OpenMRF: A Vendor-Agnostic Open-Source Framework for Standardized Cardiac Magnetic Resonance Fingerprinting

M. Gram (Würzburg)1, P. Albertova (Würzburg)2, T. Griesler (Ann Arbor)3, S. Kaplan (Ann Arbor)4, X. Wang (Freiburg)5, M. Zaitsev (Freiburg)5, T. Wech (Würzburg)6, P. M. Jakob (Würzburg)7, N. Seiberlich (Ann Arbor)4, P. Nordbeck (Würzburg)1
1Universitätsklinikum Würzburg Medizinische Klinik und Poliklinik I Würzburg, Deutschland; 2University Hospital Würzburg Department of Internal Medicine I Würzburg, Deutschland; 3University of Michigan Department of Biomedical Engineering Ann Arbor, USA; 4University of Michigan Department of Radiology Ann Arbor, USA; 5University Medical Center Freiburg, Faculty of Medicine, University of Freiburg Division of Medical Physics, Department of Radiology Freiburg, Deutschland; 6University Hospital Würzburg Department of Diagnostic and Interventional Radiology Würzburg, Deutschland; 7University of Würzburg Experimental Physics 5 Würzburg, Deutschland
BACKGROUND
Quantitative MRI enables the measurement of tissue-specific relaxation times such as T1, T2, and T1ρ, allowing the assessment of myocardial fibrosis or edema without the need for contrast agents [1]. Magnetic Resonance Fingerprinting (MRF) is a promising approach for clinical application, as it allows simultaneous quantification of multiple parameters within reduced scan times [2,3]. However, reproducibility between different vendors, platforms, and field strengths remains limited due to non-standardized acquisition and reconstruction pipelines. To overcome these limitations, we developed OpenMRF - a vendor-agnostic, open-source framework for the design, simulation, and reconstruction of cardiac MRF (cMRF) sequences, enabling standardized and reproducible multi-center studies.

METHODS

OpenMRF builds on the Pulseq format for vendor-independent MRI sequence programming [4] and unifies the entire cMRF workflow, from sequence definition to image reconstruction. The main modules include: (1) flexible readouts supporting spiral, radial, and rosette trajectories (2) standardized T1, T2, and T1ρ preparations (3) automated Bloch simulation for accurate dictionary generation (4) reconstruction using an iterative low-rank algorithm.
Validation was performed using the NIST phantom on multiple Siemens scanners at 0.55T, 1.5T, and 3T. The unified T1–T2–T1ρ cMRF protocol was evaluated in a multi-site study. In-vivo feasibility was further demonstrated in healthy subjects.

RESULTS

Phantom measurements (Fig. 1) showed excellent agreement with gold-standard references, with minor deviations of +3.1% (T1) and -1.2% (T2). T1ρ values obtained from MRF were slightly lower compared to conventional mapping (-7.4%). Parameter maps demonstrated high inter-site reproducibility. In-vivo experiments confirmed robust performance, showing clear delineation of the left-ventricular myocardium and consistent quantitative values within the expected physiological range (Fig. 2).

DISCUSSION & CONCLUSION

OpenMRF enables reproducible and standardized cardiac tissue characterization using identical open-source Pulseq sequences. By integrating standardized preparations, Bloch simulations, and low-rank reconstruction, the framework bridges technical sequence development and clinical translation. The presented results demonstrate that quantitative cardiac MRI can be performed in a harmonized and vendor-independent manner, providing reliable relaxation time maps. The long-term goal is to link functional and quantitative cardiac imaging within a unified cMRF approach, enabling the comprehensive assessment of both structural and functional myocardial changes in a single examination.

The OpenMRF project establishes a harmonized and fully transparent workflow for quantitative cardiac MRI. Its reproducibility across scanners, field strengths, and research sites provides a solid foundation for future multi-center validation of native myocardial tissue characterization.

[1] Christodoulou A et al. doi: 10.1016/j.jocmr.2024.100997 [2] Ma D et al. doi: 10.1038/nature11971 [3] Hamilton JI et al. doi: 10.1002/mrm.26216 [4] Layton KJ et al. doi: 10.1002/mrm.26235