Phase-locked CEST – Introducing dynamic B0-correction to gagCEST
Johannes Windschuh1, Moritz Zaiss2, Jae-Seung Lee1,3, Alexej Jerschow3, and Ravinder Regatte1

1Center of Biomedical Imaging, New York University Langone Health, New York, NY, United States, 2Magnetic Resonance Center, Max Planck Institute for Biological Cybernetics, Thuebingen, Germany, 3Department of Chemistry, New York University, New York, NY, United States


Even a small frequency drift of less than 1Hz/min of the MRI scanner can have a strong impact on gagCEST measurements. We propose a dynamic B0-correction that tracks the frequency shift using the phase images provided by the GRE readout. We show that this correction eliminates the influence of the frequency drift on gagCEST without the need of additional measurement time allowing higher accuracy, reproducibility, and comparability of gagCEST studies.


Chemical Exchange Saturation Transfer (CEST) enables the detection of glycosaminoglycan (GAG) present in human articular cartilage1, which represent an early marker of osteoarthritis. For reliable gagCEST analysis exact B0 mapping and correction are essential2. In this study we investigates the influence of a frequency drift often observed in a clinical MRI scanners3,4 on gagCEST. Furthermore, we introduce a dynamic B0-correction method called phase-locked CEST, which utilizes phase maps extracted from the GRE readout.


Measurements were performed in model solution (20 mM D-Glucose, 3.4% Agar, 9 uM Gadoteric acid) – experiment I – and in vivo in the knee of a 32-year-old healthy volunteer – experiment II. All measurement were performed on a 7T whole body scanner (MAGNETOM, Siemens, Germany) using a 28-channel knee coil (QED, USA). For precise B0 mapping four WASABI5 measurements were interleaved with three identical gagCEST measurements. The gagCEST parameters are: 5 Gaussian-shaped pulses with tp$$$~$$$=$$$~$$$100$$$~$$$ms, B1$$$~$$$=$$$~$$$1.5$$$~$$$μT, DC$$$~$$$=$$$~$$$99%, 41 equally-distributed frequency offsets from -2$$$~$$$ppm$$$~$$$to$$$~$$$2$$$~$$$ppm (Msat) and 6 far off-resonant offsets at -300 ppm (M0) for improved normalization6. Image acquisition was performed using 3D$$$~$$$GRE$$$~$$$SnapCEST7 with TR$$$~$$$=$$$~$$$4.2$$$~$$$ms, TE$$$~$$$=$$$~$$$2.3$$$~$$$ms, FOV$$$~$$$150$$$~$$$x$$$~$$$140$$$~$$$x$$$~$$$36$$$~$$$mm3, resolution 0.85$$$~$$$x$$$~$$$0.85$$$~$$$x$$$~$$$3$$$~$$$mm3. Evaluation was performed in ROIs shown in Figure 1.

For phase-locked CEST the relative B0 shift is calculated pixel-by-pixel as $$$\delta B_0(t) [ppm] = \frac{\theta(t)-\theta(t_0)}{TE\cdot\omega_0}$$$ from phase maps. Only phase maps of M0 acquisitions were used and two subsequent acquisitions were averaged to reduce noise. After linear extrapolation the relative B0 shift δB0(t) was added to the absolute B0 map from the first WASABI acquisition to create the absolute B0 shift ΔB0(t).

Three different B0-correction methods were compared for the two experiments:

$$$~~~~~$$$1) Time-independent correction using only the B0 map of the first WASABI measurement.

$$$~~~~~$$$2) Time-dependent correction using linear interpolation of all the B0 maps of the WASABI measurements.

$$$~~~~~$$$3) Time-dependent correction using phase-locked CEST, as described above.


Figure 2 shows representative Z-spectra of articular cartilage and their corresponding MTRasym curves and as a function of a simulated, artificial B0 shift, which changes the gagCEST signal at MTRasym(1.0 ppm) by 53%/ppm (0.18%/Hz).

In Figure 3 ΔB0(t) determined by WASABI acquisitions is compared to ΔB0(t) from the new phase-locked CEST approach in experiments I & II. Both methods show very good agreement. The variation of ΔB0 during Msat acquisitions look similar to the direct water saturation in a Z-spectrum, making most Msat phase images unusable for B0-correction. In experiment II ΔB0(t) increased linearly with approx. 0.003 ppm/min (0.9 Hz/min). In experiment I ΔB0(t) decreased with -0.002 ppm/min (-0.6 Hz/min).

