Introduction

Preclinical investigations using mouse models, including genetically manipulated mouse models, provide the opportunity for the mechanistic understanding of disease pathophysiology, as well as for the longitudinal evaluation of disease progression and the response to novel therapeutic interventions. MR fingerprinting (MRF) allows simultaneous T₁ and T₂ mapping with unprecedented efficiency.^1,2 Previously, we developed a single-slice MRF sequence that enabled simultaneous tracking of two MRI contrast agents in a single dynamic contrast-enhanced (DCE)-MRI scan.^3,4 In the current study, we present further development of a 3D MRF sequence for brain-wide T₁ and T₂ mapping in mice. The potential of accelerated data acquisition for DCE-MRI studies was examined by comparing T₁ and T₂ mapping using retrospectively undersampled data.

Methods

Sequence Design: The FISP-based 3D MRF sequence acquires a total of 768 frames partitioned into 16 segments. Segments 1, 5, 9, 13 were preceded by an inversion module with an inversion time of 21, 56, 400, 150 ms, respectively. Eight segments were preceded by a T₂ prep module with a mixing time of either 45 ms (Segments 3, 7, 11, 15) or 60 ms (Segments 4, 8, 12,16). Data acquisition in each segment used flip angles ramped up sinusoidally 6° to a maximum value between 8° and 17°. Constant echo time (2.0 ms) and repetition time (10.0 ms) were used. The acquisition time of a single fingerprint was ~21 s. 3D MRF data were acquired using a stack-of-spiral trajectory.^5,6 The spiral trajectory sampled the entire k-space in 48 interleaves with a FOV of 30 × 30 mm², yielding an in-plane resolution of 0.23 x 0.23 mm². Data acquisition in the slice direction covered 9 mm thickness in 9 phase-encoding steps, yielding a through-plane resolution of 1 mm.
Data Acquisition: All experiments were performed on a Bruker 9.4T scanner with a 35-mm 1H volume coil. Fully-sampled 3D MRF data were acquired in vitro from a phantom and in vivo from 3-month-old male C57BL/6 mice. The phantom consisted of 7 compartments with varied manganese concentrations ranging from 50 to 350 µM. For in vivo scans, the mouse was anesthetized with 1% isoflurane mixed with 100% oxygen delivered via a nose cone. A custom-built cradle was used to hold the mouse in a prone position, with the brain positioned at the isocenter of the magnet. The respiration rate was monitored and maintained at 80 to 100 breaths/minute by adjusting the anesthesia level. The body temperature was monitored and maintained at 36.5°C to 37°C by circulating warm air (37.5°C) into the scanner bore.
Data Analysis: All images were reconstructed in MATLAB using the NUFFT toolbox. A dictionary was generated by solving the Bloch equation according to the MRF sequence. T₁ and T2 values were derived from the best matching dictionary entry that was determined by calculating the inner product between the acquired fingerprint and all the dictionary entries. For evaluation of the undersampling capacity of the 3D MRF sequence, retrospective undersampling was performed on the fully sampled datasets. In-plane undersampling factors of 4, 8, 16, and 24, and a through-plane undersampling factor of 3 were evaluated. Normalized root-mean-square error (NRMSE) was used to evaluate the accuracy of undersampling.

Results

Reconstructed images, fingerprints, and T₁ and T₂ maps generated from both fully sampled and retrospectively undersampled phantom data are shown in Figure 1. With high in-plane undersampling factors (R_xy= 8, 16), aliasing artifacts became dominant in the images, and aliasing noise increased drastically in the corresponding fingerprints (Figure 1A-B). Adding through-plane undersampling (R_xy= 8, R_z= 3; R_xy= 16, R_z= 3) further increased the aliasing noise in both the reconstructed images and the fingerprints. However, a total undersampling factor of 48 had minimal impact on the estimated T₁ and T₂ maps (Figure 1C-D). While NRMSE increased with increased undersampling factor, it was below 10% for a total undersampling factor of 48 (Figure 1E). Figure 2 shows T₁ and T₂ maps from both fully sampled and retrospectively undersampled data acquired from a mouse brain. The T₁ and T₂ maps demonstrate that a 16-fold in-plane acceleration combined with a 3-fold through-plane acceleration still preserved the majority of the details with no obvious artifacts.

Discussion and Conclusion

A 3D MRF sequence was developed for simultaneous T₁ and T₂ mapping of the entire mouse brain. Our results demonstrate that up to 16-fold in-plane and 3-fold through-plane undersampling can be achieved with adequate accuracy. This undersampling capacity will enable whole-brain T₁ and T₂ mapping in 3 to 5 minutes.

Acknowledgements

This work was supported by a grant from the U.S. National Institute of Health (R01 EB23704).