Free-breathing 3D CEST MRI of Human Liver at 3.0 T

Abstract

Purpose:

To develop a novel 3D abdominal CEST MRI technique at 3T using MR Multitasking, which enables entire-liver coverage with free-breathing acquisition.

Methods:

K-space data were continuously acquired with repetitive steady-state CEST (ss-CEST) modules. The stack-of-stars acquisition pattern was used for k-space sampling. MR Multitasking was used to reconstruct motion-resolved 3D CEST images of 53 frequency offsets with entire-liver coverage and 2.0 x 2.0 x 6.0 mm³ spatial resolution. The total scan time was 9 min. The sensitivity of APT-CEST (MTR_asym at 3.5 ppm) and glycoCEST (mean MTR_asym around 1.0 ppm) signals generated with the proposed method was tested with fasting experiments.

Results:

Both APT-CEST and glycoCEST signals showed high sensitivity between post-fasting and post-meal acquisitions. APT-CEST and glycoCEST MTRasym signals from post-mean scans were significantly increased (APT-CEST: −0.019 ± 0.017 in post-fasting scans, 0.014 ± 0.021 in post-meal scans, p < 0.01; glycoCEST: 0.003 ± 0.009 in post-fasting scans, 0.027 ± 0.021 in post-meal scans, p < 0.01).

Conclusion:

The proposed 3D abdominal steady-state CEST method using MR Multitasking can generate CEST images of the entire liver during free breathing.

1. Introduction

Chemical exchange saturation transfer (CEST) imaging is a novel MRI technique that allows indirect detection of exchangeable protons in the water pool. In recent years, several studies have explored various CEST applications in the liver, such as amide proton transfer weighted (APTw) imaging [1, 2, 3, 4], glycogen CEST (glycoCEST) imaging [1, 2], glycogen nuclear Overhauser enhancement (glycoNOE) imaging [5], and extracellular pH measurement with Ioversol injection [6, 7], to list a few examples. However, CEST MRI of the liver in human studies is still challenging. Breath-holding is currently needed in abdominal CEST imaging to reduce motion artifacts, which limits its applications. First, successive breath-holds can cause discomfort of the patients. Some severely ill patients may not be able to perform voluntary breath-holding. Second, it limits the spatial resolution and volume coverage or the scan. 3D coverage is almost impossible with current techniques: even with multiple breath holds, a 3D scan would require an impractically long scan time. Ideally, abdominal CEST MRI can be performed with (a) continuous free-breathing acquisition, (b) 3D coverage, and (c) clinically acceptable scan time. These requirements call for improvements in sequence design and image reconstruction methods.

Steady-state CEST (ss-CEST) is a method that performs pre-saturation and k-space sampling in an interleaved pattern with repeated modules [8, 9]. Recently, we developed a ss-CEST method in the brain using MR Multitasking to accelerate image acquisition and enhance image quality [10, 11]. In this work, we propose a novel respiration-resolved 3D abdominal Multitasking ss-CEST technique at 3.0 T, which enables whole-liver coverage with free-breathing acquisition. The feasibility of the proposed technique was tested in healthy volunteers. The sensitivity of APT-CEST and glycoCEST measurements between post-overnight-fasting and post-meal imaging were evaluated with fasting experiments.

2. Methods

2.1. Sequence design

The continuous-acquisition pulse sequence consists of repetitive ss-CEST modules. Each ss-CEST module contains a single-lobe Gaussian saturation pulse, followed by a spoiler gradient and eight fast low-angle shot (FLASH) readouts [10]. FLASH readout with the option ‘water excitation only’ was used for fat suppression. The total time of one ss-CEST module is 72 ms, consisting of 30 ms for the saturation pulse and 42 ms for right FLASH readouts. The module repeated several times at each frequency offset until the steady state is reached and maintained, and then switched to another frequency without any additional delay between modules.

