Introduction

Magnetic resonance fingerprinting (MRF) offers a way to acquire relaxation time maps in an efficient manner and has shown its use in many 1.5T and 3T applications¹. The T₁ and T₂ relaxation times can potentially be used to generate a variety of image contrasts retrospectively². This could be particularly relevant at low field³, where scans take many minutes due to a lower signal-to-noise ratio (SNR). Portable point-of-care permanent magnet systems often have large B₀ inhomogeneities, which can cause image distortions and signal dropout. MRF is a flexible acquisition framework⁴ that can encode the B₀ field inhomogeneities in the acquisition process, to incorporate them in the reconstruction process. In this work, we use an alternating TE pattern in the MRF sequence that supports ΔB₀ estimation from the MRF data. We use the reconstructed ΔB₀ map to correct for B₀-induced image distortions of the MRF images, before matching them to the calculated dictionary. This method reduces distortions in the relaxation time maps without needing to acquire an additional ΔB₀ map.

Methods

Hardware: We used a 50 mT Halbach magnet consisting of 23 rings filled with neodymium boron iron magnets, shown in Fig. 1A, with custom-built RF and gradient amplifiers and a Magritek Kea2 spectrometer (Fig. 1B). A 15 cm diameter, 15 cm long solenoid coil was used in transceive mode (Fig. 1C). A mechanical robot with a magnetic field probe was used to map the B₀ field during system installation.

Fingerprinting implementation: MRF data were acquired with a FISP sequence⁵, using an optimized flip angle pattern of 240 RF shots, a constant TR of 12 ms and an alternating TE pattern to support ΔB₀ estimation, shown in Fig. 2A. The dictionary was generated using the extended phase graph formalism, with T₁ values ranging from 20-500 ms (5 ms steps), T₂ values ranging from 10-500 ms (5 ms steps) and B₁⁺ fraction ranging from 0.05-1.50 (0.05 steps).

MR Data acquisition: Experiments were performed in the lower leg of 2 healthy volunteers with informed consent obtained. Three-dimensional MRF scans were acquired with a Cartesian sampling scheme shown in Fig. 2B. Scan parameters: FOV = 170×150×99 mm³, resolution = 2.5×2.5×3.0 mm³, TE₁/TE₂/TR = 6.00/6.15/12.0 ms, imaging bandwidth (BW) = 294 Hz/pixel, undersampling factor = 8.6, scan time = 13 min and 6 s. Inversion recovery (inversion times = 25/50/75/100/150/300 ms) and multi-echo-spin-echo (TEs = 20/40/60/80/100/120/140/160/180/200 ms) were used for reference T₁ and T₂ mapping, respectively. A TSE-based ΔB₀ map was acquired for comparison⁶.

Reconstruction: MRF images were reconstructed using a matrix completion-based reconstruction on the two separate TE series⁷. A ΔB₀ map was estimated from the two series using total variation regularization and spherical harmonic decomposition and used in a model-based reconstruction framework to correct for image distortions⁶. The resulting combined image series were matched to the dictionary using the simulated B₁⁺ excitation profile (Fig. 1E) as prior information for each slice. This pipeline is schematically shown in Fig. 2C.

Results

Figure 3 shows that the relaxation times obtained with MRF (constant TE, without distortion correction) are close to those obtained with the reference techniques, but the SNR of the MRF maps is lower compared to that of the reference maps, as expected. Using an alternating TE pattern results in very similar parameter maps compared to using a constant TE (both without distortion correction), as shown in Fig. 4. Estimating a ΔB₀ map from the alternating TE pattern and using it in a model-based MRF reconstruction results in matched parameter maps with reduced distortions. The estimated ΔB₀ map is in the same range as that estimated from a TSE sequence, with a maximum error of 105 Hz.

Discussion

The acquired MRF ΔB₀ map corresponds with the map measured with a robot and that measured with a fully encoded TSE sequence. The difference between the MRF-based ΔB₀ map and the TSE-based ΔB₀ map could have been introduced by hardware imperfections (spectrometer timing jitter, eddy currents, etc.) leading to phase errors in the MRF sequence or by heating of the magnet during the imaging session. Using the ΔB₀ map in a model-based reconstruction framework reduces the image distortions along the readout direction of the parameter maps. The remaining signal void at the left side of the M₀ map is a result of through-plane B₀-induced dephasing, for which the current reconstruction algorithm does not correct. Future work could resolve this by incorporating a ΔB₀ map with a higher through-plane resolution into the signal model.

Conclusion

It is possible to integrate ΔB₀ estimation and correction in an MRF acquisition and reconstruction framework for a 50 mT permanent magnet MR system. The proposed framework reduces the image distortions of the MRF relaxation time maps without increasing the scan time.

Acknowledgements

This work is supported by the following grants: Horizon 2020 ERC FET-OPEN 737180 Histo MRI, Horizon 2020 ERC Advanced NOMA-MRI 670629, Simon Stevin Meester Award and NWO WOTRO Joint SDG Research Programme W 07.303.101. We would also like to thank Hanna Liu for sharing the optimized flip angle train with us.