BART Online Open-Source Sequence Toolbox for Computational MRI

arXiv:2607.19099 · physics.med-ph · Submitted 2026-07-21 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.

Kai: We're looking at the paper: "BART Online Open-Source Sequence Toolbox for Computational MRI".

Mira: Here's the gist of the paper. In this work, an end-to-end open-source framework for sequence design and model-based reconstruction based BART can be used to develop advanced qMRI techniques Online integration with a vendor-specific interpreter enables online adjustment of sequence parameters, while Pulseq export allows an exact offline reproduction of the sequence Framework Integration and Capabilities The paper introduces BART as an open-source toolbox for computational MRI and extends it to include a sequence development framework This integrated framework allows for the development of pulse sequences, including a vendor-specific driver sequence that enables online adjustment of all relevant sequence parameters Sequences can be exported to Pulseq files, which are used to reproduce the exact same sequence offline using a widely used vendor-neutral open-source standard The framework supports quantitative MRI methods for T1 and joint water/fat R∗2, B0 mapping using radial FLASH and model-based reconstruction as proof-of-concept Sequence Definition and Event Composition The specific configuration of an MRI sequence depends on a high-level parameter set comprising fixed system parameters as well as adjustable parameters Sequences are composed from different event types, namely gradient events, RF pulses, sampling (Analog-to-digital converter, ADC) periods, wait and trigger events Gradients are decomposed into triangles to describe their amplitude at midtime in phase-, read- and slice-encoding direction RF events include a pulse with flip angle, phase, frequency shift, and a mapping to a pulse shape ADC events are defined with a dwell-time and number of samples, as well as frequency and phase information A set of events is combined into a sequence block, which are classified into magnetic preparation or imaging blocks Online Acquisition Workflow For online acquisition, BART is compiled as a dynamic library and linked with a vendor-specific driver sequence that can interpret the generated sequence events The interface is designed in a generalized way to completely disentangle the actual sequence logic from the driver sequence Vendor-specific user interface (UI) parameters are mapped to sequence parameters using a generalized interface that also allows the definition of custom parameters During execution, sequence events are computed just-in-time and interpreted/converted to actual vendor-specific hardware instructions for every sequence block A log file containing a synthesized command-line is automatically saved each time a sequence is started for reproducibility Model-Based Reconstruction Validation Consistency between online and offline acquisition is validated in phantom and in vivo experiments by comparing the results when using the online driver sequence with results obtained when reproducing the acquisition using exported Pulseq files Quantitative parameter maps from model-based reconstruction agree for both online and offline regenerated Pulseq acquisitions The framework enables reproducibility of advanced computational MRI methods within a comprehensive end-to-end open-source framework Model-based reconstructions of a FLASH sequence utilizing 15 echoes show good agreement in the parameter maps for both R⋆2 and B0 maps between Pulseq and online measurements Good agreement between Pulseq and online acquisitions can be observed for both quantitative maps demonstrated by low differences across the whole brain Reproducibility and Maintenance The integration into the development workflow of the BART toolbox ensures long-term reproducibility and maintenance for sequences A continuous integration pipeline is used to automatically run tests when the software is changed to be able to identify and fix bugs early on The code to reproduce the results of this paper is available at https://gitlab.tugraz.at/ibi/mrirecon/papers/bart-sequence BART is available at https://codeberg.org/mrirecon/bart The Pulseq interpreter can be obtained from the University of Freiburg The BART Driver Sequence (BOOST: BART Online Open-Source Sequence Toolbox) can be obtained from the Graz University of Technology via Siemens Healthineers’ C2P platform An end-to-end open-source framework for sequence design and model-based reconstruction based BART can be used to develop advanced qMRI techniques Online integration with a vendor-specific interpreter enables online adjustment of sequence parameters, while Pulseq export allows an exact offline reproduction of the sequence Conflict of Interest The authors declare no competing interests Data Availability Statement In the spirit of reproducible research, the code to reproduce the results of this paper is available at https://gitlab.tugraz.at/ibi/mrirecon/bart-sequence BART is continuously being developed and maintained since 2013 The Institute of Biomedical Imaging at Graz University of Technology and BART developers are committed to supporting the toolbox in the future Acknowledgement This work has been supported by the DZHK (German Centre for Cardiovascular Research), funding code: 81Z0300115 This research was funded in 24 part by NIH under grant U24EB029240 This research was funded in whole or in part by the Austrian Science Fund (FWF) 10.55776/F100800 The authors thank the ISMRM Reproducible Research Study Group for conducting a code review of the code (Version 0.1) supplied in the Data Availability Statement This research was funded in whole or in part by the Austrian Science Fund (FWF) 10.55776/F100800 The authors thank the ISMRM Reproducible Research Study Group for conducting a code review of the code (Version 0.1) supplied in the Data Availability Statement The scope of the code review covered only the code’s ease of download, quality of documentation, and ability to run, but did not consider scientific accuracy or code efficiency The Institute of Biomedical Imaging at Graz University of Technology and BART developers are committed to supporting the toolbox in the future References

1: Stikov N, Trzasko JD, Bernstein MA.

Kai: Reproducibility and the future of MRI research.

