Synopsis

We propose a fast 3D whole-brain steady-state chemical exchange saturation transfer (CEST) method using MR Multitasking. Exploiting the correlation between images throughout the space, time and frequency dimensions, the low-rank tensor framework shows possibility for further acceleration of steady-state CEST, so that a whole-brain Z-spectrum acquisition can be done within 5 min.

Introduction

Chemical exchange saturation transfer (CEST) can indirectly detect exchangeable protons in the water pool by pre-saturation at different frequency offsets. A steady-state CEST approach has been developed to efficiently acquire the whole Z-spectrum¹, but it still requires more than 10 min scan time for a 3D volume, which is too long for practical use.

In this work, we propose a fast 3D steady-state CEST method using MR Multitasking². This approach images not only the entire Z-spectrum, but also an additional time dimension describing the transient process towards steady-state within each offset frequency. We use a low-rank tensor model to leverage the correlation between images throughout the multi-dimensional array, resulting in a 5 min whole-brain steady-state CEST scan.

Methods

Image model and reconstruction

Image data are represented as a 5D image $$$A(\mathbf{x},z,\tau)$$$, where $$$\mathbf{x}$$$ represents 3 spatial dimensions, $$$z$$$ indexes different frequency offsets of pre-saturation, and $$$\tau$$$ is the time since the last saturation pulse. This image is reshaped as a 3-way tensor $$$\mathcal{A}$$$ (grouping together the three spatial dimensions), which can be modeled as low rank and partially separable⁵, i.e. $$\mathbf{A}_{(1)}=\mathbf{U_x}\mathbf{C}_{(1)}(\mathbf{U}_z\otimes \mathbf{U}_{\tau})^{\text{T}}$$ where columns of each $$$\mathbf{U}_{(\cdot)}$$$ contain the basis functions for corresponding dimensions, $$$\otimes$$$ denotes the Kronecker product, $$$\mathbf{A}_{(1)}$$$ and $$$\mathbf{C}_{(1)}$$$ are mode-1 matricized image tensor and core tensor respectively.

$$$\mathbf{A}_{(1)}$$$ is reconstructed by firstly estimating $$$\mathbf{\Phi} = \mathbf{C}_{(1)}(\mathbf{U}_z\otimes \mathbf{U}_{\tau})^{\text{T}}$$$ from frequently sampled training data, and then recovering $$$\mathbf{U_x}$$$ by fitting $$$\mathbf{\Phi}$$$ to the remaining imaging data: $$\mathbf{U_x}=\arg\min_{\mathbf{U}}\left\|\mathbf{d}-E[\mathbf{U\Phi}]\right\|_2^2+R(\mathbf{U})$$ where $$$\mathbf{d}$$$ is measured data, $$$E$$$ is the k-space encoding operator, and $$$R$$$ is an optional regularization functional.

A 4D image $$$B(\mathbf{x},z) = A(\mathbf{x},z,\tau_{\max})$$$ is extracted to be analyzed, representing the Z spectrum images at steady-state.

Sequence design

Fig.1 illustrates the pulse sequence and k-space sampling pattern. Each module (TR=70ms) contains a single-lobe Gaussian saturation pulse (t_sat=30ms, flip angle=1200°), followed by a spoiler gradient and 8 FLASH readout lines (flip angle=5°). The module repeats 80 times at one offset frequency, and then switches to another frequency without delay. K-space lines were sampled using 3D Cartesian encoding. The center line (k_y=k_z=0) was first sampled as training data, and 7 random lines were then sampled with Gaussian distribution in k_y and k_z direction as imaging data.

Experiments

MRI data were acquired on a 3T Siemens Vida system for a phantom study and human studies (n=4). Phantoms of [0, 0.15, 0.375, 0.75, 1.2, 1.5] mM BSA were used.

