Synopsis

Keywords: Low-Field MRI, Data Acquisition

Point-of-care imaging with low-field MRI (<0.1T) is a potential game changer for low-income countries and the intensive care unit. The main challenge is the low SNR. Balanced steady-state free precession (bSSFP) sequences are fast and SNR-efficient. However, bSSFP is very sensitive to B₀ inhomogeneities resulting in banding artifacts. We evaluated the feasibility of using bSSFP on our 46 mT MRI scanner. By acquiring 17 bSSFP datasets with linearly increasing frequency offsets (from 0 to 1/TR), we could reconstruct banding-free maximum-intensity images as well as F₀ and F_-1 SSFP configurations, which we employed for rapid relaxation time mapping.

Introduction

Development of low-field MRIs (with B₀<0.1T) for point-of-care (POC) applications has recently become increasingly widespread¹. The reduced cost and improved portability drastically increase their accessibility compared to conventional systems. However, low-field brain images suffer from low SNR and poor tissue contrast². The balanced steady-state free precession (bSSFP) imaging sequence is fast, offers high SNR, and produces a T₂/T₁-weighted image contrast³. However, bSSFP is very sensitive to B₀ inhomogeneities (ΔB₀), which lead to banding artifacts³. Banding-free images can be achieved by acquiring multiple images using RF phase-cycling increments or frequency offsets³. Although optimized passive and active shimming schemes have been developed to minimize ΔB₀⁴, compact low-field permanent MRI systems based on cylindrical Halbach arrays have intrinsically large ΔB₀ (~thousands ppm) due to the finite diameter-to-length ratio⁵ compared to larger low-field fixed location systems on which bSSFP has been implemented⁶. Here, we wanted to evaluate the potential of bSSFP on our 46mT imaging system.

Materials and Methods

Hardware: We used a 46mT Halbach array-based magnet (outer/inner diameter=60/30.1cm, length=49.2cm, weight=111kg), with custom-built RF and Bruker gradient amplifiers and a Magritek Kea2 spectrometer⁷. Imaging was performed using a solenoid and an elliptical spiral-solenoid head coil^2,7.
bSSFP implementation: The 3D bSSFP sequence consisted of a train of hard pulses with alternating flip angle (α), following an α/2 ramp preparation (Fig.1A). A zig-zag phase-encode trajectory was used to smoothly sample k-space⁸ (Fig.1B). To correct for banding artifacts we used a range of equidistant frequency offsets between 0 and 1/TR.
Sequence validation (n=3): 5 tubes were filled with water-based agarose gel (ranging from 0 to 4%) and 2mM copper-sulfate to obtain different T₁/T₂ ratios (Fig.2A). T₂ mapping was performed using a 3D Carr-Purcell-Meiboom-Gill (CPMG) imaging sequence (resolution=3x3x10mm³, TR/echo-spacing=1250/20ms, 10 echoes, acq. time~15min). T₁ mapping was performed using 3D inversion recovery (IR) scans with TSE readout (identical resolution, TR/TE/TEeff=1250/14/14ms, ETL=6, inversion times=50/100/150/200/300/400ms, acq. time~15min). A series of 11 bSSFP datasets (identical resolution, TR/TE=16/8ms, acq. time~11x3.4min) were acquired with a flip angle varying from 10º to 170º. Maximum-intensity bSSFP images were reconstructed for each flip angle using 17 frequency offset increments. The signal was averaged over 8 slices and a region-of-interest (5x5 pixels) was defined to extract proton-density (PD), T₂, T₁ and bSSFP values within each sample (Fig.2/3). BSSFP signal evolution as a function of flip angle was fitted with the theoretical expression³: $$$\frac{M_{SS}}{M_{0}}=\frac{sin(\alpha)}{1+cos(\alpha)+(1-cos(\alpha))*\frac{T_{1}}{T_{2}}}$$$. Optimal flip angle and steady-state level obtained from the fit were compared to values derived from T₁/T₂ mapping performed with CPMG/IR sequences using the formula³: $$$FA_{optim}=arcos(\frac{\frac{T_{1}}{T_{2}}-1}{\frac{T_{1}}{T_{2}}+1})$$$ and $$$M_{SS}=\frac{1}{2}*M_{0}*\sqrt{\frac{T_{2}}{T_{2}}}$$$.
Brain-like (BrainLo) phantom experiment: BrainLo was built using a 3D printer and filled with different solutions of agarose/copper-sulphate doped-water/deuterated-water to imitate brain tissue relaxation properties (Fig.3A). T₂/T₁ mapping was performed as described above with (resolution: 2x2x10 mm³, total acq time~10/8 mins for CPMG/IR sequences). bSSFP data (identical resolution, total acq. time~2 mins) were acquired with α=90º. First, the sum of squares, complex sum, magnitude sum, and maximum-intensity images were calculated, neglecting any additional frequency drift during multi-offset series’ acquisition (Fig.4). Secondly, F₀ and F_-1 modes were retrieved via a Fourier transform of the complex bSSFP frequency response, incorporating the measured frequency drift. Finally, a T₂ map was calculated using the F₀ and F_-1 configurations in an approach similar to MIRACLE relaxometry^9-12.