Mira: Magnetic Resonance in Medicine 2019; 82:1981–1983

2: Knoll F, Zbontar J, Sriram A, et al.

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

Paper summary: Kai: So to wrap up on "BART Online Open-Source Sequence Toolbox for Computational MRI," it’s this end-to-end open-source framework that lets you design sequences and reconstruct images using a model.

Mira: It claims that online integration with a vendor-specific interpreter allows you to adjust sequence parameters while acquiring data, and the Pulseq export lets you reproduce the exact same sequence offline using an open standard two <ref:2607.19099#pg1>.

Lev: The implication for users is that they get reproducibility of advanced computational MRI methods, which was a major challenge because pulse sequences are usually proprietary three <ref:2607.19099#pg1>.

Kai: It’s about creating a tool that meets those demands for long-time reproducibility and maintenance in clinical settings one <ref:2607.19099#pg1,long-time reproducibility and maintenance>.

Mira: Essentially, it gives researchers an open path to develop and share these complex techniques without being locked into one vendor's system two <ref:2607.19099#pg1>.

Conclusion: Kai: So we're wrapping up on this BART Open-Source Sequence Toolbox paper by Stikov and colleagues.

Mira: It’s about taking a whole framework for sequence design and model based reconstruction, and making it open source so everyone can use it.

Lev: The core idea is this end to end system that lets you build sequences and then reconstruct images using a model.

Kai: And the big thing they’re showing is how you can run that sequence online while also having a vendor specific tool adjust the parameters on the fly.

Mira: Plus, you get this Pulseq export file so you can take those exact parameters offline and reproduce the acquisition perfectly using a standard format.

Lev: So it’s about connecting the online adjustment capability with an offline reproduction capability for quantitative MRI.

Kai: It seems like they’re really focusing on making advanced computational MRI techniques reproducible without being locked into one specific machine setup.

Mira: They show this consistency in phantom and in vivo experiments when you compare the online acquisition results with the offline reconstruction using those exported Pulseq files.

Lev: For me, that validation is key because it means the model based reconstructions for things like T1 and joint water fat R star two B zero mapping actually agree between the two methods.

Kai: I mean, they even show good agreement in the parameter maps for both R star two and B zero maps when using a radial FLASH sequence with fifteen echoes.

Mira: It proves that this whole framework can reliably support quantitative MRI methods for T1 and joint water fat R star two B zero mapping.

Lev: It’s a solid proof of concept, but the real question is how much overhead this online dynamic library adds to the actual running time on real hardware.

Kai: That's what we need to look at next—the actual performance trade offs when you run this kind of online sequence generation.

Institute of Biomedical Imaging, Graz University of Technology · Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital Department of Radiology · Department of Radiology, Harvard Medical School Department of Radiology German Centre for Cardiovascular Research Partner Site Göttingen BioTechMed-Graz

physics.med-ph

Submitted: 2026-07-21

Updated: 2026-11-10

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 98/100

The gist: In this work, an end-to-end open-source framework for sequence design and model-based reconstruction based BART can be used to develop advanced qMRI techniques Online integration with a…

Key concepts

BART Framework
BART is an end-to-end open-source toolbox for computational MRI that allows users to develop pulse sequences. It supports quantitative MRI methods like T1 and R*2 mapping by defining sequences through a set of structured events, enabling the creation of complex imaging protocols from scratch.
Online Acquisition Workflow
For real-time imaging, BART is compiled as a dynamic library linked with a vendor-specific driver. This setup lets users adjust sequence parameters online using a vendor interface. Sequence events are computed just-in-time and converted into hardware instructions for immediate execution.
Pulseq Export
Sequences can be exported to Pulseq files, which are a widely used, vendor-neutral standard. This export enables the exact offline reproduction of the sequence using external software. This ensures that results obtained online can be verified by running the same sequence offline.
Sequence Events
MRI sequences are built from fundamental events: gradient events (describing magnetic field changes), RF pulses (defining excitation), sampling periods (ADC information), and wait/trigger events. These individual components are combined into blocks to form a complete imaging sequence.

Terminology

Summary

In this work, an end-to-end open-source framework for sequence design and model-based reconstruction based BART can be used to develop advanced qMRI techniques Online integration with a vendor-specific interpreter enables online adjustment of sequence parameters, while Pulseq export allows an exact offline reproduction of the sequence

Framework Integration and Capabilities

The paper introduces BART as an open-source toolbox for computational MRI and extends it to include a sequence development framework This integrated framework allows for the development of pulse sequences, including a vendor-specific driver sequence that enables online adjustment of all relevant sequence parameters Sequences can be exported to Pulseq files, which are used to reproduce the exact same sequence offline using a widely used vendor-neutral open-source standard The framework supports quantitative MRI methods for T1 and joint water/fat R∗2, B0 mapping using radial FLASH and model-based reconstruction as proof-of-concept

