Ultrafast Z-spectroscopic Imaging in vivo at 3T using through-slice spectral encoding (TS-UFZ)

Abstract

Purpose:

Acquisition of high-resolution Z-spectra for CEST or MTC MRI requires excessive scan times. Ultrafast Z-spectroscopy (UFZ) has been proposed to address this, however, the quality of in vivo UFZ spectra has been insufficient. Here, we present a simple approach to improve this.

Theory and Methods:

UFZ imaging acquires full Z-spectra by encoding the spectral dimension spatially via a gradient applied concurrently with the RF saturation pulse. Different from previous implementations, both this saturation gradient and its readout were applied in the slice direction, resulting in a relatively uniform voxel composition. Phase-encoding was applied in both in-plane directions, allowing additional under-sampling and acceleration.

Results:

In phantoms, UFZ imaging with through-slice Z-spectral encoding (TS-UFZ) provided Z-spectra of salicylic acid and egg white in excellent agreement with conventional acquisition. In vivo brain Z-spectra were influenced by flow through the imaging slice which affected the Z-spectral baseline. Still, CEST signals could be quantified after baseline fitting and mapping the residual CEST signal. Amide proton transfer (APT) contrast intensities obtained by TS-UFZ were on the same order of magnitude as conventional CEST but with different contrast across slice which likely is a result of different tissue regions contributing.

Conclusion:

TS-UFZ approach improves signal stability and spectral uniformity over previous implementations and allows high spectral-resolution imaging of saturation transfer effects in the human brain at 3T. This implementation allows for further acceleration by reducing phase encoding steps and thus opens up the possibility of mapping dynamic CEST signals in vivo with a practical temporal resolution.

1. Introduction

Chemical exchange saturation transfer (CEST) MRI has generated great interest for molecular imaging applications since it can detect low concentration solute molecules via the water proton signal.^1–5 In CEST MRI, exchangeable solute protons are selectivity saturated using radiofrequency (RF) irradiation and after the exchange of these protons to the bulk water pool, a reduction of water signal is observed. Through continuous RF saturation and repeated chemical exchange, multiple saturation transfer events lead to an enhanced reduction of the bulk water signal thus CEST MRI is able to amplify signals from low concentration molecules by several orders of magnitude^1,6.

However, in vivo, multiple signal sources such as semi-solid magnetization transfer contrast (MTC)⁷, direct water saturation (DS)⁸, multiple types of exchangeable protons⁹, and relayed Nuclear Overhauser effect (rNOEs) of mobile macromolecules^1,10–14 complicate the interpretation of CEST images. CEST images are therefore often acquired as a function of saturation frequency offset to create voxel-based water saturation spectra containing all of these effects.¹⁵ These so-called Z-spectra, which, similar to MR spectroscopy, are frequency-referenced (but here to the water solvent at 0 ppm) aid in the discrimination of signal components and allows their assessment using spectral quantification methods. A larger number of frequency offsets results in higher spectral resolution but the sequential acquisition of a proportionally larger number of images leads to excessive scan times.

To accelerate the acquisition of full Z-spectra for the purpose of measuring MTC effects, Swanson proposed applying a constant magnetic field gradient concurrently with the saturation RF pulse to encode the frequency offsets spatially.¹⁶ Xu et al¹⁷ and Boutin et al¹⁸ also showed the utility of this approach to measure CEST and hyperpolarized agents, respectively, named ultrafast Z-spectroscopy (UFZ). These initial implementations demonstrated the feasibility of this approach in single uniform samples in vitro. Subsequent studies showed that the incorporation of slice selection can allow multiple sample tubes to be measured simultaneously^19–21. Although these approaches have been shown useful to accelerate the screening of a large number of samples in vitro, they are not yet easily applicable to map dynamic processes in vivo. More recently, several groups have implemented UFZ approaches in vivo using different strategies.²² Liu et al.²³ implemented the UFZ approach in a point-resolved spectroscopy (PRESS) sequence, making it available for in vivo single-voxel spectroscopy. Wilson et al.²⁴ further extended the UFZ method by using the localized PRESS sequence with reversed gradient for the acquisition of in vivo Z-spectroscopy. However, these previous in vivo implementations of UFZ could only obtain Z-spectra of a single voxel and not CEST maps across a slice.

Here, we improve on previous UFZ implementations by applying the saturation gradient and its imaging readout in the slice direction, while phase-encoding is applied in-plane in two directions. Application of the saturation gradient and its readout across the relatively narrow slice thickness, produces a Z-spectrum over a relatively uniform tissue volume. In addition, this approach allows further acceleration of the phase-encoding applied in-plane. This UFZ imaging with through-slice Z-spectral encoding (TS-UFZ) implementation was validated on a 3T clinical scanner using samples of salicylic acid and egg white. Finally, in vivo feasibility was shown by mapping amide proton transfer (APT) contrast in the human brain.

