Introduction

Both quantitative MR relaxometry and metabolic imaging using MRSI have long been of great interest in both scientific research and clinics. But traditional imaging methods for each modality require long scan times (more than 20 minutes), and they are usually acquired and processed separately, thus leading to longer total scan times.^1,2 Recently, advanced methods have been developed for fast parametric mapping like MRF and EPTI,^3,4 and SPICE has demonstrated the capability of simultaneous high-resolution imaging of water and metabolite signals.^5-8 In this work, we proposed a new method to achieve fast T₁ and T₂ mapping with metabolic imaging leveraging these advances. The basic SPICE sequence was extended with variable flip angles and T₂-preparation pulses for T₁ and T₂ mapping; and the auxiliary data with limited and sparse coverage of (k, t)-space were integrated with the unsuppressed water signals from the SPICE scan in image reconstruction. As a result, quantitative T₁ and T₂ maps at 1.0×1.0×2.0 mm³ resolution were obtained in only one-minute extra scan time.

Methods

Extended SPICE sequence for T₁, T₂ mapping: As illustrated in Figure 1, the proposed acquisition sequence kept the basic features of the SPICE sequence while extending it with variable flip angles and T₂-preparation modules for T₁ and T₂ mapping. The excitation pulses with different flip angles (FAs) were used to collect signals with different T₁ weightings while the preparation pulses with different preparation times (TPs) were used for measuring signals with different T₂ weightings. The T₂ preparation pulses included three pairs of hard pulses to reduce sensitivity to B₀ and B₁ inhomogeneities. Figure 2 shows the proposed sampling of (k, t)-space: first, the SPICE sequence was used to acquire the metabolite signals and high-resolution water signals;⁷ then it was followed by the T₁ frames with different FAs and T₂ frames with different TPs; (k, t)-spaces of both the T₁ and T₂ frames were sparsely sampled using CAIPIRINHA trajectories (effective acceleration factors were 72, with a factor of 3 in ky and a factor of 24 in (kz, t)-space). The acquisition parameters of the current implementation were: FOV: 240×240×72 mm³, TR/TE: 160/1.6 ms, echospace: 1.76 ms, matrix size of SPICE frame = 110×218×72, matrix size of the central region of SPICE frame and T₁/T₂ frames = 110×78×24, TP = 0/20/40/60/80 ms and FA = 12/17/22/27/32˚, scan time: 7 min for SPICE frame, 1 min for T₁ and T₂ frames in total.
Reconstruction from sparse and limited (k, t)-space data: Since the T₁/T₂ frames and the SPICE frame shared the same acquisition setup except for the use of variable flip angles or T₂-preparation, the differences between the signals of different frames were only T₁/T₂ weightings which did not affect their temporal patterns. Therefore, the relationship between the spatiotemporal functions of the signals in the SPICE frame (denoted as $$$\rho_r(x,t)$$$ ) and in the T₁/T₂ frames (denoted as $$$\rho_i(x,t)$$$ ) could be expressed using a low-order generalized series (GS) model:⁹
$$\rho_i(\mathbf{x},t)=\rho_r(\mathbf{x},t)\cdot \sum_{n_i=-N_i/2}^{N_i/2}c_{n_i}(\mathbf{x})e^{i2{\pi}{n_i}{\Delta}ft}$$
where $$$i$$$ denotes a specific T₁/T₂ frame, $$$c_{n_i}(\mathbf{x})$$$ the model coefficients and $$$N_i$$$ the GS model order. Reconstruction of $$$\rho_i(\mathbf{x},t)$$$ from the sparsely measured data (denoted as $$$d_i$$$) became solving the following optimization problem:
$$\widehat{c}_i= \mathrm{arg}\mathrm{min}_{c_i}{\left \| d_i-{\Omega}{FBS}(G(\rho_r)c_i)) \right \|^2_2+R(c_i)}$$
where $$$\widehat{c}_i$$$ and $$$\rho_r$$$ are the vector forms of $$$\left \{ c_{n_i}(\mathbf{x}) \right \}$$$ and $$$\rho_r(\mathbf{x},t)$$$, $$$\Omega,F,B,S,G,R$$$ represent operators associated with k-space sampling, Fourier transform, field inhomogeneity effect, coil sensitivity weighting, GS modeling and regularization, respectively. After the GS coefficients $$$\widehat{c}_i$$$ were determined, the reconstructed signals $$$\widehat{\rho}_i(\mathbf{x},t)$$$ could be synthesized using the GS model and the reference SPICE signals. With the reconstructed signals of all the T₁/T₂ frames, T₁, T₂ and PD maps could be generated by fitting to the relaxation model.
The generation of metabolite maps from the SPICE frame followed the existing methods.^5-12

Results

Both phantom and in vivo experiments on healthy subjects were carried out on a 3T Siemens scanner. Figure 3 shows the T₁ and T₂ maps generated from the NIST/ISMRM system phantom. The T₁ and T₂ maps reconstructed from the sparse data were compared with those estimated from fully sampled data, which showed good consistency. Figure 4 shows some representative T₁ and T₂ maps obtained from a healthy subject using the proposed method. Note that the extra scan time for obtaining the T₁ and T₂ maps at 1.0×1.0×2.0 mm³ resolution was one minute. Figure 5 shows a complete set of results from an 8-minute scan, including the quantitative T₁, T₂, PD maps and spatiospectral distributions of metabolite signals (NAA, Cr, Cho, …).

Conclusion

Quantitative T₁ and T₂ maps of the brain at 1.0×1.0×2.0 mm³ resolution were successfully obtained in one-minute extra scan time along with SPICE. This new capability enables simultaneous high-resolution parametric mapping and metabolic imaging of the human brain, which can provide quantitative T₁, T₂, PD maps at 1.0×1.0×2.0 mm³ resolution and metabolite signals (NAA, Cr, Cho, …) at 2.0×3.0×3.0 mm³ resolution in a total 8-minute scan.

Acknowledgements

This work reported in this paper was supported, in part, by the National Institutes of Health (R21-EB023413).