Sequence Definition and Event Composition

The specific configuration of an MRI sequence depends on a high-level parameter set comprising fixed system parameters as well as adjustable parameters Sequences are composed from different event types, namely gradient events, RF pulses, sampling (Analog-to-digital converter, ADC) periods, wait and trigger events Gradients are decomposed into triangles to describe their amplitude at midtime in phase-, read- and slice-encoding direction RF events include a pulse with flip angle, phase, frequency shift, and a mapping to a pulse shape ADC events are defined with a dwell-time and number of samples, as well as frequency and phase information A set of events is combined into a sequence block, which are classified into magnetic preparation or imaging blocks

Online Acquisition Workflow

For online acquisition, BART is compiled as a dynamic library and linked with a vendor-specific driver sequence that can interpret the generated sequence events The interface is designed in a generalized way to completely disentangle the actual sequence logic from the driver sequence Vendor-specific user interface (UI) parameters are mapped to sequence parameters using a generalized interface that also allows the definition of custom parameters During execution, sequence events are computed just-in-time and interpreted/converted to actual vendor-specific hardware instructions for every sequence block A log file containing a synthesized command-line is automatically saved each time a sequence is started for reproducibility

Model-Based Reconstruction Validation

Consistency between online and offline acquisition is validated in phantom and in vivo experiments by comparing the results when using the online driver sequence with results obtained when reproducing the acquisition using exported Pulseq files Quantitative parameter maps from model-based reconstruction agree for both online and offline regenerated Pulseq acquisitions The framework enables reproducibility of advanced computational MRI methods within a comprehensive end-to-end open-source framework Model-based reconstructions of a FLASH sequence utilizing 15 echoes show good agreement in the parameter maps for both R⋆2 and B0 maps between Pulseq and online measurements Good agreement between Pulseq and online acquisitions can be observed for both quantitative maps demonstrated by low differences across the whole brain

Reproducibility and Maintenance

The integration into the development workflow of the BART toolbox ensures long-term reproducibility and maintenance for sequences A continuous integration pipeline is used to automatically run tests when the software is changed to be able to identify and fix bugs early on The code to reproduce the results of this paper is available at https://gitlab.tugraz.at/ibi/mrirecon/papers/bart-sequence BART is available at https://codeberg. p. 0

Where the paper says it

BART Online Open-Source Sequence Toolbox for Computational MRI Daniel Mackner1 , Philip Schaten1 , Markus Huemer1,2,3, Viktoria Buchegger1 , Moritz Blumenthal1,3, Xiaoqing Wang3 , and Martin Uecker1,4,5 1 Institute of Biomedical Imaging, Graz University of Technology, Graz, Austria 2Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Charlestown, USA 3Department of Radiology, Harvard Medical School, Boston, Massachusetts, USA 4German Centre for Cardiovascular Research (DZHK), Partner Site G¨ottingen, G¨ottingen, Germany 5BioTechMed-Graz, Graz, Austria September 30, 2026 Abstract Purpose: In advanced computational MRI techniques, acquisition and reconstruction techniques are jointly designed.

org/mrirecon/bart The Pulseq interpreter can be obtained from the University of Freiburg The BART Driver Sequence (BOOST: BART Online Open-Source Sequence Toolbox) can be obtained from the Graz University of Technology via Siemens Healthineers’ C2P platform

Conclusion

An end-to-end open-source framework for sequence design and model-based reconstruction based BART can be used to develop advanced qMRI techniques Online integration with a vendor-specific interpreter enables online adjustment of sequence parameters, while Pulseq export allows an exact offline reproduction of the sequence

Conflict of Interest

The authors declare no competing interests

Data Availability Statement

In the spirit of reproducible research, the code to reproduce the results of this paper is available at https://gitlab.tugraz.at/ibi/mrirecon/bart-sequence BART is continuously being developed and maintained since 2013 The Institute of Biomedical Imaging at Graz University of Technology and BART developers are committed to supporting the toolbox in the future

Acknowledgement

This work has been supported by the DZHK (German Centre for Cardiovascular Research), funding code: 81Z0300115 This research was funded in 24 part by NIH under grant U24EB029240 This research was funded in whole or in part by the Austrian Science Fund (FWF) 10.55776/F100800 The authors thank the ISMRM Reproducible Research Study Group for conducting a code review of the code (Version 0. p. 2

Where the paper says it

For this reason, reproducibility became an important topic within the MRI community [1–4]. Pulse sequences are typically implemented using a vendor-specific proprietary platform, hampering translation across systems and sharing of source code within the research community.

p. 2
Where the paper says it

For this reason, reproducibility became an important topic within the MRI community [1–4]. Pulse sequences are typically implemented using a vendor-specific proprietary platform, hampering translation across systems and sharing of source code within the research community.

1) supplied in the Data Availability Statement This research was funded in whole or in part by the Austrian Science Fund (FWF) 10.55776/F100800 The authors thank the ISMRM Reproducible Research Study Group for conducting a code review of the code (Version 0.1) supplied in the Data Availability Statement The scope of the code review covered only the code’s ease of download, quality of documentation, and ability to run, but did not consider scientific accuracy or code efficiency The Institute of Biomedical Imaging at Graz University of Technology and BART developers are committed to supporting the toolbox in the future

References

[1] Stikov N, Trzasko JD, Bernstein MA. Reproducibility and the future of MRI research. p. 0

Where the paper says it

Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Charlestown, USA 3Department of Radiology, Harvard Medical School, Boston, Massachusetts, USA 4German Centre for Cardiovascular Research (DZHK), Partner Site G¨ottingen, G¨ottingen, Germany 5BioTechMed-Graz, Graz, Austria September 30, 2026 Abstract Purpose: In advanced computational MRI techniques, acquisition and reconstruction techniques are jointly designed. For reproducibility, it is therefore important to provide an open implementation of both.