Images were acquired in axial orientation covering the whole liver region. K-space lines were sampled using a stack-of-stars acquisition pattern, as shown in Supporting Information Figure S1. In each ss-CEST module, the center k-space line was first sampled as “training data” to estimate temporal basis functions, and seven k-space lines with golden angle ordering in-plane (radial) and Gaussian-density randomized ordering in the partition direction (Cartesian) were sampled as “imaging data” to recover spatial basis functions. In this work, the training data was acquired in the superior-inferior direction (k_x = k_y = 0) to detect respiratory motion with more sensitivity.

2.2. In-vivo experiments

The experiment was approved by the institutional review board of Cedars-Sinai Medical Center. Written informed consent was obtained for all participating subjects. Data were acquired in eight (n = 8) healthy volunteers (five males + three females, mean age = 36.4 y, age range 21–59 y). To test the sensitivity of the proposed CEST imaging protocol to fasting and compared with previous methods [2, 3, 5], two separate scans were done for each volunteer: a ‘post-meal’ scan which was done 1.5 hours after a full lunch, and a ‘post-fasting’ scan which was done after overnight (more than 12 hours) fasting.

MRI data were acquired on a Siemens 3T MR system (MAGNETOM Vida, Siemens Healthcare, Erlangen, Germany) with an 18-channel phase array body coil. Imaging parameters were: FOV = 512 x 512 x 192 mm³, matrix size = 256 x 256 x 32, spatial resolution = 2.0 x 2.0 x 6.0 mm³. CEST parameters were: TR = 72 ms, saturation pulse duration t_sat = 30 ms, saturation pulse flip angle = 500° (effective B₁ = 0.93 μT). For FLASH readouts, echo spacing = 4.6 ms, TE = 2.1 ms, flip angle = 5°, water-excitation for fat suppression. The module was repeated 128 times at each frequency offset (9.2 sec in total), and then switched to another frequency without delay. Data were acquired at 53 frequency offsets (−40.0, −30.0, −20.0, −15.0, −10.0, −9.0, −8.0, −7.0, −6.0, −5.5, −5.0, −4.5, −4.0, −3.5, −3.0, −2.5, −2.0, −1.7, −1.4, −1.2, −1.0, −0.8, −0.6, −0.4, −0.2, −0.1, 0, 0.1, 0.2, 0.4, 0.6, 0.8, 1.0, 1.2, 1.4, 1.7, 2.0, 2.5, 3.0, 3.5, 4.0, 4.5, 5.0, 5.5, 6.0, 7.0, 8.0, 9.0, 10.0, 15.0, 20.0, 30.0, 40.0 ppm) from upfield to downfield, with prolonged unsaturated acquisition S₀ (300 ppm) at the beginning and the end. The total imaging time was 9 min.

2.3. Image reconstruction

The abdominal CEST image is modeled as a 6D image a(x, z, τ, r), with voxel location x, pre-saturation frequency offset z, time within each frequency offset τ (indexing the approach to steady-state), and respiratory motion phase r. It can be modeled as a 4-way low-rank tensor 𝒜, which is similar to our previous work [10] but adding a respiratory dimension:

$$𝒜=𝒢{\times }_1{\mathbf{U}}_{\mathbf{x}}{\times }_2{\mathbf{U}}_z{\times }_3{\mathbf{U}}_{\tau }{\times }_4{\mathbf{U}}_r$$ (1)

where columns of U_x, U_z, U_τ, U_r contain basis functions for space, frequency offset, approach to steady-state, and respiratory motion respectively. 𝒢 denotes the core tensor.

Components of the temporal factor tensor Φ = 𝒢 ×₂ U_z ×₃ U_τ ×₄ U_r were estimated from the training data d_tr with tensor subspace estimation. The training data d_tr can be reshaped into a four-way multichannel tensor 𝒟_tr in (k, z, τ, r)-space, where the first mode comprises k-space locations k from all receiving coils, z and τ represent the temporal indices which can be read naturally from chronological order, and r indicates the respiratory motion state. In this work, respiratory motion was binned into 5 states with the help of recorded Siemens Physiologic Monitoring Unit (PMU) data, which was extracted from the raw data.

