To collect ultrafast z-spectra in vivo at 7T where voxel homogeneity cannot be assured.

CEST contrast is generated by selective saturation of a labile proton group and measurement of the subsequent reduction in water signal.¹ Standard methods collect a single offset frequency per acquisition which leads to a trade-off between spectral resolution and scan time.

Saturating in the presence of a gradient spreads the offset frequency spatially across a voxel. Effectively, each voxel is split into slivers that are encoded for each offset frequency. This encoding is resolved by applying a similar gradient during readout and allows for ultrafast² collection of a full z-spectrum in a single shot and has been applied in vitro^3,4,5 and recently, localized in vivo in human brain (UCEPR)⁶. Spectral resolution is determined by the number of readout points and does not affect overall scan time. In conventional z-spectroscopy, inhomogeneity and partial volume effects are averaged over the entire voxel for each frequency, whereas in ultrafast z-spectroscopy (UFZS), different offsets experience different effects, resulting in unpredictable errors in analysis.

Here, we demonstrate an extension to UCEPR⁶ that makes it more robust to voxel inhomogeneity and partial volume effects by acquiring an additional scan with the saturating gradient polarity reversed. Reversing the gradient polarity reverses the direction of offset frequencies relative to isocenter. Physical locations that were saturated at a positive offset originally are saturated at a negative offset during reversal. We call this approach gradient-reversed ultrafast z-spectroscopy (GRUFZS).

The pulse sequence is a modified version of PRESS in which excitation was performed with a hard, nonselective 90^o pulse followed by three slice selective refocusing pulses and is shown in Fig 1. Saturation used Hanning-filtered rectangular pulses with a total saturation time of 800 ms with peak B1 power 220 Hz. The saturation pulse was frequency shifted such that the voxel center was on resonance and the saturation bandwidth was $$$\pm$$$5 ppm over the voxel. The readout was oversampled by a factor of 10 to ensure adequate sampling of T2*-weighted gradient echo. TR/TE = 8000/30 ms. Voxel size was 15x15x15 mm³. Scans without saturation were acquired and used to normalize for proton density differences across the voxel. All scans were taken on a Siemens 7T whole body scanner.

Time domain data was filtered using a binary threshold at the noise level since the readout was much longer than the actual echo duration. Data was Fourier transformed to the spatial/z-spectral domain and normalized with the unsaturated scan. Mixed z-spectra were created by taking the positive offset frequencies from the positive polarity scan and combining with the negative offset frequencies from the negative polarity scan and vice versa. $$$\text{MTR}_{asym}$$$ values were calculated and normalized by the negative offset intensity.

Figure 2 shows an example reconstruction for a cortical white matter voxel. The white matter voxel shown is fairly homogeneous with little partial voluming. Nevertheless, the asymmetry plots from the mixed z-spectra in Fig 2g are much closer than those from the acquired spectra in Fig 2f. For an inhomogeneous voxel with significant partial voluming, UFZS results in widely different, sometimes negative asymmetry values depending on the gradient polarity as shown in Fig 3f. The mixed asymmetry plots calculated from GRUFZS in Fig 3g show much better agreement, especially farther from water.

Table 1 shows results of different voxel placements as well as different ultrafast directions and compares to conventional z-spectroscopy. It is clear that regular UFZS fails in many of the cases since $$$\text{MTR}_{asym}$$$ can be positive or negative depending on the gradient polarity choice, while GRUFZS values are much closer and consistently reflect positive asymmetry values.

Conventional z-spectroscopy at the same digital spectral resolution as GRUFZS would take 25x longer, and though there is an SNR penalty associated with ultrafast scanning, the high sensitivity of CEST along with the large voxel sizes here mean that SNR is not a limiting factor. Compared to previously-reported UFZS, GRUFZS requires additional scans. However, there is no associated penalty in scanning efficiency since the mixed z-spectra can be averaged to give a single $$$\text{MTR}_{asym}$$$ value. Alternatively, the mixed z-spectra can be kept separate, as they are effectively acquired from two adjacent half voxels.

We have presented a method to acquire ultrafast z-spectra in vivo that requires only an additional scan with the gradient polarity reversed. The improved tolerance to inhomogeneity and partial voluming is evident in the asymmetry plots of the mixed z-spectra compared to the originally acquired ones. This method offers a fast, robust way to record full z-spectra in vivo.