Magnetic Resonance in Medicine 2019; 82:1981–1983

[2] Knoll F, Zbontar J, Sriram A, et al. fastMRI: A Publicly Available Raw k-Space and DICOM Dataset of Knee Images for Accelerated MR Image Reconstruction Using Machine Learning. p. 2

Where the paper says it

BART is an open-source toolbox for computational MRI. It includes algorithms for iterative optimization, efficient calibration and image reconstruction for parallel imaging and compressed sensing, advanced model-based reconstruction for quantitative MRI, as well as for machine learning.

Radiology: Artificial Intelligence 2020; 2:e190007

[3] Veldmann M, Ehses P, Chow K, Nielsen JF, Zaitsev M, St¨ocker T. Open-source MR imaging and reconstruction workflow. p. 0

Where the paper says it

BART Online Open-Source Sequence Toolbox for Computational MRI Daniel Mackner1 , Philip Schaten1 , Markus Huemer1,2,3, Viktoria Buchegger1 , Moritz Blumenthal1,3, Xiaoqing Wang3 , and Martin Uecker1,4,5 1 Institute of Biomedical Imaging, Graz University of Technology, Graz, Austria 2Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Charlestown, USA 3Department of Radiology, Harvard Medical School, Boston, Massachusetts, USA 4German Centre for Cardiovascular Research (DZHK), Partner Site G¨ottingen, G¨ottingen, Germany 5BioTechMed-Graz, Graz, Austria September 30, 2026 Abstract Purpose: In advanced computational MRI techniques, acquisition and reconstruction techniques are jointly designed.

Magn. Reson. Med. 2022; 88:2395–2407

[4] Tamir JI, Blumenthal M, Wang J, Oved T, Shimron E, Zaiss M. MRI acquisition and reconstruction cookbook: recipes for reproducibility, served with real-world flavour. p. 0

Where the paper says it

Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Charlestown, USA 3Department of Radiology, Harvard Medical School, Boston, Massachusetts, USA 4German Centre for Cardiovascular Research (DZHK), Partner Site G¨ottingen, G¨ottingen, Germany 5BioTechMed-Graz, Graz, Austria September 30, 2026 Abstract Purpose: In advanced computational MRI techniques, acquisition and reconstruction techniques are jointly designed. For reproducibility, it is therefore important to provide an open implementation of both.

MAGMA Magnetic Resonance Materials in Physics, Biology and Medicine 2025; 38:367–385

[5] Jochimsen TH, von Mengershausen M. ODIN—Object-oriented Development Interface for NMR. p. 5

Where the paper says it

A flow chart of the interaction between the library and driver is shown in Figure 2. The interface is designed in a generalized way to completely disentangle the actual sequence logic from the driver sequence to ensure the flexibility for further development of new sequence types without modifying the driver sequence.

Journal of Magnetic Resonance 2004; 170:67–78

[6] Magland JF, Li C, Langham MC, Wehrli FW. Pulse sequence programming in a dynamic visual environment: SequenceTree. p. 2

Where the paper says it

gammaSTAR [21] is a vendor-agnostic framework for pulse sequences that allows online adjustment of sequence parameters, but it is not open source. In this work, we describe an open-source sequence programming framework, which is developed as an extension of the Berkeley Advanced Reconstruction Toolbox (BART) [22].

Magnetic Resonance in Medicine 2016; 75:257–265

[7] Nielsen JF, Noll DC. TOPPE: A framework for rapid prototyping of MR pulse sequences. p. 1

Where the paper says it

Methods: A software framework to develop pulse sequences is added to BART , an open-source toolbox for computational MRI. In addition, a vendor-specific driver sequence is developed that can be used to run the sequence on a clinical MRI scanner enabling online adjustment of all relevant sequence parameters.

Magnetic Resonance in Medicine 2018; 79:3128–3134

[8] Layton KJ, Kroboth S, Jia F, et al. Pulseq: A rapid and hardwareindependent pulse sequence prototyping framework. p. 2

Where the paper says it

While highly flexible in the sequence design and fully reproducible, the Pulseq format represents a pre-computed MRI sequence with limited possibilities to adapt the scan to each patient, which hampers its use in clinical studies. gammaSTAR [21] is a vendor-agnostic framework for pulse sequences that allows online adjustment of sequence parameters, but it is not open source.

Magn. Reson. Med. 2017; 77:1544–1552

[9] Ravi KS, Potdar S, Poojar P, et al. Pulseq-Graphical Programming Interface: Open source visual environment for prototyping pulse sequences and integrated magnetic resonance imaging algorithm development. p. 2

Where the paper says it

seq-files that can be executed on MRI scanners of different vendors using vendor-specific driver sequences and can be used as input for Bloch simulations [16–18]. Environments for open-source pulse sequence development now usually offer the possibility to convert to Pulseq [19, 20].

