Chemical exchange saturation transfer (CEST MRI) is a technique attracting growing interest in the molecular imaging arena, mainly due to its capability to investigate selectively in vivo metabolites (1). Variations in the static magnetic field B0 throughout the volume of interest and muscle deformations due to exercise can corrupt the measurement and must be accounted for. Currently, corrections are performed by mapping the B0 offset and shifting the CEST spectrum in the frequency domain accordingly (2). However, this method is not applicable in dynamic studies requiring high temporal resolution. Moreover, pixelwise analyses are particularly susceptible to motion artifacts. In muscle studies, motion can result from activity as well as from structural deformation due to muscular swelling after exercise, in a timescale comparable to a single saturation experiment. Here we propose a correction approach based on B0-CEST calibration applied to a longitudinal series of CEST scans during workout.

Exercise consisted of a regular repetition of high intensity plantar flexions for two minutes. Scanning at 3T was performed at rest and immediately after workout, allowing 15 minutes for recovery. Imaging protocol included dynamic CEST MRI, B0 and B1 mapping at the beginning and at the end of the series. Higher-order shimming was performed prior to the series acquisition. Static field inhomogeneities were assessed by WASSR (2). During CEST experiment, a 5 cm thick slab of tissue received a saturation pulse of 150 Hz for 500 ms. An offset of ±2 ppm from water frequency was chosen for the saturation, according to previous studies investigating Creatine under stress (3). A fast single-shot FLASH readout was ten applied, with an acquisition time for each CEST dataset being 12 sec, leading to a total series time of 30 minutes.

All images were first registered to match the anatomy of the model, consisting of the average of three baseline scans. A rigid transformation was applied to correct for high angular movement. Then, an algorithm based on demon transformation was used to map velocity gradient in both static and deforming images, determine displacement vectors pixelwise and deform the moving images accordingly (4). B1 correction was not required since no changes was found between map at beginning and end of the experiment. CEST signal at baseline in morphologically homogenous regions was plotted against B0 offsetsand a calibration curve obtained from the linear regression between pixel CEST contrasts and B0 offsets. All images were then B0-calibrated according to such curve. CEST asymmetry, defined as the ratio of signal with saturation at the frequency offset to the signal at symmetrical offset with respect to water frequency, was computed and plotted over time for different muscles.

Experiment was repeated with the conventionally used approach to correct for field inhomogeneities. The method consists in Z-spectra acquisition over a ±0.75ppm range around creatine saturation frequency and shifting the spectra in the frequency domain according to WASSR offsets map.

In order to assess the effectiveness of registration, Dice coefficient of similarity was computed for each image-model pair before and after transformation. Images difference was reduced and good similarity was achieved (average Dice 0.98 from 0.94). Baseline CEST asymmetry map before calibration showed a marked dependence on B0 value, which was minimized after correction (Pearson’s linear correlation coefficient r=0.7 before and r=0.3 after registration) and led to homogeneous signal within muscle tissue, as expected at rest condition and confirmed by comparison with standard correction data.

Our results show a steep increase in CEST asymmetry right after exercise, followed by an exponential decay during recovery. Average baseline and peak values in uncorrected map are 6% and 15% respectively, whereas after correction the values are rescaled to 0% and 12%, which are in accord with published data (3).

Our results confirm the observations reported in literature of a detectable metabolic change within muscles undergoing stress or activity. Moreover, we have demonstrated a simple yet efficient tool to correct for motion and B0 artifacts in dynamic CEST series, without the need for additional acquisitions or loss in temporal resolution and coverage. With further confirmation of its effectiveness, this technique can become a useful tool for in vivo study of muscle bioenergetics using CEST at a clinical B0-field strength.