benchmarks for acceler- ated MRI

    Zbontar Jure, Knoll Florian, Sriram Anuroop, et al. fastMRI: An open dataset and benchmarks for acceler- ated MRI. arXiv preprint arXiv:1811.08839. 2018

    2018 arXiv

  28. [36]

    Bruce, Evans Alan C., Collins D

    Aubert-Broche Bereng` ere, Griffin Mark, Pike G. Bruce, Evans Alan C., Collins D. Louis. Twenty new digi- tal brain phantoms for creation of validation image data bases. IEEE Transactions on Medical Imaging. 2006;25(11):1410–1416

    2006

  29. [37]

    M4Raw: A multi-contrast, multi-repetition, multi-channel MRI k- space dataset for low-field MRI research

    Lyu Mengye, Mei Lifeng, Huang Shoujin, et al. M4Raw: A multi-contrast, multi-repetition, multi-channel MRI k- space dataset for low-field MRI research. Scientific Data. 2023;10(1):264

    2023

  30. [38]

    Multi-domain convolutional neural network (MD-CNN) for radial reconstruction of dynamic cardiac MRI

    El-Rewaidy Hossam, Fahmy Ahmed S., Pashakhanloo Farhad, et al. Multi-domain convolutional neural network (MD-CNN) for radial reconstruction of dynamic cardiac MRI. Magnetic Resonance in Medicine. ;85(3):1195– 1208

  31. [39]

    Modified Look-Locker Inversion Recov- ery (MOLLI) for High-Resolution T 1 Mapping of the Heart

    Messroghli Daniel R., Radjenovic Aleksandra, Kozerke Sebastian, Higgins David M., Sivananthan Mohan U., Ridgway John P.. Modified Look-Locker Inversion Recov- ery (MOLLI) for High-Resolution T 1 Mapping of the Heart. Magnetic Resonance in Medicine. 2004;52(1):141– 146

    2004

  32. [40]

    Simultaneous Mapping of Water Shift and B1 (W ASABI)-Application to Field-Inhomogeneity Correction of CEST MRI Data

    Schuenke Patrick, Windschuh Johannes, Roeloffs Volkert, Ladd Mark E., Bachert Peter, Zaiss Moritz. Simultaneous Mapping of Water Shift and B1 (W ASABI)-Application to Field-Inhomogeneity Correction of CEST MRI Data. Magnetic Resonance in Medicine. 2017;77(2):571–580

    2017

  33. [41]

    Extended phase graphs: Dephasing, RF pulses, and echoes - Pure and simple.Journal of Magnetic Resonance Imaging

    Weigel Matthias. Extended phase graphs: Dephasing, RF pulses, and echoes - Pure and simple.Journal of Magnetic Resonance Imaging. 2015;41:266–295

    2015

  34. [42]

    MR fingerprinting for rapid quantification of myocardial T1, T2, and proton spin density

    Hamilton Jesse I., Jiang Yun, Chen Yong, et al. MR fingerprinting for rapid quantification of myocardial T1, T2, and proton spin density. Magnetic Resonance in Medicine. 2017;77:1446–1458

    2017

  35. [43]

    Open-source cardiac magnetic resonance finger- printing

    Schuenke Patrick, Redshaw Kranich Catarina, Lutz Max, et al. Open-source cardiac magnetic resonance finger- printing. Magnetic Resonance Materials in Physics, Biol- ogy and Medicine. 2025

    2025

  36. [44]

    Adam: A Method for Stochastic Optimization

    Kingma Diederik P., Ba Jimmy. Adam: A Method for Stochastic Optimization. In: Bengio Yoshua, LeCun Yann, eds.. 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings 2015

    2015

  37. [45]

    2D Cartesian MR raw data [dataset]

    Kolbitsch Christoph. 2D Cartesian MR raw data [dataset]. Zenodo, 2025, doi: 10.5281/zenodo.15223816

    2025 doi

  38. [46]

    Motion Artifacts in MRI: A Complex Problem with Many Par- tial Solutions

    Zaitsev Maxim, Maclaren Julian, Herbst Michael. Motion Artifacts in MRI: A Complex Problem with Many Par- tial Solutions. Journal of Magnetic Resonance Imaging. 2015;42:887–901

    2015

  39. [47]

    Batchelor Philip G., Atkinson David, Irarrazaval Pablo, Hill D. L. G., Hajnal Joseph, Larkman D.. Matrix Description of General Motion Correction Applied to Multishot Images. Magnetic Resonance in Medicine. 2005;54:1273–1280

    2005

  40. [48]

    3D Undersampled Golden- Radial Phase Encoding for DCE-MRA Using Inherently Regularized Iterative SENSE

