Multicolor metabolic quantitative CEST (mmqCEST): high resolution imaging of brain metabolites
Vitaliy Khlebnikov1, Alex Bhogal1, Mark Schuppert2, Moritz Zaiss2, Tobias Lindig3, Benjamin Bender3, Ulrike Ernemann3, Klaus Scheffler2,4, Peter Luijten1, Hans Hoogduin1, Dennis Klomp1, and Jeanine J Prompers1

1University Medical Center Utrecht, Utrecht, Netherlands, 2Max Planck Institute for Biological Cybernetics, Tübingen, Germany, 3Eberhard-Karls University, Tübingen, Germany, 4Eberhard-Karls University Tübingen, Tübingen, Germany


Multicolor metabolic quantitative CEST (mmqCEST): high resolution imaging of brain metabolites


The relatively low spatial resolution (ca. 250 mm3) of 1H-MRSI hinders its application towards probing heterogeneous diseased tissue, e.g. tumor tissue. We hypothesize that advanced calibration models between CEST and 1H-MRSI can be established, whereby quantitative metabolic maps can be extracted from Z-spectra. In this work, we develop mmqCEST, a metabolic imaging technique based on the saturation transfer from RF-tagged metabolites to the bulk water [1]. mmqCEST decodes a frequency-dependent metabolite-weighted contrast in the Z-spectra into (multicolor) quantitative metabolic maps with a high spatial resolution (ca. 3.4 mm3).


An informed consent was obtained from all participants in this study.

1H-MRSI at 7T

Steady-state free precession (SSFP) pulse acquire MRSI data was acquired on a 7T Philips MRI using the following parameters: FA/TE/TR=35°/2.5/300 ms, FOV 220x220mm2, acquisition matrix 44x44, resolution 5x5x10 mm3, BW 3000 Hz, NSA 2, scan duration 10m59s, lipid suppression using external crusher coil, tailored spiral in-out spectral-spatial water suppression pulses, two slices acquired at the level of the ventricles and basal ganglia, respectively, in 23 healthy subjects. LC-model was used for data quantification.

CEST at 7T

3D CEST experiments (8 healthy subjects) were done on a 7T Philips MRI platform by modifying the sequence in [2] to a single shot acquisition and using parallel RF transmission to achieve a duty cycle for the presaturation module of 100% [3,4]. A total of three acquisitions with different presaturation schemes (block pulse shape) were performed: (1) Tsat=1.4s and B1=1.48µT; (2) T1sat=1.4s and B1=2.12µT; and (3) Tsat=0.5s and B1=3.6µT. (1) and (2) were shown to provide a high sensitivity to myo-inositol, creatine, phosphocreatine, and mobile (poly)peptides, whereas (3) provides a high sensitivity to glutamate [5]. A FOV of 224x224x24mm3 with a voxel size of 2x2x2mm3 was used. CEST data were motion corrected.

CEST at 9.4T

3D CEST experiments were done on a 9.4T Siemens MRI platform in four healthy volunteers and one glioblastomas (GBM, grade IV) patient. A CEST prepulse of 4.5s was used at four B1 levels (0.6, 0,9, 1.2 and 1.6uT). A FOV of 220x180x32mm3 with a voxel size of 1.5x1.5x1.8mm3 was used. Other sequence details are in [3]. The patient study was approved by the local ethics committee. CEST data were motion corrected.

Data processing

To reduce artifacts in MRSI data, e.g. due to B1 inhomogeneity, all metabolic maps were generated as ratios over total Cr (tCr).Normalized Z-spectra were B0-corrected by searching for the minimum of spline-interpolated (to 1Hz) spectra and shifting them accordingly.All data (CEST and MRSI) were transferred to the standard MNI 152 space (1x1x1mm3) by a three-step co-registration algorithm.

