Introduction

X-nuclei based imaging (e.g. ¹³C, ²³Na, ¹²⁹Xe) usually suffers from low signal relative to ¹H MRI. Increasing the SNR by means of conventional averaging is very time-consuming especially when a series of images is required. Such image series are for example acquired in chemical exchange saturation transfer (CEST), where each image is encoded after applying a saturation pulse with a different frequency. In this study, we present a new approach to increase the SNR and/or accelerate the acquisition of CEST measurements with hyperpolarized ¹²⁹Xe.

Methods

Our proposed CAVKA method combines variable flip angle (VFA) excitations¹ with the “keyhole” technique and averaging. Keyhole is a view-sharing approach that includes the acquisition of two components: a fully sampled reference image and a series of undersampled images². Undersampling is achieved by encoding only a square in the center of k-space (called keyhole). Missing data in the periphery of the undersampled images is retrieved computationally from the reference image. In the proposed CAVKA approach, this reference image is acquired using averaging and the keyhole series is encoded with VFA. Merging the k-space data of the keyhole and reference images and the subsequent image reconstruction are done in self-written software in Python. All measurements were performed on a 9.4T pre-clinical MRI scanner (Bruker, Germany). Xenon hyperpolarization is achieved through spin exchange optical pumping³ using a home-built polarizer. All results shown come from a two-compartment phantom experiment, where the outer compartment is filled with H₂O and the inner one with a CEST agent solution (10 μM Cryptophane-A in H₂O + 0.2% DMSO). Hyperpolarized ¹²⁹Xe gas was bubbled directly into the solutions.

Results

Our approach increases SNR since VFA divides the available magnetization according to the chosen number of phase encoding steps. The lower this number, as in an undersampled image, the higher the magnetization per readout line. Additional de-noising is achieved, for the entire CEST image series, by the averaged reference image. Fig.1 shows three images: a conventional (non-hybrid) image (A), a hybrid image with VFA-enhanced center and non-averaged periphery (B) and the CAVKA image with VFA-enhanced center and averaged periphery (C). The transition from A to B shows the signal intensity increase due to VFA encoding and the transition from B to C shows the noise reduction due to averaging of the periphery. Importantly, the combination of both enables a ~4-fold SNR increase from A to C. Quantification of the SNR in the images that were composited according to our method is presented in Fig.2. The plot shows the SNR as a function of the number of averages of the reference image for three different undersampling factors of R = 1.33 (green), R = 2 (orange) and R = 4 (blue). In addition, a linear fit is included (dashed gray line) to show the theoretical SNR values when both, the periphery and the keyhole region, were averaged. For a fixed undersampling factor, the SNR increases for increasing number of averages. More important, the SNR further increases with increasing undersampling factors and almost reaches the level of the reference line for CAVKA-4 (i.e., R = 4). The outcome of applying the method to an image series is presented in Fig. 3. It shows the comparison between two z-spectra: one based on the proposed CAVKA method (blue) and the other based on fully sampled and averaged images (orange). It displays a very good agreement between the two spectra by their almost complete overlap. In particular, this match applies to the CEST contrast at -132 ppm. However, using CAVKA leads to a significant reduction (7-fold) in the acquisition time compared to the fully sampled and averaged series.

Discussion

The two components of CAVKA optimize the utilization of the available magnetization according to the number of phase encoding steps. In addition, the accelerated acquisition of this method is achieved by the reuse of averaged data from the reference image for the entire undersampled series. The CAVKA method is particularly useful to accelerate image acquisitions for setups that suffer from low SNR and demands for averaging. Some limitations apply for capturing dynamic changes of small areas along the image series where an excessive undersampling factor corresponds to a small keyhole and might not capture such dynamic changes that manifest in the periphery. However, in the presented CEST case, a keyhole size of just 4x4 was still enough to capture the CEST contrast between the two different compartments.

Conclusion

In this work, we have introduced a method that deals with low SNR in MRI series. Such conditions particularly occur for X-nuclei or limited magnetization from dilute hyperpolarized agents. Such scenarios often do not allow for long acquisition times. In addition, future in vivo applications of the HyperCEST method face reduced SNR compared to in vitro experiments where a large potential of this method has been demonstrated.

Acknowledgements

No acknowledgement found.