    Prieto Claudia, Uribe Sergio, Razavi Reza, Atkinson David, Schaeffter Tobias. 3D Undersampled Golden- Radial Phase Encoding for DCE-MRA Using Inherently Regularized Iterative SENSE. Magnetic Resonance in Medicine. 2010;64:514–526

    2010

  41. [49]

    Nonrigid Motion Mod- eling of the Liver From 3-D Undersampled Self-Gated Golden-Radial Phase Encoded MRI

    Buerger Christian, Clough Rachel E., King Andrew P., Schaeffter Tobias, Prieto Claudia. Nonrigid Motion Mod- eling of the Liver From 3-D Undersampled Self-Gated Golden-Radial Phase Encoded MRI. IEEE Transactions on Medical Imaging. 2012;31:805–815

    2012

  42. [50]

    Free-breathing MRI of the liver [dataset]

    Kolbitsch Christoph, Schattenfroh Jakob, Jaitner Noah. Free-breathing MRI of the liver [dataset]. Zenodo, 2025, doi: 10.5281/zenodo.15288250

    2025 doi

  43. [51]

    Suppres- sion of MRI Truncation Artifacts Using Total Variation Constrained Data Extrapolation

    Block Kai Tobias, Uecker Martin, Frahm Jens. Suppres- sion of MRI Truncation Artifacts Using Total Variation Constrained Data Extrapolation.. Int J Biomed Imaging. 2008;2008:184123

    2008

  44. [52]

    I., Hayes C., Hill D

    Rueckert D., Sonoda L. I., Hayes C., Hill D. L. G., Leach M. O., Hawkes D. J.. Nonrigid Registration Using Free- Form Deformations: Application to Breast MR Images. IEEE Transactions on Medical Imaging. 1999;18:712– 721

    1999

  45. [53]

    On and beyond total variation regu- larization in imaging: the role of space variance

    Pragliola Monica, Calatroni Luca, Lanza Alessandro, Sgallari Fiorella. On and beyond total variation regu- larization in imaging: the role of space variance. SIAM Review. 2023;65(3):601–685

    2023

  46. [54]

    Learning regularization parameter-maps for varia- tional image reconstruction using deep neural networks Zimmermann et al 13 and algorithm unrolling

    Kofler Andreas, Altekr¨ uger Fabian, Antarou Ba Fatima, et al. Learning regularization parameter-maps for varia- tional image reconstruction using deep neural networks Zimmermann et al 13 and algorithm unrolling. SIAM Journal on Imaging Sci- ences. 2023;16:2202–2246

    2023

  47. [55]

    In vivo 3D brain and extremity MRI at 50 mT using a permanent mag- net Halbach array

    O’Reilly Thomas, Teeuwisse Wouter M., Gans Danny, Koolstra Kirsten, Webb Andrew G.. In vivo 3D brain and extremity MRI at 50 mT using a permanent mag- net Halbach array. Magnetic resonance in medicine. 2021;85(1):495–505

    2021

  48. [56]

    Open-source magnetic resonance imaging: improving access, science, and education through global collabora- tion

    Winter Lukas, Periquito Joao, Kolbitsch Christoph, et al. Open-source magnetic resonance imaging: improving access, science, and education through global collabora- tion. NMR in Biomedicine. 2024;37(7):e5052

    2024

  49. [57]

    Nexus: A Versatile Console for Advanced Low-fieldMRI

    Schote David, Silemek Berk, O’Reilly Thomas, et al. Nexus: A Versatile Console for Advanced Low-fieldMRI. Magnetic Resonance in Medicine. 2025;93(5):2224–2238

    2025

  50. [58]

    Magnetic resonance fingerprinting

    Ma Dan, Gulani Vikas, Seiberlich Nicole, et al. Magnetic resonance fingerprinting.. Nature. 2013;495:187–192

    2013

  51. [59]

    A medical device-grade T1 and ECV phantom for global T1 mapping quality assurance - the T1 Mapping and ECV Standardization in cardiovascular magnetic res- onance (T1MES) program

    Captur Gabriella, Gatehouse Peter, Keenan Kathryn E., et al. A medical device-grade T1 and ECV phantom for global T1 mapping quality assurance - the T1 Mapping and ECV Standardization in cardiovascular magnetic res- onance (T1MES) program. Journal of Cardiovascular Magnetic Re...

