NMR microscopy provides a variety of image contrast with a high spatial resolution. However, NMR multi-parameter (T1, T2, diffusion, etc.) mapping with large matrix sizes is time consuming, especially for 3D mapping, and is practically difficult to apply to living samples because of their perishable nature. In this study, we introduced magnetic resonance fingerprinting (MRF) [1] technique to 3D NMR multi-parameter microscopy, which could reduce the scan time largely. We verified the feasibility of the relaxation and proton density mapping using a vertical wide bore superconducting magnet.

The MRI system consisted of a vertical wide bore SC magnet (Oxford Instruments, field strength = 4.74 T, bore diameter = 88.3 mm) and a saddle-type RF probe and shielded gradients (Doty Scientific Inc.) (Fig. 1(a)). The MRF images were acquired from CuSO₄–doped water phantoms with varied concentrations and a seedless grape berry (Fig. 1(b)).

The 3D MRF acquisition was developed from a 3D fast imaging with steady-state free precession (FISP) [2,3]. The unbalanced gradient moments along the readout direction was used to destroy the coherence of the transverse magnetization and avoid the banding artifact. The MRF acquisition was initiated with an inversion pulse, which was followed by 100 successive FISP acquisition periods with varying FA and TR. TE was held constant (8 ms). FA and TR were determined by pseudo-randomly according to the method proposed by Gao et al. [3]. FA was varied sinusoidally (5 to 90o) and TR was varied using a Perlin noise pattern (20 to 30 ms). For the acquisition of the MRF signal evolutions, the same line of k space for each of 100 sequential images during each MRF scan repetition period (5 s) was measured. The number of excitation (NEX) was 1. The matrix size was 256 × 256 × 8, FOV was 24.1 mm × 24.1 mm × 40 mm, and the voxel size was 94 um × 94 um × 5 mm. The measurement time was 2.85 hours.

The MRF dictionary was created using a home-written Bloch simulator considering the intravoxel dephasing effect, which was implemented in a C/C++ programming environment. The MRF dictionary consisted of 202500 profiles generated from 100 T₁ values (20-2000 ms: increment, 20ms) and 25 T₂ values (20-500 ms: increment, 20ms) for the grape measurement. Unrealistic profiles with T₁<T₂ were excluded. The dictionary was matched to the acquired MRF signal evolution profile using vector-based inner product comparisons.

Figure 2 shows T₁, T₂, and proton density maps of the CuSO₄-doped water phantoms from the conventional 2D spin echo method and MRF. The MRF-based T1 estimates for all phantoms were not significantly different from the conventional SE T₁ maps, whereas the MRF- and SE-based T₂ estimates were different from each other. For the MRF, the T₂ values were not uniform in each phantom, which may be caused by the B₁ inhomogeneity. The SE measurements showed that the T₂ difference between the phantoms were fairly small (34 ms), and this could not be distinguished by the particular MRF sequence used in this study.

Figure 3 shows axial slices of MRF T₁, T₂, and proton density maps of the grape berry. In the T₁ and T₂ maps, network structures with the small relaxation times were observed. This network would be assigned to the vascular bundles which consist of the smaller tissues than the surrounding region and have the smaller relaxation times because of the restricted diffusion. At the peripheral region, the relaxation times were decreased largely, which may also result from the B₁ inhomogeneity. In the proton density image, anatomical structures such as a degenerated seed in the center and small dark spots in the flesh were observed.

The measurement time was reduced largely by the MRF acquisition. For example, the measurement time of 2D SE for the phantom experiments was 5.18 hours for T₁ and 11.38 hours for T₂. The total time was 16.56 hours which corresponds to 46.8 hours for a 3D SE measurement (256 × 256 × 8 matrices) with the equivalent slice thickness (5 mm) and signal-to-noise ratio (SNR). This is 5.8 times and 13.71 hours longer than the MRF acquisition. As the SNR becomes small, the time difference between the conventional SE and MRF becomes much longer.

The 3D MRF-FISP for the grape berry provided the T₁, T₂, and proton density maps in the short measurement time that can be used to extract important structural information. We concluded that the MRF microscopy is a promising tool for 3D multi-parameter mapping of living samples with large matrix sizes.