Magnetic Resonance Imaging 2018; 52:9–15

[10] Ravi KS, Geethanath S, Vaughan JT. PyPulseq: A Python Package for MRI Pulse Sequence Design. p. 2

Where the paper says it

Environments for open-source pulse sequence development now usually offer the possibility to convert to Pulseq [19, 20]. While highly flexible in the sequence design and fully reproducible, the Pulseq format represents a pre-computed MRI sequence with limited possibilities to adapt the scan to each patient, which hampers its use in clinical studies.

Journal of Open Source Software 2019; 4:1725

[11] Schuenke P, Kranich C, Lutz M, et al. Open-source cardiac magnetic resonance fingerprinting. p. 2

Where the paper says it

While many frameworks did not find traction in the community so far, Pulseq [8] together with pypulseq [9, 10], became a vendor-neutral open-source standard for sharing MRI sequences. Numerous applications have been published using the format, including MR fingerprinting [11, 12] cardiac T1 mapping [13, 14], and Chemical Exchange Saturation Transfer (CEST) [15].

MAGMA Magnetic Resonance Materials in Physics, Biology and Medicine 2025; 38:665–677

[12] Griesler T, Stebani J, Kaplan S, et al. OpenMRF: A Modular, VendorNeutral Open-Source Framework for Magnetic Resonance Fingerprinting Using Pulseq. p. 2

Where the paper says it

While many frameworks did not find traction in the community so far, Pulseq [8] together with pypulseq [9, 10], became a vendor-neutral open-source standard for sharing MRI sequences. Numerous applications have been published using the format, including MR fingerprinting [11, 12] cardiac T1 mapping [13, 14], and Chemical Exchange Saturation Transfer (CEST) [15].

Magnetic Resonance in Medicine 2026; n/a

[13] Gaspar AS, Silva NA, Ferreira AM, Nunes RG. Repeatability of OpenMOLLI: An open-source inversion recovery myocardial T1 mapping sequence for fast prototyping. p. 2

Where the paper says it

While many frameworks did not find traction in the community so far, Pulseq [8] together with pypulseq [9, 10], became a vendor-neutral open-source standard for sharing MRI sequences. Numerous applications have been published using the format, including MR fingerprinting [11, 12] cardiac T1 mapping [13, 14], and Chemical Exchange Saturation Transfer (CEST) [15].

Magnetic Resonance in Medicine 2024; 92:741–750

[14] Gaspar AS, Silva NA, Price AN, Ferreira AM, Nunes RG. Open-source myocardial T1 mapping with simultaneous multi-slice acceleration: Combining an auto-calibrated blipped-bSSFP readout with VERSEMB pulses. p. 6

Where the paper says it

Gradients are prepared as trapezoids and a failed preparation returns an error, feasible gradient amplitudes are projected onto corresponding axes. Preparation of RF pulses include mapping to a shape and complex modulation depending on the spoiling phase and the slice shift.

Magnetic Resonance in Medicine 2023; 90:539–551

[15] Herz K, Mueller S, Perlman O, et al. Pulseq-CEST: Towards multisite multi-vendor compatibility and reproducibility of CEST experiments using an open-source sequence standard. p. 2

Where the paper says it

While many frameworks did not find traction in the community so far, Pulseq [8] together with pypulseq [9, 10], became a vendor-neutral open-source standard for sharing MRI sequences. Numerous applications have been published using the format, including MR fingerprinting [11, 12] cardiac T1 mapping [13, 14], and Chemical Exchange Saturation Transfer (CEST) [15].

Magnetic Resonance in Medicine 2021; 86:1845–1858

[16] Loktyushin A, Herz K, Dang N, et al. MRzero - Automated discovery of MRI sequences using supervised learning. p. 2

Where the paper says it

BART is an open-source toolbox for computational MRI. It includes algorithms for iterative optimization, efficient calibration and image reconstruction for parallel imaging and compressed sensing, advanced model-based reconstruction for quantitative MRI, as well as for machine learning.

Magnetic Resonance in Medicine 2021; 86:709–724

[17] Castillo-Passi C, Coronado R, Varela-Mattatall G, Alberola-L´opez C, Botnar R, Irarrazaval P. KomaMRI.jl: An open-source framework for general MRI simulations with GPU acceleration. p. 2

Where the paper says it

seq-files that can be executed on MRI scanners of different vendors using vendor-specific driver sequences and can be used as input for Bloch simulations [16–18]. Environments for open-source pulse sequence development now usually offer the possibility to convert to Pulseq [19, 20].

Magnetic Resonance in Medicine 2023; 90:329–342

[18] Weinm¨uller S, Endres J, Dang N, Stollberger R, Zaiss M. MR-zero meets FLASH – controlling the transient signal decay in gradient- and RF-spoiled gradient echo sequences. p. 9

Where the paper says it

When running the sequence, events are prepared and interpreted block-wise. (B) Implementation of a gradient-echo imaging block including start times computation, spoiling phase and radial projection angle calculation.