Results and Discussion

Sequence validation: As illustrated in Fig.2A, our samples had T₁/T₂ ratios ranging from ~1:1 to ~4.7:1. Banding-free maximum-intensity image could be reconstructed (Fig.2B). Fig.3A shows the evolution of the signal as a function of the flip angle. The sample with the lowest and highest T₁/T₂ ratio shows the highest and lowest signal respectively, which is in agreement with literature³. The optimal flip angle, steady-state signal level and T₁/T₂ ratio were close to expected values (Fig.3B). Note however, that the fit slightly deviated from data for low flip angle since in this regime, the maximum might not occur in the passband but at different frequency offsets. The steady-state signal at the highest T₁/T₂ ratio was slightly higher than predicted from the model.
Brain-like (BrainLo) phantom: As shown in Fig.4A, GM/WM/CSF compartments had T₂ and T₁ values measured using conventional techniques which match in vivo values at low-field². Slight residual banding was observed in each reconstruction, except for the maximum-intensity image (Fig.4B). The cause is probably small drifts in B₀ during acquisition. By calculating the difference between each dataset and the first dataset (Fig.5A), we estimated this additional drift, corrected it and produced banding-free F₀ and F_-1 modes (Fig.5B), which were used to reconstruct the T₂ map. The relative T₂ values are in the right ratio, but absolute values in WM/GM are ~40% lower than those obtained using the much slower conventional techniques. In future experiments we will investigate whether this is a systematic error, or simply reflects the fact that different acquisition schemes result in different apparent relaxation times.

Conclusion

POC imaging with low-field MRI (B₀<0.1T) is a potential game changer for low-income countries and patients in the intensive care unit. However, POC systems have intrinsically low SNR and relatively high ΔB₀. We illustrated the feasibility to reconstruct banding-free bSSFP images at 46mT on a portable POC system and to rapidly quantify T₂ based on the retrieved F₀ and F_-1 configurations.

Acknowledgements

This project has received funding from Horizon 2020 ERC Advanced PASMAR 101021218, ERC advanced grant SpreadMRI 834940 and the Dutch Science Foundation Open Technology 18981.

References

¹Saracanie, Front. Phys. (2020) ; ²O’Reilly et al., MRM (2021) ; ³Scheffler et al., Eur Radiol (2003) ; ⁴O’Reilly et al., ISMRM #63 (2022); ⁵Koolstra et al., MAGMA (2021); ⁶Saracanie et al., Sci. Rep. (2015) ; ⁷O’Reilly et al., MRM (2020); ⁸Scheffler et al., ISMRM #294 (2003) ; ⁹Heule et al., MRM (2013) ; ¹⁰Welsch et al., MRM (2009) ; ¹¹Heule et al., MRM (2014) ; ¹²Nguyen et al., MRM (2017)