Calibration of CEST to MRSI

A calibration model between 1H-MRSI (metabolic ratios) and CEST (Z-spectra frequency profile) was established in the MNI-space by means of neural networks (NN) using 1H-MRSI data at 7T from 23 healthy subjects, CEST data at 7T from 7 healthy subjects, and CEST data at 9.4T from 3 healthy subjects. B1-inhomogeneity correction of CEST data was built into the calibration model by providing B1-uncorrected data during the calibration step. The validity of this atlas-based B1 correction of CEST data was validated in both simulations and in vivo experiments.Validation of mmqCEST at both 7T and 9.4T field strengths was done by applying the calibration model to CEST data from 1 healthy subject (unseen during the calibration) and comparing the output to the 1H-MRSI metabolic ratios averaged across 23 healthy subjects.

Results and Discussion

The Pearson correlation coefficient (R) between 1H-MRSI and the corresponding mmqCEST metabolic maps for the validation dataset at 7T (unseen during the calibration) was above 0.95 (Fig. 1). Interesting to note that only two subjects was enough for building the calibration model (R≈0.95).Some of the mmqCEST metabolic maps at 7T and 9.4T are shown in Fig. 2 and Fig. 3, respectively. In healthy subjects, 1H-MRSI and mmqCEST demonstrate a similar contrast for all metabolites studied (Pearson correlation coefficient, R>0.95). In a GBM patient at 9.4T, mmqCEST revealed a hotspot in the tumor area (GPC&Cho/tCr and GSH/tCr), which is likely an aggressive, metabolically active part of the tumor.


The initial results of calibration CEST frequency profile to MRSI metabolic ratios by means of artificial neural networks are promising. We showed a high contrast similarity between the original MRSI and mmqCEST metabolic maps (Pearson correlation coefficient above 0.95). To our knowledge, this is the first report demonstrating quantitative metabolic mapping with CEST.


No acknowledgement found.


[1] van Zijl, P. C. M. & Yadav, N. N. Chemical exchange saturation transfer (CEST): what is in a name and what isn’t? Magn Reson Med 65, 927–948 (2011).

[2] Khlebnikov, V. et al. Comparison of pulsed three-dimensional CEST acquisition schemes at 7 tesla: steady state versus pseudosteady state. Magn Reson Med 77, 2280–2287 (2017).

[3] Hoogduin H et al. Semi continuous wave CEST with alternating sets of 4 transmit channels at 7T. MAGMA. 30, S1–S152 (2017)

[4] Keupp J et al. Parallel RF Transmission based MRI Technique for Highly Sensitive Detection of Amide Proton Transfer in the Human Brain at 3T. Proc Intl Soc Mag Reson Med. 19:710 (2011)

[5] Khlebnikov V. et al. Multicolor metabolic quantitative CEST (mmqCEST) imaging:possibility and limitations. Proc. Intl. Soc. Mag. Reson. Med. 26, 0416 (2018).

[6] Zaiss, M., Ehses, P. & Scheffler, K. Snapshot-CEST: Optimizing spiral-centric-reordered gradient echo acquisition for fast and robust 3D CEST MRI at 9.4 T. NMR in Biomedicine 31, e3879 (2018).


Fig. 1. The Pearson correlation coefficient between 1H-MRSI and the corresponding mmqCEST maps for the validation dataset at 7T as a function of the number of subjects used in building the calibration model. Shown are results for different CEST acquisitions as well as for the output averaged from all three acquisitions.

Fig. 2. Metabolic maps generated using 1H-MRSI (1st row, averaged over 23 healthy subjects) and mmqCEST at 7T (Tsat=1.4s and B1=2.12µT) for a healthy subject (2nd row, validation dataset). CSF was masked out. All metabolic maps are overlaid on a T1w MNI image.

Fig. 3. Metabolic maps generated using 1H-MRSI (1st row, averaged over 23 healthy subjects) and mmqCEST at 9.4T (Tsat=4.5s and B1=0.6µT) for a healthy subject (2nd row, validation dataset) and a GBM (grade IV) patient (3rd row). CSF was masked out. The metabolic maps in the healthy subjects and in the patient are overlaid on a T1w MNI and FLAIR images, respectively. The quality of the patient data is reduced because of motion. The white arrows show the metabolic hotspot in the tumor.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)