Magnetic Resonance in Medicine 2025; 93:942–960

[19] St¨ocker T, Vahedipour K, Pflugfelder D, Shah NJ. High-performance computing MRI simulations. p. 2

Where the paper says it

Pulseq files are human-readable . seq-files that can be executed on MRI scanners of different vendors using vendor-specific driver sequences and can be used as input for Bloch simulations [16–18].

Magnetic Resonance in Medicine 2010; 64:186–193

[20] Artiges A, Saimbhi AS, Castillo-Passi C, Lattanzi R, Block KT. mtrk—A flexible environment for developing open-source MRI pulse sequences. p. 2

Where the paper says it

While highly flexible in the sequence design and fully reproducible, the Pulseq format represents a pre-computed MRI sequence with limited possibilities to adapt the scan to each patient, which hampers its use in clinical studies. gammaSTAR [21] is a vendor-agnostic framework for pulse sequences that allows online adjustment of sequence parameters, but it is not open source.

Magnetic Resonance in Medicine 2026; 95:1089–1097

[21] Konstandin S, G¨unther M, Hoinkiss DC. gammaSTAR: A framework for the development of dynamic, real-time capable MR sequences. p. 2

Where the paper says it

While highly flexible in the sequence design and fully reproducible, the Pulseq format represents a pre-computed MRI sequence with limited possibilities to adapt the scan to each patient, which hampers its use in clinical studies. gammaSTAR [21] is a vendor-agnostic framework for pulse sequences that allows online adjustment of sequence parameters, but it is not open source.

Magnetic Resonance in Medicine 2025; 94:1485–1499

[22] Uecker M, Virtue P, Ong F, et al. Software toolbox and programming library for compressed sensing and parallel imaging. p. 2

Where the paper says it

BART is an open-source toolbox for computational MRI. It includes algorithms for iterative optimization, efficient calibration and image reconstruction for parallel imaging and compressed sensing, advanced model-based reconstruction for quantitative MRI, as well as for machine learning.

In: ISMRM Workshop on Data Sampling and Image Reconstruction. p. 2
Where the paper says it

BART is an open-source toolbox for computational MRI. It includes algorithms for iterative optimization, efficient calibration and image reconstruction for parallel imaging and compressed sensing, advanced model-based reconstruction for quantitative MRI, as well as for machine learning.

Sedona; 2013

[23] Fessler JA, Sutton BP. Nonuniform fast Fourier transforms using minmax interpolation. p. 4

Where the paper says it

Gradients are decomposed into triangles, each described by its amplitude at midtime in phase-, read- and slice-encoding direction. Hence, typical trapezoidal gradients have a compact representation using two overlapping asymmetrical triangles, while arbitrary gradient shapes can be designed by superposition of multiple gradient events corresponding to a linear spline interpolation.

IEEE Trans. Signal Processing 2003; 51:560–574

[24] Graff C, Li Z, Bilgin A, Altbach MI, Gmitro AF, Clarkson EW. Iterative T2 estimation from highly undersampled radial fast spin-echo data. p. 9

Where the paper says it

When running the sequence, events are prepared and interpreted block-wise. (B) Implementation of a gradient-echo imaging block including start times computation, spoiling phase and radial projection angle calculation.

Proc. Int. Soc. Mag. Reson. Med. 2006; 14:925

[25] Olafsson VT, Noll DC, Fessler JA. Fast Joint Reconstruction of Dynamic R∗2 and Field Maps in Functional MRI. p. 1

Where the paper says it

Using the Pulseq format, the exact same sequence can also be reproduced offline using a widely used vendor-neutral opensource standard. As proof-of-concept, quantitative MRI methods for T1 and joint water/fat R∗ 2 , B0 mapping using radial FLASH and model-based reconstruction are implemented in the proposed framework.

IEEE Trans. Med. Imag. 2008; 27:1177–1188

[26] Block KT, Uecker M, Frahm J. Model-Based Iterative Reconstruction for Radial Fast Spin-Echo MRI. p. 2

Where the paper says it

BART is an open-source toolbox for computational MRI. It includes algorithms for iterative optimization, efficient calibration and image reconstruction for parallel imaging and compressed sensing, advanced model-based reconstruction for quantitative MRI, as well as for machine learning.

IEEE Trans. Med. Imaging 2009; 28:1759–1769

[27] Sumpf TJ, Uecker M, Boretius S, Frahm J. Model-based nonlinear inverse reconstruction for T2-mapping using highly undersampled spinecho MRI. p. 1

Where the paper says it

Using the Pulseq format, the exact same sequence can also be reproduced offline using a widely used vendor-neutral opensource standard. As proof-of-concept, quantitative MRI methods for T1 and joint water/fat R∗ 2 , B0 mapping using radial FLASH and model-based reconstruction are implemented in the proposed framework.

J. Magn. Reson. Imaging 2011; 34:420–428

[28] Wang X, Tan Z, Scholand N, Roeloffs V, Uecker M. Physics-based Reconstruction Methods for Magnetic Resonance Imaging. p. 1

Where the paper says it

Using the Pulseq format, the exact same sequence can also be reproduced offline using a widely used vendor-neutral opensource standard. As proof-of-concept, quantitative MRI methods for T1 and joint water/fat R∗ 2 , B0 mapping using radial FLASH and model-based reconstruction are implemented in the proposed framework.