A dictionary of full-course signal curves was generated with Bloch simulation, consisting of 41 T1 values of the water pool logarithmically spaced from 100 ms to 3000 ms and 18 exchange rates (k_ex) of the APT pool at 3.5 ppm linearly spaced from 100 ms to 3000 ms. The temporal factor U_τ which characterizes the signal evolution to reach steady-state was predetermined from the singular value decomposition (SVD) of this dictionary.

The undersampled tensor 𝒟_tr was first completed by solving:

$${\widehat{𝒟}}_{\text{tr}}=\text{arg}\underset{{\mathbf{D}}_{\text{tr},(2)}\in \text{span}\{{\mathbf{U}}_z\}}{\text{min}}\Vert {\mathbf{d}}_{\text{tr}}-{\Omega }_{\text{tr}}({𝒟}_{\text{tr}}){\Vert }_2^2+\lambda \sum _{i=1,3,4}{\Vert {\mathbf{D}}_{\text{tr},(i)}\Vert }_{\ast }$$ (2)

where d_tr is the sampled training data, D_tr,(i) is the mode-i matricization of the tensor 𝒟_tr, ∥·∥_* denotes the nuclear norm, λ is the weighting parameter, and Ω_tr indicates that only the 𝒟_tr with sampled index combinations are considered in the data fidelity term.

With the ${\widehat{𝒟}}_{\text{tr}}$ completed, the core tensor 𝒢, respiratory basis functions U_r and Z-spectra characterization basis functions U_z can be extracted from the higher-order SVD [12] of 𝒟_tr, which finally recovers Φ. Then, spatial basis functions U_x were determined from the imaging data d_im by solving the wavelet-regularized least-squares fitting problem [11].

2.4. CEST analysis

A 4D image $\tilde{a}(\mathbf{x},z)=a\left(\mathbf{x},z,{\tau }_{\max },r_{\text{opt}}\right)$ was extracted for CEST analysis, where τ_max is the last sampling time point (or the average of all time points when the steady state is reached) at each frequency offset, and r_opt is the optimal respiratory phase.

Voxel-wise B₀ correction was performed using the central part of the Z-spectrum (nominal frequency offset ∣Δω∣ < 1ppm) with a Lorentzian model [13]. The CEST effect was then measured with the magnetization transfer asymmetry ratio (MTR_asym):

$${\text{MTR}}_{\text{asym}}(\Delta \omega )=Z(-\Delta \omega )-Z(\Delta \omega )$$ (3)

where Δω is the frequency offset. APT-CEST values were quantified as MTR_asym at Δω = 3.5 ppm, while glycoCEST values were quantified as the average MTR_asym of 0.6, 0.8, 1.0, 1.2, and 1.4 ppm.

2.5. Image analysis

All image reconstruction and image processing were performed with MATLAB R2018a (MathWorks, Natick, Massachusetts, USA) on a Linux workstation with two 2.7-GHz 12-core Intel Xeon CPUs, one NVIDIA Quadro K6000 GPU, and 256 GB RAM.

Regions of interest (ROIs) were manually drawn to include the liver region in the central slice of the imaging volume. The boundary of the liver was excluded from the ROIs to avoid severe B₀ inhomogeneities. For each scan, single APT-CEST and glycoCEST MTR_asym values were reported as mean ± standard deviation (SD) within the ROI.

Wilcoxon signed rank test was performed to compare mean APT-CEST and glycoCEST MTR_asym values in post-fasting and post-meal scans. A two-tailed value of p < 0.05 was considered to be statistically significant. The repeatability was tested in all eight volunteers for both post-fasting and post-meal scans, and evaluated by performing the Bland-Altman plot. Statistical graphs were generated using GraphPad Prism 8 (GraphPad Software, La Jolla, California, USA).