    2016

  52. [60]

    Open-Source Cardiac MR Fingerprinting [dataset]

    Redshaw Kranich Catarina. Open-Source Cardiac MR Fingerprinting [dataset]. Zenodo, 2025, doi: 10.5281/zen- odo.15726937

    2025 doi

  53. [61]

    Image quality assessment: From error visibility to structural similarity

    Wang Zhou, Bovik Alan Conrad, Sheikh Hamid Rahim, Simoncelli Eero P.. Image quality assessment: From error visibility to structural similarity. IEEE Transactions on Image Processing. 2004;13(4):600–612

    2004

  54. [62]

    Operator Dis- cretization Library (ODL)

    Adler Jonas, Kohr Holger, ¨Oktem Ozan. Operator Dis- cretization Library (ODL). 2017

    2017

  55. [63]

    DeepInverse: A Python package for solving imag- ing inverse problems with deep learning

    Tachella Juli´ an, Terris Matthieu, Hurault Samuel, et al. DeepInverse: A Python package for solving imag- ing inverse problems with deep learning. arXiv preprint arXiv:2505.20160. 2025

    2025 arXiv

  56. [64]

    SigPy: A Python Pack- age for High-Performance Iterative Reconstruction

    Ong Frank, Lustig Michael. SigPy: A Python Pack- age for High-Performance Iterative Reconstruction. In: Proceedings of the ISMRM 27th Annual Meeting 2019

    2019

  57. [65]

    DIRECT: Deep Image REConstruction Toolkit

    Yiasemis George, Moriakov Nikita, Karkalousos Dim- itrios, Caan Matthan, Teuwen Jonas. DIRECT: Deep Image REConstruction Toolkit. Journal of Open Source Software. 2022;7(73):4278

    2022

  58. [66]

    Jorge, Li Wenqi, Brown Richard, et al

    Cardoso M. Jorge, Li Wenqi, Brown Richard, et al. MONAI: An open-source framework for deep learning in healthcare. 2022

    2022

  59. [67]

    L., Graham Mark S., Kerfoot Eric, et al

    Pinaya Walter H. L., Graham Mark S., Kerfoot Eric, et al. Generative AI for medical imaging: extending the MONAI framework. arXiv preprint arXiv:2307.15208. 2023

    2023 arXiv

  60. [68]

    TorchIO: a Python library for efficient loading, preprocessing, augmentation and patch-based sampling of medical images in deep learning

    P´ erez-Garc ´ ıa Fernando, Sparks Rachel, Ourselin S´ ebastien. TorchIO: a Python library for efficient loading, preprocessing, augmentation and patch-based sampling of medical images in deep learning. Computer Methods and Programs in Biomedicine. 2021;:106236

    2021

  61. [69]

    PyQMRI: An accelerated Python based Quantitative MRI toolbox

    Maier Oliver, Spann Stefan M., B¨ odenler Markus, Stoll- berger Rudolf. PyQMRI: An accelerated Python based Quantitative MRI toolbox. Journal of Open Source Soft- ware. 2020;5(56):2727

    2020

  62. [70]

    qMRLab: Quantitative MRI analysis, under one umbrella

    Karakuzu Agah, Boudreau Mathieu, Duval Tanguy, et al. qMRLab: Quantitative MRI analysis, under one umbrella. Journal of Open Source Software. 2020;5(53)

    2020

  63. [71]

    MRI acquisition and reconstruction cookbook: recipes for reproducibil- ity, served with real-world flavour

    Tamir Jonathan I., Blumenthal Moritz, Wang Jiachen, Oved Tal, Shimron Efrat, Zaiss Moritz. MRI acquisition and reconstruction cookbook: recipes for reproducibil- ity, served with real-world flavour.. Magnetic Resonance Materials in Physics, Biology and Medicine. 2025

    2025

  64. [72]

    Quantitative Study Group: T1 [dataset]

    Tatman Ben. Quantitative Study Group: T1 [dataset]. Zenodo, 2024, doi: 10.5281/zenodo.10868350

    2024 doi

  65. [73]

    Quantitative Study Group: T2* [dataset]

    Tatman Ben. Quantitative Study Group: T2* [dataset]. Zenodo, 2024, doi: 10.5281/zenodo.10868361

    2024 doi

  66. [74]

    MRzero - Automated discovery of MRI sequences using supervised learning

    Loktyushin A., Herz K., Dang N., et al. MRzero - Automated discovery of MRI sequences using supervised learning. Magnetic Resonance in Medicine. 2021;86:709– 724

    2021

  67. [75]

    Ten- sorflow: Large-scale machine learning on heterogeneous distributed systems

    Abadi Mart ´ ın, Agarwal Ashish, Barham Paul, et al. Ten- sorflow: Large-scale machine learning on heterogeneous distributed systems. arXiv preprint arXiv:1603.04467. 2016

    2016 arXiv

  68. [76]

    ESPIRiT-an Eigenvalue Approach to Autocalibrating Parallel MRI: Where SENSE Meets GRAPPA..Magnetic Resonance Imaging

    Uecker Martin, Lai Peng, Murphy Mark J., et al. ESPIRiT-an Eigenvalue Approach to Autocalibrating Parallel MRI: Where SENSE Meets GRAPPA..Magnetic Resonance Imaging. 2013

    2013

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.