Introduction

The rosette trajectory¹ is a sampling scheme that crosses the center of k-space multiple times during the acquisition. Recently, the rosette trajectory has been used with cardiac MRF^2,3 to enable T₁ and T₂ mapping with fat or water suppression⁴. In addition, the multiple k=0 crossings of the rosette trajectory have been used to create individual echo images, which can be used to fit R₂*⁵ and B₀ values. While previous work has investigated R₂* estimation from Dixon MRF⁶, this work investigates an alternate sampling method. This abstract proposes an approach for using rosette MRF data for estimation of T₁, T₂, and R₂* properties simultaneously.

Methods

MRF data were acquired on a 1.5T Sola (Siemens Healthineers, Erlangen, Germany) using a five-lobe rosette trajectory. The sequence was based on the cardiac sequence described in ³ and used the following parameters: TR = 17.4 ms, TE (start of sampling) = 1 ms, readout duration = 15.2 ms, 768 repetitions, total scan duration = 14.7 s, flip angles ranging between 4.5° and 15°, 192 x 192 matrix size, and 300 mm² FOV. Inversion pulses were used with inversion times between 21 and 260 ms to sensitize the signal to T₁; T₂ prep pulses were used with echo times between 40 and 80 ms. The sequence was run continuously without cardiac gating. Data were acquired on the T₂ array of the ISMRM/NIST MRI system phantom. Reference T₁ and T₂ values were measured using a cardiac MRF sequence with a spiral trajectory, and reference R₂* values were measured using a series of 6 gradient echo images with echo time of 2.38n ms for the n^th image in the series.
Quantitative tissue property maps were reconstructed in a two-step process, shown in Figure 1. The first step of this process is to estimate R₂* and B₀ values from a set of multi-echo images collected during the rosette readout. First, the rosette trajectory was separated into 10 half-lobes, with effective echo times of 1+3.03n ms for a half-lobe around the n^th k=0 crossing, and data from half-lobes was reconstructed using the NUFFT⁷ to generate 10 images per TR. These images were summed over all TRs to create a set of 10 composite images with different echo times. R₂* and B₀ values were then fit pixelwise to these composite echo images using a nonlinear least-squares algorithm.
The second step of this process involves applying the MRF pattern matching algorithm to estimate T₁ and T₂ maps. The reconstruction dictionary was simulated with the following tissue property value ranges for T₁ = [50:10:750 775:25:1200 1250:50:3000], T₂ = [5:1:50, 55:5:200 210:10:450 475:25:700], B₀ = [-15:5:15], and included corrections for slice profile and inversion efficiency⁸.
To evaluate the accuracy of the T₁, T₂, and R₂* maps, quantitative values were determined from the mean value of a sphere within the NIST phantom and compared to reference values. Linear regression was performed between estimated and reference values in each case to assess the quality of the estimation. Spheres with T₂ values greater than 300 ms were excluded from analysis, as the sequence used was not designed to be sensitive to high T₂ values.

Results

Reference and estimated R₂* maps are shown in Figure 2, with a regression plot and table of mean sphere values. For R₂* values, the linear regression had a slope of 1.087, and an intercept of 4.62 s^-1. T₁ and T₂ maps, regression plots, and mean sphere values are shown in Figure 3. For T₁, the linear regression had a slope of 0.964, and an intercept of 8.08 ms. For T₂, the linear regression had a slope of 0.828 and an intercept of 4.52 ms.

Discussion

In this work, half-lobes of rosette MRF data were separated and then combined over the acquisition to extract T₂* and B₀ maps from the rosette data in addition to T₁ and T₂ maps. While T₁ and T₂ can be determined using the traditional pattern matching MRF pipeline, R₂* is fit as an exponential function of readout time as in more traditional mapping approaches.
For R₂* values, the linear regression slope of 1.087 indicates that the values measured show the correct trend, although rosette MRF tends to slightly overestimate R₂* values compared to the GRE sequence. This effect increases at higher R₂* values, which could indicate that the rosette k=0 crossing points do not occur frequently enough with this trajectory to capture rapid T₂* decay effects. The regression for T₁ shows that this approach offers similar values when compared to spiral MRF; T₂ was underestimated as compared to the values measured with spiral MRF, as shown in ⁴.
The main difference between sequences used in previous work on rosette MRF⁴ and this approach is the sampling time. A longer rosette was chosen for this work to allow for more T₂* decay to occur during the readout, theoretically allowing for more accurate estimation of shorter R₂* values. This increased scan time by approximately 70%, but this could likely be reduced depending on desired R₂* accuracy and with further research.

Conclusion

R₂* and B₀ values can be estimated alongside T₁ and T₂ values when using a rosette readout to collect MRF data.

Acknowledgements

NSF/CBET 1553441, NIH/NIDDK 1R01CA208236, NIH/NHLBI R01HL094557, Siemens Healthineers