Philos. Trans. R. Soc. A. 2021; 379:20200196

[29] Maier O, Spann SM, B¨odenler M, Stollberger R. PyQMRI: An accelerated Python based Quantitative MRI toolbox. p. 2

Where the paper says it

BART is an open-source toolbox for computational MRI. It includes algorithms for iterative optimization, efficient calibration and image reconstruction for parallel imaging and compressed sensing, advanced model-based reconstruction for quantitative MRI, as well as for machine learning.

Journal of Open Source Software 2020; 5:2727

[30] Zimmermann FF, Schuenke P, Brahma S, et al. MRpro - PyTorchbased MR image reconstruction and processing package. p. 2

Where the paper says it

BART is an open-source toolbox for computational MRI. It includes algorithms for iterative optimization, efficient calibration and image reconstruction for parallel imaging and compressed sensing, advanced model-based reconstruction for quantitative MRI, as well as for machine learning.

Version v0. p. 5
Where the paper says it

Hence, sequence libraries implementing new sequences can be linked to the driver sequence without any need to update this vendorspecific component. Compatibility of driver sequence and library is ensured by a mutual version check at run-time (bart seq version check) which is performed during an initial handshake.

260204

[31] Scholand N, Schaten P, Graf C, et al. Rational approximation of golden angles: Accelerated reconstructions for radial MRI. p. 1

Where the paper says it

Using the Pulseq format, the exact same sequence can also be reproduced offline using a widely used vendor-neutral opensource standard. As proof-of-concept, quantitative MRI methods for T1 and joint water/fat R∗ 2 , B0 mapping using radial FLASH and model-based reconstruction are implemented in the proposed framework.

Magn. Reson. Med. 2025; 93:51–66

[32] Magland J, Wehrli F. General algorithm for automated off-center MRI. p. 0

Where the paper says it

BART Online Open-Source Sequence Toolbox for Computational MRI Daniel Mackner1 , Philip Schaten1 , Markus Huemer1,2,3, Viktoria Buchegger1 , Moritz Blumenthal1,3, Xiaoqing Wang3 , and Martin Uecker1,4,5 1 Institute of Biomedical Imaging, Graz University of Technology, Graz, Austria 2Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Charlestown, USA 3Department of Radiology, Harvard Medical School, Boston, Massachusetts, USA 4German Centre for Cardiovascular Research (DZHK), Partner Site G¨ottingen, G¨ottingen, Germany 5BioTechMed-Graz, Graz, Austria September 30, 2026 Abstract Purpose: In advanced computational MRI techniques, acquisition and reconstruction techniques are jointly designed.

Magnetic Resonance in Medicine 2006; 56:230–233

[33] Shafiekhani M, Hucker P, Zaitsev M. Vendor-Neutral FOV Positioning for Non-Cartesian MRI in the Pulseq Framework. p. 1

Where the paper says it

Using the Pulseq format, the exact same sequence can also be reproduced offline using a widely used vendor-neutral opensource standard. As proof-of-concept, quantitative MRI methods for T1 and joint water/fat R∗ 2 , B0 mapping using radial FLASH and model-based reconstruction are implemented in the proposed framework.

In: Book of Abstracts ESMRMB 2025 Online 41st Annual Scientific Meeting. p. 0
Where the paper says it

Correspondence to: Martin Uecker, Graz University of Technology, Institute of Biomedical Imaging, Stremayrgasse 16/3, 8010 Graz, Austria, Email: uecker@tugraz. at Part of this work has been presented at the ISMRM Annual Conference 2026 (Cape Town).

Marseille: Springer; 2025:105–106

[34] Uecker M, Hohage T, Block KT, Frahm J. Image reconstruction by regularized nonlinear inversion-joint estimation of coil sensitivities and image content. p. 2

Where the paper says it

BART is an open-source toolbox for computational MRI. It includes algorithms for iterative optimization, efficient calibration and image reconstruction for parallel imaging and compressed sensing, advanced model-based reconstruction for quantitative MRI, as well as for machine learning.

Magn. Reson. Med. 2008; 60:674–682

[35] Bernstein MA, King KF, Zhou XJ. Handbook of MRI Pulse Sequences. p. 1

Where the paper says it

Methods: A software framework to develop pulse sequences is added to BART , an open-source toolbox for computational MRI. In addition, a vendor-specific driver sequence is developed that can be used to run the sequence on a clinical MRI scanner enabling online adjustment of all relevant sequence parameters.

Elsevier Science; 2004

[36] Frahm J, Haase A, Matthaei D. Rapid NMR imaging of dynamic processes using the FLASH technique. p. 1

Where the paper says it

Using the Pulseq format, the exact same sequence can also be reproduced offline using a widely used vendor-neutral opensource standard. As proof-of-concept, quantitative MRI methods for T1 and joint water/fat R∗ 2 , B0 mapping using radial FLASH and model-based reconstruction are implemented in the proposed framework.

