Background and Purpose

CEST imaging shows great potentials in many clinical applications, such as the classification and staging of tumors. This technique has been widely used in brain^{1, 2} and breast³. However, there are still many challenges when applying CEST in the liver disease diagnosis⁴.

Firstly, the effects of fat and NOE usually contaminate the positive side of Z-spectra in the conventional asymmetry analysis of MTR (MTR_asym), and consequently affect the quantitative analysis of APT signals. Secondly, B₀ inhomogeneity is usually severe in the abdomen, which deviates the water peak from the center frequency and biases the APT estimation. In brain APT imaging, the B₀ shift is small and can be effectively corrected by using the Water Saturation Shift Referencing (WASSR) ⁵ sequence. In liver imaging, the WASSR sequence is very time-consuming for the acquisition between -2ppm and 2ppm frequency bands. It may also generate “field drifts” among different sequences due to temperature variation, airflow, and homogenization mechanism⁶, which consequently affects the accuracy of B₀ corrections. To address these issues, we proposed a deep learning approach for CEST imaging techniques which simultaneously corrects for B₀ inhomogeneity and separates APT signals from other effects.

Materials and Methods

The pipeline of data processing is shown in Figure 1. We recruited four healthy volunteers by using a 3T MRI system (Ingenia CX, Philips Medical, Netherlands) and a 32-channel body coil. The CEST acquisitions were acquired by using a Fast Spin-Echo (FSE) sequence. Saturation transfer was applied by using RF-excitation pulses with an amplitude of 1µT lasting for 1000ms. The imaging parameters were field of view (FOV) = 302 × 380 mm² and matrix size = 280 × 352, slice thickness = 7 mm, repetition time (TR) = 45ms, echo time (TE) = 4.6ms, flip angle (FA) = 10°, and bandwidth = 320 Hz/pixel. The whole process was repeated 32 times with different off-resonance frequency offsets in the range between −10 ppm and 10 ppm with a variable step size. An additional unsaturated reference image was collected.
We implemented our approach in two steps. Firstly, considering that the raw Z-spectrum (Figure 1b) was affected by the large B₀ inhomogeneity, we extracted the three minima points of Z-spectrum to calculate the estimated B₀ shift based on point-to-point comparisons, and interpolated it with a cubic spline function to represent the preliminary corrected Z-spectrum (Figure 1c). Secondly, we used a neural network model for a pixel-by-pixel prediction of the APT effect and the accurate B₀ shift simultaneously. The neural network was trained based on simulated background reference Z-spectrum (Figure1a), which was reconstructed by using the two-pool Bloch-McConnell equation. During model training, 13 reference points from each Z-spectrum without fat or CEST effects (red solid dots in Figure 1d) were selected to fit the entire background reference Z-spectrum (blue line in Figure1d), after taking into account the magnetization transfer and direct saturation effects ⁷, and the extra random B₀ shifts. During the model evaluation, the neural network took in the preliminary corrected Z-spectrum (Figure1c) and predicted the corresponding background reference Z-spectrum as well as the accurate B₀ shift (bluish violet dashed curve in Figure 1f). The APT effect was then estimated by calculating the difference between the predicted background reference Z-spectrum and the corrected Z-spectrum (shaded area in Figure 1f). A pixel-by-pixel reconstruction algorithm was applied to generate the APT image of the abdomen (Figure 1g).

Results

Figure 2 shows an example of the corrected abdominal B₀ maps and the corresponding MTR_asym maps from a healthy subject. Compared to the traditional method, our proposed method generated a smoother version of the B₀ map (Figure 2b&e) along with more accurate corrections for the effect of “field shift” (Figure 2c&f) and resulted in a better estimation of CEST imaging, e.g. MTR_asym(Figure 2d&g).
In addition, the proposed method separated the APT effect from other effects and showed better tissue contrasts between the liver and right kidney (as shown in Figure 3). Moreover, a higher contrast map of APT_REX was generated, both APT_REX and APT maps obtained from our method showed a better Contrast-Noisy Ratio (CNR) between tissues compared with the traditional MTR_asym map.

Discussion and conclusions

In-vivo liver CEST imaging is challenged and usually affected by various factors, for instance, the limited scanning time due to the breath-hold scan strategy, the mixed effect of fat and NOE, and severe B₀ inhomogeneity. In this study, we proposed a deep learning approach for CEST image analysis which corrects for B₀ inhomogeneity and separates APT signals from mixed effects of MTR_asym. Comparing to the traditional approaches, our method not only corrected B₀ inhomogeneity within a shorter scanning time by discarding additional correction sequences, but also generated a cleaner map of APT signals which showed a uniform distribution within the same tissue and higher contrasts between tissues. Our method also showed highly reproducible results on four healthy volunteers. This study provides new avenues for in-vivo liver CEST imaging in clinical applications.

Acknowledgements

No acknowledgement found.

References

1. Zhang JX, Zhu WZ, Tain RW, Zhou XJ, Cai KJ. Improved differentiation of low-grade and high-grade gliomas and detection of tumor proliferation using APT contrast fitted from Z-spectrum. Mol Imaging Biol 2018;20:623-631.

2. Cai K, Singh A, Poptani H, et al. CEST signal at 2ppm (CEST@2ppm) from Z-spectral fitting correlates with creatine distribution in brain tumor. NMR Biomed 2015;28:1-8.

3. Zhang S, Seiler S, Wang X, et al. CEST-Dixon for human breast lesion characterization at 3 T: A preliminary study. Magn Reson Med 2018;80:895-903.

4. Seo N, Jeong HK, Choi JY, Park MS, Kim MJ, Chung YE. Liver MRI with amide proton transfer imaging: feasibility and accuracy for the characterization of focal liver lesions. Eur Radiol 2021;31:222-231.

5. Kim M, Gillen J, Landman BA, Zhou J, van Zijl PC. Water saturation shift referencing (WASSR) for chemical exchange saturation transfer (CEST) experiments. Magn Reson Med 2009;61:1441-1450.

6. Sun PZ, Farrar CT, Sorensen AG. Correction for artifacts induced by B-0 and B-1 field inhomogeneities in pH-Sensitive chemical exchange saturation transfer (CEST) Imaging. Magn Reson Med 2007;58:1207-1215.

7. Zaiss M, Zu Z, Xu J, et al. A combined analytical solution for chemical exchange saturation transfer and semi-solid magnetization transfer. NMR Biomed 2015;28:217-230.