3. Results

The proposed protocol with free-breathing acquisition was successfully applied to all healthy volunteers. Videos showing CEST image series along the whole Z-spectrum can be found in Supporting Information Video S1 (axial view) and Video S2 (coronal view). It can be seen that the position of the liver is consistent in images of different frequency offsets within a specific respiratory bin.

Figure 1 shows representative Z-spectra from a healthy volunteer within the ROI for both post-fasting scan and post-meal scan after B₀ correction.

Figure 1:

Representative B₀ corrected Z-spectra within the ROI for (a) post-fasting scan and (b) post-meal scan of Volunteer 4.

Figure 2 presents B_0, APT-CEST (MTR_asym at 3.5 ppm), and glycoCEST (MTR_asym at 1.0 ppm) maps in post-fasting and post-meal scans of another volunteer. Figure 3 shows the distribution of APT-CEST and glycoCEST MTR_asym values within the liver region in the same volunteer.

Figure 2:

Representative B₀, APT-CEST and glycoCEST maps of the liver region at the central slice from (a-c) post-fasting scan and (d-f) post-meal scan of Volunteer 1.

Figure 3:

Histograms of MTR_asym values of APT-CEST and glycoCEST from the same volunteer shown in Figure 2.

Results of average APT-CEST and glycoCEST measurements from post-fasting and post-meal scans in all subjects are shown in Figure 4a and Figure 4b respectively. The upward trends of both APT-CEST and glycoCEST MTR_asym values after meal can be clearly seen in the figure, which is consistent with previous study [2]. In all volunteers, both APT-CEST and glycoCEST MTR_asym signals from post-mean scans were significantly increased (APT-CEST: −0.019 ± 0.017 in post-fasting scans, 0.014 ± 0.021 in post-meal scans, p < 0.01; glycoCEST: 0.003 ± 0.009 in post-fasting scans, 0.027 ± 0.021 in post-meal scans, p < 0.01).

Figure 4:

Line plot of mean MTR_asym values of APT-CEST and glycoCEST in the liver region from post-fasting scans and post-meal scans.

The Bland-Altman plots of APT-CEST and glycoCEST measurements from two intra-session scans are shown in Figure 5a and Figure 5b respectively. The 95% limits of agreement of the average APT-CEST signal were −0.055 to 0.080, while the 95% limits of agreement of the average glycoCEST signal were −0.037 to 0.044.

Figure 5:

Bland-Altman plots for mean MTR_asym values of APT-CEST and glycoCEST in two intra-session measurements. The dashed lines show 95% limits of agreement.

4. Discussion

The Multitasking ss-CEST technique was able to acquire 3D liver CEST images without breath-holds. The in-plane spatial resolution of 2.0 mm was comparable with previous 2D studies [2, 3, 4], while the slice coverage of 192 mm was much larger, covering the whole liver.

In this work, k-space data were continuously acquired during free-breathing within the 9-min scan, and all the data were used for image reconstruction. Compared with breath-hold or navigator-gated acquisition, the proposed free-breathing method not only makes it possible to apply 3D abdominal CEST imaging in human studies for the first time, but also greatly improves the acquisition efficiency (equivalent to an 100% gating efficiency). In free-breathing acquisition, data need to be binned into different groups corresponding to real-time respiratory motion states, so that the tensor 𝒟_tr can be correctly completed in the respiratory dimension. Respiratory binning is usually done only using image data (similar to self-gating strategies) in previous MR Multitasking applications, either directly in the image domain [14] or from real-time temporal weighting functions. However, it is difficult to apply such binning methods in ss-CEST acquisition, because both real-time images and temporal weighting functions show very low signal intensities when the saturation frequency offset is close to 0 ppm. Therefore, the Siemens PMU data was used in our proposed protocol as auxiliary data to finish the respiratory binning process.

