Introduction

Fast Magnetic Resonance Fingerprinting (MRF) acquisitions rely on accelerated k-space acquisition schemes such as undersampled spiral readouts. In the presence of both aqueous and fatty tissues, however, even short spiral readout gradients cause heavy blurring artifacts in the images. This hampers correct mapping of T₁, T₂ or the fat signal fraction¹. On fully sampled spiral images, a conjugate phase reconstruction (CPR) has successfully been used after Dixon water-fat separation to correct for blurring². CPR was performed separately on the water and the fat image. The off-resonance map needed for CPR is, conveniently, a by-product of the 3-point Dixon processing. For undersampled spiral MRF, the use of CPR has improved the matching results of MRF for aqueous tissues^3,4. In this work, we correct for water and fat blurring in undersampled spiral MRF by combining MRF with a Dixon acquisition and CPR correction.

Methods

We applied a spiral Dixon multi-acquisition MRF sequence on the upper right leg of a healthy volunteer and acquired 3 MRF trains S₀, S₁ and S₂, which consisted each of i = 1..500 acquisitions using RF pulses of variable flip angles and a constant repetition time (Figure 1). Signal was acquired with one rotating spiral interleave of 5 ms length, undersampling the k-space by a factor of 25. The MRF trains differed in their echo times (TE₀ = 4.61 ms, TE₁ = 5.76 ms, TE₂ = 6.91 ms) and had a duration of 13.5 ms each. For post-processing, a mean off-resonance map was calculated from the temporal averages, denoted by $$$<>$$$, over the undersampled signals S₀ and S₂: $$$<f_0> = arg\left(\frac{<S_2>}{<S_0>}\right)$$$. Proceeding similarly to², $$$<f_0>$$$ was smoothed and subsequently used to separate the data into its water and fat contribution S_w^blurred and S_f^blurred, which consisted of 500 complex images each. These were deblurred by CPR, using $$$<f_0>$$$ for S_w^blurred and $$$<f_0>-440$$$ Hz for S_f^blurred. Finally, the deblurred water and fat MRF data S_w^CPR and S_f^CPR were recombined to one MRF dataset. A dictionary of possible signal evolutions was calculated using extended phase graphs⁵.The best matching parameters (T₁, T₂) were selected for each voxel by inner product matching between the deblurred MRF data and the dictionary. Furthermore, water and fat signal images were approximated by temporally averaging over the CPR-corrected as well as over the uncorrected MRF data. A Cartesian 3-point Dixon sequence was acquired as a comparison data set to judge the accuracy of our post processing method.

Results

$$$<f_0>$$$ shows similar large-scale features to the Cartesian Dixon off-resonance map, but some finer ring-like subfeatures which mostly disappear after smoothing (Figure 2). After Dixon separation, the fat image <S_f^blurred> shows heavy blurring artifacts (Figure 4b). The T₁ and T₂ maps reflect these artifacts when matching the uncorrected MRF data to the dictionary (Figure 3a+b). After CPR, the boundary between muscle tissue and subcutaneous fat is well distinguishable and the relaxation parameter maps (Figure 3c+d) show improved homogeneity. Average values T₁^muscle=(840±90) ms, T₂^muscle =(41±15) ms, T₁^fat =(300±60) ms and T₂^fat =(120±40) ms are similar to literature⁶. In the parameter maps, finer substructures in the muscle are not visible and a spiral flow artifact remains around the vessel. Yet, the deblurred water and fat maps (Figure 4c+d) show the same features as the Cartesian 3-point Dixon scan (Figure 4e+f). When calculating the fat signal fraction $$$S_f / (S_w+S_f)$$$, the CPR-deblurred MRF result slightly overestimates the fat signal fraction in the area of muscle tissue in comparison to the Cartesian Dixon sequence (Figure 5).

Discussion

Parameter maps were improved after a Dixon water-fat separation and subsequent CPR. Finer substructures in the T₁ and T₂ maps are probably non-visible due to their low SNR, e.g. in gaps or areas of low fat fraction. Some remaining signal after imperfect deblurring may also play a role. This may be improved by employing a CPR with multi-frequency interpolation. Obtaining water and fat images by temporally averaging over the signal evolutions remains an approximation: As aqueous tissue has a larger T₁ than fatty tissue, a larger portion of its fingerprint signal is in fact negative. Thereby, the fat fraction is slightly overestimated.

Conclusion

This work shows that it is possible to correct for water-fat blurring in spiral MRF data without the use of fat-suppression methods. This makes spiral MRF more suited for clinical applications like breast imaging, which else suffer from blurring artifacts. According to the original idea of MRF, the undersampled acquisition results in a sequence duration of only 1:36 min (13.5s/MRF train). Future work includes the application to other anatomical areas.

Acknowledgements

This research project is supported by the European Commission through the Marie Sklodowska-Curie Actions, Innovative Training Program - European Industrial Doctorate, Project Nr. 642445.