Figure 4 shows the comparison of the B0-correction methods 1-3 in experiments I & II. MTRasym curves of method 1 show strong time-dependence (Fig. 4a,d), whereas MTRasym curves of methods 2 and 3 both reduce this variation dramatically. Even the gagCEST value of the first measurement obtained with method 1 deviates about 1% from the gagCEST value obtained with methods 2 and 3 (dotted line).


The steep slope of the direct water saturation in the Z-spectrum around Δω = 1.0 ppm is responsible for the observed high sensitivity on ΔB0. While the small B0 drift measured here is negligible for APT and NOE around $$$\pm$$$3.5 ppm for CEST effects closer to the water resonance accuracy and reproducibility can be improved if these drifts can be measured and corrected for.

The linear drift observed in experiment II is consistent with linear drifts of 0.003$$$~$$$-$$$~$$$0.005 ppm/min observed in previous experiments on the same scanner. The negative non-linear drift in experiment I may be due to the scanner cooling down from a preceding gradient intensive fMRI experiment.

The origin of the strong phase variation as a function of saturation offset in Msat acquisitions could not be identified yet. Increasing the spoiling time before GRE readout resulted in no change (data not shown).

Using only the M0 acquisitions can be understood as a virtual re-adjustment of the center frequency, thus a real re-adjustment of the center frequency of the scanner before each measurement may also reduce overall ΔB0(t). However, it does not reduce ΔB0(t) within the gagCEST acquisition, which is not negligible as can be seen in Figure 4. Additionally, it is not sure that the scanner adjusts the frequency in the same way as for the B0 map acquisition, introducing further uncertainty to the experiment.


Phase-locked CEST is a simple and robust dynamic B0-correction method allowing higher accuracy, reproducibility, and comparability of gagCEST studies by eliminating errors due to frequency shifts of the MRI scanner. This might be especially relevant for long measurements as well as measurements after gradient-intensive imaging.


This work was supported in part by the National Institutes of Health (grants R01EB016045, R01AR067156 and R01AR068966).


1. Ling W, Regatte R R, Navon G, et al. Proc. Nat. Acad. Sci. USA 2005;105(7):2266-2270

2. Krishnamoorthy G, Nanga Reddy R P, Bagga P, et al. High Quality Three-Dimensional gagCEST Imaging of In Vivo Human Knee Cartilage at 7 Tesla. MRM 2017;7(5):1866-1873

3. Smith A M, Lewis B K, Ruttimann U E, et al. Investigation of Low Frequency Drift in fMRI Signal. NeuroImage 1999;9(5):526-533

4. De Poorter J, De Wagter C, De Deene Y, et al. Noninvasive MRI Thermometry with the Proton Resonance Frequency (PRF) Method: In Vivo Results in Human Muscle. MRM 1995;33(1):74-81

5. Schuenke P, Windschuh J, Roeloffs V, et al. Simultaneous Mapping of Water Shift and B1 (WASABI) - Application to Field-Inhomogeneity Correction of CEST MRI Data. MRM 2017;77(2):571-580

6. Jones C K, Huang A, Xu J, Edden R AE, et al. Nuclear Overhauser enhancement (NOE) imaging in the human brain at 7 T. NeuroImage 2013;77:114-124

7. Zaiss M, Ehses P, Scheffler K, et al. snapCEST - A single-shot 3D CEST sequence for motion corrected CEST MRI. Proc. ISMRM 2017;25:3768


Figure 1: M0 image of experiment I in model solution with ROI in magenta (a), M0 image of experiment II in vivo in human knee of a healthy volunteer with ROI in magenta (b).

Figure 2: Simulation of the influence of a B0 shift ΔB0 on ROI-averaged in vivo gagCEST data acquired in articular cartilage at 7T. (a) Z-spectra with artificial B0 shifts from -0.05 to +0.05 ppm (blue to red) in steps of 0.0025 ppm. (b) MTRasym curves of the shifted Z-spectra with the black line indicating data with no shift. (c) MTRasym evaluated at 1.0 ppm as a function of B0 shift. The black line represents a linear fit with a slope of 53%/ppm.

Figure 3: ΔB0(t) of the B0 maps from WASABI acquisitions (red), phase maps of M0 acquisitions (green), and from phase maps of Msat acquisitions (blue) in ROIs of model solution (a) and in vivo in articular cartilage(b). Since phase maps provide only relative δB0(t) the curves are shifted with an offset calculated by extrapolation to t = 0, the time of the first WASABI acquisition and are then added to the B0 map of this WASABI acquisition to generate ΔB0(t).

Figure 4: MTRasym after B0-correction with one B0 map at time t = 0 (a,d), with interpolated B0 maps from WASABI measurements (b,e), and with the phase-locked CEST approach (c,f) in ROIs of model solution (a-c) and in vivo in articular cartilage (d-f) at 7T.

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)