OpenMRF: A Pulseq-based Magnetic Resonance Fingerprinting Framework for simultaneous T1, T2 and T1ρ Quantification

Presenting Author(s)

• 

  Maximilian Gram, PhD

  Postdoc
  University Hospital Wuerzburg
  Würzburg, Bayern, Germany

Primary Author(s)

• 

  Maximilian Gram, PhD

  Postdoc
  University Hospital Wuerzburg
  Würzburg, Bayern, Germany

Co-Author(s)

• TG

  Tom Griesler, MSc

  PhD Candidate
  University of Michigan
  Ann Arbor, Michigan, United States
• JS

  Jannik Stebani, MSc

  PhD Candidate
  University of Würzburg, Germany
• IA

  Ivaylo Angelov, MSc

  PhD Student
  University of Würzburg
  Würzburg, Bayern, Germany
• PA

  Petra Albertova, MSc

  PhD Student
  University Hospital Würzburg
  Würzburg, Bayern, Germany
• 

  Sydney Kaplan, MSc

  Student
  University of Michigan
  Ann Arbor, Michigan, United States
• CS

  Calder D. Sheagren, PhD

  Postdoctoral Fellow
  University of Michigan
  Ann Arbor, Michigan, United States
• MB

  Martin Blaimer, PhD

  Postdoc
  Fraunhofer Institute for Integrated Circuits IIS
  Würzburg, Bayern, Germany
• PJ

  Peter Michael Jakob, PhD

  Professor
  University of Würzburg
  Würzburg, Bayern, Germany
• NS

  Nicole Seiberlich

  Professor
  University of Michigan
  Ann Arbor, Michigan, United States
• 

  Peter Nordbeck, MD

  Senior physician
  University Hospital Würzburg
  Würzburg, Bayern, Germany

Description: Background

Cardiac MR Fingerprinting (cMRF) enables simultaneous quantification of multiple tissue parameters in a time-efficient manner and holds considerable potential for clinical translation [1]. However, established cMRF techniques are vendor-specific and lack multi-platform transferability [2,3]. The open-source Pulseq framework provides a flexible, vendor-neutral environment for sequence development [4]. Yet, no integrated solution currently exists for the implementation and reconstruction of advanced cMRF methods particularly those incorporating T_1ρ encoding [5,6]. To address this gap, we introduce OpenMRF: a modular framework that combines flexible sequence implementation, automated dictionary simulation, and state-of-the-art low-rank reconstruction. OpenMRF supports a comprehensive library of contrast preparation modules - including T_1ρ - and enables platform-independent cMRF for advanced tissue characterization.



OpenMRF

OpenMRF is available as part of the Harmonized MRI project [7]. Sequences are defined using the Pulseq format, enabling modular integration of contrast preparation modules such as adiabatic inversions, T₂-preparations, and B₀/B₁-insensitive spin-locking for T_1ρ encoding [8]. Dictionary generation is completely automated by converting .seq files into simulation instructions for Bloch-based modeling, accounting for slice profile effects, relaxation during adiabatic or spin-lock pulses, and subject-specific ECG timings. Rawdata are processed using a low-rank subspace reconstruction following compression of the dictionary along the temporal dimension [1].



Examples

The source code for T₁-T₂-T_1ρ cMRF sequences is available at [9]. To validate OpenMRF for cardiac applications, we performed measurements on the ISMRM/NIST phantom [10] using Siemens MRI systems (0.55T, 1.5T, and 3T). In vivo validation was conducted in a healthy volunteer at 3T. Figure 1 illustrates T₁, T₂, and T_1ρ maps of the NIST phantom acquired with OpenMRF sequences. Quantitative comparison with gold-standard mapping techniques demonstrated high agreement (averaged R²=0.996) across all three field strengths (Figure. 2). T₁ and T₂ deviations remained within acceptable limits (±2.7% and ±6.1%) for physiologically relevant relaxation time ranges, while only T_1ρ revealed a significant bias at 3T (-13%). Myocardial relaxation times were consistent with previously reported values (Figure 3).



Conclusion

OpenMRF provides a modular framework for advanced cMRF, enabling robust and reproducible quantification of T₁, T₂, and T_1ρ across different platforms and field strengths. Our open-source project is an end-to-end solution that covers the entire MRF workflow with just a single Matlab-based repository (Table 1). By introducing standardized sequence implementation, dictionary simulation and reconstruction, OpenMRF has the potential to pave the way for broader application of multiparametric mapping in routine clinical practice.

Access Instructions: https://github.com/HarmonizedMRI/OpenMRF

OpenMRF Features

Category	Feature	Status
Basic imaging	Spiral readoutRosette readoutECG trigger	Ready to useWork in progressReady to use
Preparation modules	T1-InversionT1-SaturationT2-MLEVT1ρ: continuous waveT2ρ-MLEVT1ρ: adiabaticDiffusionCESTFat suppression	Ready to useReady to useReady to useReady to useReady to useWork in progressWork in progressWork in progressReady to use
Dictionary simulation	Automatic .seq converterBloch isochromatsExtended Phase GraphsSlice profilesAdiabatic efficienciesContinous wave spin-lockingAdiabatic spin-lockingDiffusionCESTGPU translation	Ready to useReady to useReady to useReady to useReady to useReady to useReady to useReady to useWork in progressWork in progress
Recontruction & matching	Brute Force NUFFT & MatchingSVD-compressed dictionariesLow-Rank recontructionDeep-Image-Prior recontructionGPU translation	Ready to useReady to useReady to useWork in progressWork in progress
Accomplisehd Tests	Siemens Free.Max 0.55 TSiemens Sola 1.5 TSiemens Avanto 1.5 TSiemens Vida 3.0 TSiemens Cima.X 3.0 TSiemens Prisma 3.0 TSiemens Terra.X 7.0TGE Signa 3.0 T	CompletedCompletedWork in progressCompletedCompletedWork in progressWork in progressWork in progress

Figure 1) Results from NIST phantom measurements on a 0.55 T FreeMax, 1.5 T Sola, and 3 T Vida system. A cMRF sequence with 16 segments and integrated T1, T2, and T1ρ preparations was used across all platforms. Identical .seq files were applied on the Sola and Vida scanner, while reduced RF and gradient amplitudes were required on the FreeMax due to hardware limitations. Relaxation time maps were obtained via dictionary matching (~1000000 entries) following low-rank reconstruction.
Figure 2) Deviation of T1, T2, and T1ρ values from gold-standard reference sequences. Across all parameters and field strengths, high agreement was observed. Only very short or very long relaxation times, which are of limited physiological relevance, showed increased deviations.
Figure 3) Results from in vivo scans on a 3 T Siemens Skyra. Panels A–C illustrate relaxation time maps in a four-chamber view (upper row) and a short-axis view (bottom row). The matched parameter maps exhibit good image quality. Panel D compares left ventricular values with those from a widely used cMRF sequence encoding only T1 and T2 [1]. No systematic differences were found between the conventional and extended T1–T2–T1ρ variants.