In this study, asymmetric analysis was used to performed for CEST quantification. Although MTR_asym is easy to calculate, mixed contrast is usually produced in MTR_asym with contributions of different chemical component sources, reflecting CEST effect, NOE effect, or even MT effect. For instance, glycoCEST signal (MTR_asym at 1.0 ppm) can be contaminated with the glycogen NOE signal at around −1.0 ppm [5]. APT-CEST signal (MTR_asym at 3.5 ppm) is contaminated with the lipid NOE signal from around −3.5 ppm, though can possibly be reduced with fat suppression [2]. Such mixed contrast makes it difficult to explain the origin of the signal variation in post-fasting and post-meal scans. Advanced CEST quantification methods, such as multi-pool fitting or the Lorentzian difference method [15, 16, 17], may help to separate different sources of contribution. Nevertheless, a robust Lorentzian fitting requires careful selection of the number of pools, dominant components of each pool, as well as initial values and bounds of the fitting parameters. To our knowledge, very few research have been done in in-vivo human liver CEST imaging, and there is no consensus or recommended parameters for multi-pool fitting in human liver. Therefore, we were still using the simple asymmetric analysis in this work. It is an important and interesting future work to develop appropriate and robust Z-spectrum fitting methods for abdominal CEST imaging. MTR_asym may also be contaminated by T1 changes of the water pool [18], especially with relatively low saturation powers. We would also like to explore how T1 changes will affect in-vivo abdominal CEST signals in our future studies, and perform T1 normalization if necessary.

Motion handling is still not perfect in the current free-breathing ss-CEST protocol. First, the B₀ field is constantly changing during free-breathing acquisition, which can cause severe B₀ inhomogeneities, especially in the regions close to the liver dome. This will introduce error to final CEST analysis because of incomplete B₀ correction or incorrect modeling of B₀ inhomogeneities in the low-rank model. It can possibly explain those outliers shown in Figure 5. Advanced active shimming coils may potentially provide better B₀ shimming in future studies [19, 20]. Second, residual motion is still visible in some cases after respiratory binning, which can also introduce error to pixel-wise CEST analysis. This may be addressed in the future by developing more robust binning procedures. Last, the steady-state signal may be less stable, because the tissues at the boundary of the imaging volume are continuously moving in and out. Signal averaging can be used to replace the single time point τ = τ_max to lessen this issue.

There are several limitations in this study. First, B₁ maps were not acquired in this study to perform B₁ correction. It can cause increased variabilities of CEST quantification, especially in moving organs. This issue will be investigated and improved in our future work. Second, as an initial technical validation, only healthy volunteers were included in this study, and some variates were not strictly controlled in the fasting experiments (such as diet, actual duration of fasting, etc.). Even though clear upward trends can be seen in both APT-CEST and glycoCEST signals after meal for each subject, it is hard to compare among different subjects. Therefore, we are not able to reach clinical conclusions at this stage from the results of current fasting experiments. Also, a single ROI was drawn in the liver region to report the average APT-CEST and glycoCEST for each volunteer. Analysis was not performed in different sub-regions of the liver to explore possible signal difference of various liver tissues. A more systematic patient study will be performed in the future to investigate potential clinical applications of the proposed protocol.

Future work will also focus on optimizing the saturation power B₁ and developing application-specific frequency offset sampling patterns with clinical validations. Further reduction of total scan time may be achieved by reducing number of frequency offsets and by reducing k-space data sampled within each frequency offset. Improvement of the spatial resolution will be explored as well.

5. Conclusion

The proposed 3D abdominal steady-state CEST method using MR Multitasking can generate CEST images of the entire liver during free breathing. It has the potential to push the liver CEST technique towards practical clinical use.

Supplementary Material

SUPINFO1

Figure S1: K-space sampling pattern. The k-space is continuously sampled using a stack-of-stars FLASH sequence with golden angle ordering in the x-y plane and Gaussian-density randomized ordering in the z direction, interleaved with training data (central k-space line along z-direction) every 8th readout.

SUPINFO2

Video S1: Representative CEST image series along the whole Z-spectrum (axial view).

SUPINFO4

Video S2: Representative CEST image series along the whole Z-spectrum (coronal view).