Introduction

Imaging of brain neurometabolism and oxygenation is of great significance for tissue characterization in a variety of brain diseases such as stroke and brain tumor.¹ Recently, simultaneous 3D MRSI and T₂* mapping has been achieved using SPICE (SPectroscopic Imaging by exploiting spatiospectral CorrElation).^2,3 Quantitative T₂' map that reflects tissue oxygenation could be calculated from the T₂* map and a T₂ map acquired from a spin-echo scan, thus making SPICE potentially applicable for concurrent neurometabolism and oxygenation imaging.

However, both T₂* and T₂ measurements suffer from system imperfections such as B₀ and B₁ inhomogeneities in practice. More specifically, the presence of B₀ field inhomogeneity introduces signal dephasing and distortion, thus leading to faster T₂* decay in regions with large susceptibility.^4,5 The B₁⁺ inhomogeneity causes unideal magnetization refocusing, which makes the spin echo chain deviate from the desired exponential T₂ decay.⁶ Therefore, the effects of field inhomogeneities on T₂* and T₂ mappings need to be corrected for accurate and reliable T₂' measurements.

In this work, we propose to correct the B₀ field effects on T₂* mapping using a method leveraging both high-resolution B₀ field map and pre-learned subspaces, and correct the B₁⁺ inhomogeneity effects on T₂ mapping using the dictionary-based estimation built on Bloch simulations.⁷ The results show that the proposed method could provide improved T₂' mapping and achieve reliable simultaneous brain oxygenation and neurometabolic imaging for normal and pathological human brain tissues in SPICE experiments.

Methods

To correct the signal dephasing in T₂* maps, the field-corrected water spatiotemporal functions were reconstructed using a high-resolution B₀ field map and a pre-learned subspace. More specifically, based on the partial separability model,⁸ the field-corrected spatiotemporal functions of water signals could be represented using a low-dimensional subspace, so the measured signals can be expressed as:
$$d(k,t) = Ω_{k,t}(F((\sum_{l=1}^{L}c_1(x)φ_1(t))e^{i2πB_0(x)t})) + η(k,t),$$
where $$$d$$$ is the measured limited k-space data, $$$Ω_{k,t}$$$ the sampling operator, $$$F$$$ the Fourier operator, $$$\{φ_1(t)\}$$$ the subspace basis and $$$\{c_1(x)\}$$$ the corresponding subspace coefficients.

In this work, the basis functions $$$\{φ_1(t)\}$$$ were learned from the high-resolution training data where the B₀ field inhomogeneity was effectively corrected. To generate the high-resolution field map, a low-resolution field map was first estimated from the measured water signals and then super-resolution was performed via local polynomial approximation.⁹ With the learned signal subspace and high-resolution field map, the reconstruction was performed via solving the following optimization problem:
$$\boldsymbol{\widehat{c}} = arg{\mathop \min_\boldsymbol{c}}\|\boldsymbol{d}-Ω_{k,t}FB_H \boldsymbol{Φc}\|_2^2 + λ\|W∇\boldsymbol{c}\|_2^2,$$
where $$$\boldsymbol{d}$$$, $$$\boldsymbol{Φ}$$$, $$$\boldsymbol{c}$$$ are the vector/matrix form of $$$d(k,t)$$$, $$$\{φ_1(t)\}$$$ and $$$\{c_1(x)\}$$$, $$$B_H$$$, $$$∇$$$, $$$W$$$ the operators representing high-resolution field map, total variation, and edge-preserving weights derived from the tissue field, respectively. With $$$\boldsymbol{\widehat{c}}$$$ determined, the field-corrected signals was synthesized as $$$\hat{ρ} = \boldsymbol{Φ\hat{a}}$$$, from which we derived the field-corrected T₂* map.

To correct the bias of T₂ estimation induced by the B₁⁺ inhomogeneities, a dictionary method based on Bloch simulation was used.⁷ The dictionary was built covering a range of T₂ values (1 to 1000 ms), T₁ values (500 to 5000 ms) and flip angles (120 to 180 degrees). The T₂ value of each spatial voxel was estimated by searching the dictionary to find the closest simulated curve to the measured spin-echo curve in the L₂-distance sense. Then, the T₂' map was generated using the improved T₂* and T₂ maps. The neurometabolite maps were obtained using the standard processing pipeline of SPICE.^3,10

In vivo experiments were carried out on a 3T Siemens Skyra MR scanner with local Institutional Review Board approved. The acquisition protocols included SPICE (1.2 x 1.2 x 1.9 mm³ for T₂*, 2.0 x 3.0 x 3.0 mm³ for MRSI, FOV = 240 x 240 x 72 mm³, FA = 27°, TE = 1.6 ms, TR = 160 ms) and multi-echo spin echo sequence (0.4 x 0.4 x 5.0 mm³, FOV = 230 x 230 x 126 mm³, FA = 180° , TE = 10.5, 21.0, 31.5, 42.0, 52.5, 63.0 ms, TR = 2000 ms).

Results and Discussion

Fig 1 shows the localized FID signals from the regions with large B₀ field inhomogeneities. The original FID signals suffered from noticeable distortions which could affect T₂* estimation, while the FID signals reconstructed using the proposed method showed much improved signal decay. Fig 2 shows the T₂* maps estimated from the original data and from the reconstructed data with field effects corrected. As can been see, noticeable signal loss from dephasing existed in the original T₂* map near the sinus region. In contrast, the T₂* map obtained by the proposed method had much reduced image distortions. Fig 3. shows the improved performance of T₂ quantification using dictionary-based approach over exponential fitting when comparing mean tissue T₂ values to reference.¹¹ The potential applications of the proposed simultaneous brain oxygenation and neurometabolic imaging in clinical settings are demonstrated in Fig 4. Simultaneous increased T₂' values along with reduced NAA to creatine ratio and elevated choline to creatine ratio were shown in the tumor region of a glioma patient, in line with the literature.^12,13

Conclusions

The proposed method achieved improved T₂' mapping in simultaneous neurometabolic and oxygenation imaging experiments using SPICE. It will have many potential applications in a variety of brain diseases studies.

Acknowledgements

Y. L. is funded by National Science Foundation of China (No.61671292 and 81871083) and Shanghai Jiao Tong University Scientific and Technological Innovation Funds (2019QYA12).