Steady-state imaging parameters were as follows. FOV=220×220×80mm³, matrix size=128×128×40, spatial resolution=1.7×1.7×2.0mm³. The acquisition time was 5.6s for each frequency, which is 2x faster than the original steady-state CEST method³. Images of 49 frequency offsets were acquired from -40ppm to 40ppm, with a 22.4s additional unsaturated acquisition at the beginning of the scan, for a total imaging time of 5 min. The water frequency (0ppm) was determined by Lorentzian fitting of the central Z-spectrum (|df|<1ppm).

Conventional 3D amide proton transfer-weighted (APTw) CEST images were also acquired at 6 frequency offsets (S₀, ±3/±3.5/±4ppm) as a reference^3,4. FOV=220×220×80mm³, matrix size=128×128×16, spatial resolution=1.7×1.7×2.0mm³, TR/FA=3000ms/12°, shots per slice=1, SENSE factor=2. 30 saturation pulses of 1200°, 30ms (duty cycle=50%) were used. B₀ correction was done with WASSR⁶. The total scan time was 12(6.4+5.6) min.

Results

Fig.2 shows MTR_asym(3.5ppm) maps of the phantom from the conventional method and the proposed method. The result from the proposed method shows good consistency with the conventional method (R²=0.99) and has superior SNR.

Fig.3A displays steady-state images of the central slice over the whole Z-spectrum; Fig.3B displays images of the whole volume at -20ppm.

Fig.4 shows in vivo MTR_asym(3.5ppm) maps, as well as mean MTR_asym values within ROIs chosen in WM(1) and GM(2) from healthy volunteers. Good consistency can be seen. In both methods, GM and WM have MTR_asym(3.5ppm) values around 0, with GM slightly higher.

Discussion

In the original steady-state method, further reduction of imaging time is very difficult, since enough data, according to the Nyquist criterion and limited parallel imaging (PI) factor, has to be acquired at each frequency. However, in our proposed method, the time required to reach steady state appears to be the only limiting factor for acceleration rather than the image modeling.

The proposed fast steady-state CEST method was 46x more efficient than conventional CEST: it provided images at 8.5x more frequency offsets and 2.5x as many slices in less than half the time.

Conclusion

Acknowledgements

References

1. Jones CK, Polders D, Hua J, Zhu H, Hoogduin HJ, Zhou J, Luijten P, Van Zijl PC. In vivo three-dimensional whole-brain pulsed steady-state chemical exchange saturation transfer at 7 T. Magnetic resonance in medicine. 2012 Jun;67(6):1579-89.

2. Christodoulou AG, Shaw JL, Nguyen C, Yang Q, Xie Y, Wang N, Li D. Magnetic resonance multitasking for motion-resolved quantitative cardiovascular imaging. Nature Biomedical Engineering. 2018 Apr;2(4):215-226.

3. Jiang S, Eberhart CG, Zhang Y, Heo HY, Wen Z, Blair L, Qin H, Lim M, Quinones-Hinojosa A, Weingart JD, Barker PB. Amide proton transfer-weighted magnetic resonance image-guided stereotactic biopsy in patients with newly diagnosed gliomas. European Journal of Cancer. 2017 Sep 1;83:9-18.

4. Zhou J, Blakeley JO, Hua J, Kim M, Laterra J, Pomper MG, van Zijl PC. Practical data acquisition method for human brain tumor amide proton transfer (APT) imaging. Magnetic Resonance in Medicine: An Official Journal of the International Society for Magnetic Resonance in Medicine. 2008 Oct;60(4):842-9.

5. Liang ZP. Spatiotemporal imaging with partially separable functions. In Biomedical Imaging: From Nano to Macro, 2007. ISBI 2007. 4th IEEE International Symposium on 2007 Apr 12 (pp. 988-991). IEEE.

6. Kim M, Gillen J, Landman BA, Zhou J, van Zijl PC. Water saturation shift referencing (WASSR) for chemical exchange saturation transfer (CEST) experiments. Magnetic resonance in medicine. 2009 Jun 1;61(6):1441-50.