Magn. Reson. Med. 1986; 3:321–327

[37] Roeloffs V, Voit D, Frahm J. Spoiling without additional gradients: Radial FLASH MRI with randomized radiofrequency phases. p. 1

Where the paper says it

Using the Pulseq format, the exact same sequence can also be reproduced offline using a widely used vendor-neutral opensource standard. As proof-of-concept, quantitative MRI methods for T1 and joint water/fat R∗ 2 , B0 mapping using radial FLASH and model-based reconstruction are implemented in the proposed framework.

Magn. Reson. Med. 2016; 75:2094–2099

[38] Wang X, Roeloffs V, Klosowski J, et al. Model-based T1 mapping with sparsity constraints using single-shot inversion-recovery radial FLASH. p. 1

Where the paper says it

Using the Pulseq format, the exact same sequence can also be reproduced offline using a widely used vendor-neutral opensource standard. As proof-of-concept, quantitative MRI methods for T1 and joint water/fat R∗ 2 , B0 mapping using radial FLASH and model-based reconstruction are implemented in the proposed framework.

Magn. Reson. Med. 2018; 79:730–740

[39] Wang X, Kohler F, Unterberg-Buchwald C, Lotz J, Frahm J, Uecker M. Model-based myocardial T1 mapping with sparsity constraints using single-shot inversion-recovery radial FLASH cardiovascular magnetic resonance. p. 1

Where the paper says it

Using the Pulseq format, the exact same sequence can also be reproduced offline using a widely used vendor-neutral opensource standard. As proof-of-concept, quantitative MRI methods for T1 and joint water/fat R∗ 2 , B0 mapping using radial FLASH and model-based reconstruction are implemented in the proposed framework.

J. Cardiovasc. Magn. Reson. 2019; 21:60

[40] Petzschner FH, Ponce IP, Blaimer M, Jakob PM, Breuer FA. Fast MR parameter mapping using k-t principal component analysis. p. 6

Where the paper says it

2 ARCHITECTURE AND IMPLEMENTATION Vendor-specific user interface (UI) parameters are mapped to sequence parameters using a generalized interface that also allows the definition of custom parameters. For every parameter change, feasibility of the configuration set is checked.

Magn. Reson. Med.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Precision in quantitative MRI acquisition and reconstruction is enhanced by an end-to-end open-source framework that jointly designs acquisition and reconstruction techniques for advanced computational MRI methods. This allows for long-term reproducibility of methods by ensuring Consistency between online and offline acquisition is validated in phantom and in vivo experiments.

  2. Online sequence adjustment capability can be integrated into the AI system through a vendor-specific driver sequence linked to BART, enabling online adjustment of all relevant sequence parameters during clinical MRI scans. This feature allows the system to adapt acquisition parameters dynamically based on real-time scanner capabilities or patient needs.

  3. Offline reproducibility of advanced methods is guaranteed by exporting sequences in Pulseq format, allowing them to be reproduced exactly using a widely used vendor-neutral opensource standard. This ensures that all necessary information for model-based reconstruction is faithfully preserved when regenerating the acquisition offline.

  4. The AI system can perform complex quantitative MRI tasks such as T1 mapping and joint water/fat R2 maps, validated through methods like linearized subspace reconstruction for T1 mapping with Inversion-Recovery FLASH. This enables the system to generate quantitative parameter maps that agree between online and offline regenerated Pulseq acquisitions.

  5. The framework supports non-Cartesian MRI by allowing in-plane FOV shift during acquisition, which can be corrected retrospectively using scripts provided in the toolbox. This capability allows the system to handle complex spatial variations in image acquisition, ensuring that phase inconsistencies due to the FOV shift were corrected retrospectively.

  6. Model-based reconstruction algorithms are improved by having access to full sequence details, as k-space trajectories, gradient and pulse waveforms, and timing information can be recovered directly from the sequence. This direct access ensures that reconstructions are based on faithfully preserved acquisition data.

Abstract

Purpose In advanced computational MRI techniques, acquisition and reconstruction techniques are jointly designed. For reproducibility, it is therefore important to provide an open implementation of both. At the same time, any use in a clinical environment usually requires a close integration with the MRI scanner. Ensuring long-time reproducibility and maintenance then poses additional challenges. In this work, we aim to provide a fully integrated open-source framework that can meet these demands. Methods A software framework to develop pulse sequences is added to BART, an open-source toolbox for computational MRI. In addition, a vendor-specific driver sequence is developed that can be used to run the sequence on a clinical MRI scanner enabling online adjustment of all relevant sequence parameters. Using the Pulseq format, the exact same sequence can also be reproduced offline using a widely used vendor-neutral open-source standard. Using the Pulseq format, the exact same sequence can also be reproduced offline. As proof-of-concept, quantitative MRI methods for T1 and joint water/fat R2*, B0 mapping using radial FLASH and model-based reconstruction are implemented in the proposed framework. Consistency between online and offline acquisition is validated in phantom and in vivo experiments. Results Quantitative MRI methods highlighting specific challenges of acquisition and reconstruction were successfully implemented in BART. Acquisition parameters and FOV can be adapted online on a clinical MRI system. Quantitative parameter maps from model-based reconstruction agree for online and offline regenerated Pulseq acquisitions. Conclusion This work enables reproducibility of advanced computational MRI methods within a comprehensive end-to-end open-source framework.

Related papers