2. Methods

2.1. TS-UFZ Pulse Sequence

The TS-UFZ imaging technique employs a saturation gradient G_sat applied concurrently with a pseudo-continuous wave (CW) saturation pulse as well as a turbo spin-echo (TSE) readout, both in the slice direction; phase-encoding is applied in-plane (Figure 1). When the center of the slice coincides with G_sat isocenter (Δs = 0, whereΔs is the distance of the slice center to the G_sat isocenter), the presence of G_sat along the slice shifts the resonance frequency (Δω) of each spin pool depending on the gradient strength and the location (d) from G_sat center:

$$\Delta \omega \left(d\right)=\gamma G_{\mathit{\text{sat}}}d$$ [1]

where γ is the gyromagnetic ratio of the observed nuclei (γ ≈ 267.5 × 10⁶ rad/T for protons) and G_sat is the saturation gradient amplitude. Thus, an RF saturation pulse applied at the water resonance frequency (F0) will directly saturate water at the G_sat isocenter (d = 0). In this case, the frequency of the applied saturation pulse (Δω_sat) is equal to the center frequency of the Z-spectrum (Δω_c), i.e., Δω_sat (Δs = 0) = Δω_c = 0 ppm. For other values of d, the resonance frequency for each pool will shift in line with Eq. [1] thus the saturation pulse applied at F0 will saturate a different proton pool depending on d. Therefore, an entire z-spectrum can be encoded along the spatial direction of the saturation gradient. This spectral information is recovered by applying a readout gradient in the same direction as G_sat. The CEST spectral bandwidth, BW_sat, is thus determined by the saturation gradient strength over the slice thickness (L_s) in rad/s:

$$B{W}_{\mathit{\text{sat}}}=\gamma G_{\mathit{\text{sat}}}{L}_s$$ [2]

with the frequency offset range between [γG_sat(−L_s/2), γG_sat(L_s/2)]. The number of saturation offsets is decided by the number of datapoints in the slice direction (N), so the resolution of Z-spectrum is given by:

$$\delta {\omega }_{\mathit{\text{sat}}}=B{W}_{\mathit{\text{sat}}}/N$$ [3]

When applied through a slice, each offset corresponds to a sub-slice. Thus, for a 10 mm slice thickness with 100 offsets, signal for data points in the Z-spectrum is coming from 100 μm sub-slices at each offset. Practically, the slice is often off-centered from the G_sat isocenter (Δs ≠ 0), resulting the Z-spectrum will not be centered at slice. Thus, the frequency of the applied saturation pulse Δω_sat needs to be adjusted according to the shift to ensure the Z-spectrum is at the slice middle as well as Δω_c = 0 ppm:

$$\Delta {\omega }_{\mathit{\text{sat}}}\left(\Delta s\right)=\Delta {\omega }_{\text{c}}+\frac{\Delta s}{L_s}B{W}_{\mathit{\text{sat}}}$$ [4]

Also, it is not necessary for the center of the Z-spectrum to be placed in the middle of the slice, and in fact, shifting the Z-spectra (i.e., shifting Δω_c) may be useful for sampling select Z-spectral regions of interest whilst minimizing signal loss due to spectral dispersion.

Figure 1.

Principle of TS-UFZ imaging pulse sequence. A series of short saturation gradient pulses of strength G_sat is applied in the slice direction concurrently with the saturation pulses. The saturation gradient G_sat and turbo spin-echo sequence are applied in the slice direction and phase-encoding is applied in-plane. To compensate for possible droop on the gradient amplifier during the prolonged CEST irradiation period, a series of shorter gradient pulses (a total of 10 in this paper) were applied for spatial frequency encoding during preparation instead of a single long gradient pulse.

Following the acquisition of the saturation image, a reference scan is acquired with the saturation pulse center applied far off-resonance (e.g., Δω_c=200 ppm). This reference signal (S₀) at an offset where there is no saturation difference expected over the slice thickness is required to correct the Z-spectrum for proton density differences between the sub-slices, and to eliminate effects arising from gradient non-linearity and spatial irregularities of the object. By dividing saturated signal (S_sat) at each offset with the corresponding S₀ at each, the normalized Z-spectral information can be recovered.

2.2. MRI acquisition

TS-UFZ was validated in vitro and then tested in vivo on three healthy volunteers. All MRI scans were performed on a 3T Philips Elition RX system (Philips Healthcare), with a 32-channel phased array head coil used as signal receiver and a quadrature body coil used as radiofrequency transmitter (Philips Healthcare). Gradients of up to 95 mT/m and 220 T/m/s in all three directions were available.

The TS-UFZ pulse sequence, as shown in Figure 1, used a 3D spin-echo (SE) with multi-shot TSE readout and consisted of a saturation period of 2 s with a pseudo-continuous saturation pulse consisting of a train of 40 sinc-gaussian shape pulses, each 50 ms in duration and B_1,sat = 0.5 μT. Spectral encoding was achieved using a train of 10 gradient pulses (each 200 ms in duration) applied concurrently with the RF saturation pulse. The saturation period was followed by the TSE sequence. To allow for the magnetization recovery, a delay of 3s was included before each saturation module.

For comparison, conventional Z-spectral images using a 2D SE with single-shot TSE readout were acquired with the same saturation duration and B₁ as the TS-UFZ imaging method, but Z-spectral data was acquired by varying the frequency offsets of the saturation pulse. See below for detailed parameters.

2.2.1. In vitro experiments

Salicylic acid phantom

Salicylic acid contains exchangeable protons resonating at 9.3 ppm from water.²⁵ A 50 mM salicylic acid solution (Sigma Aldrich, St Louis, MO) was prepared in phosphate buffered saline (PBS) solution (pH of 7.5).

TS-UFZ imaging was performed with the following parameters: field of view (FOV) = 144 × 144 × 20 mm³, the slice thickness along the G_sat was L_s = 20 mm, the typical number of acquired datapoints in the slice direction (N) was 200, leading to a voxel size for the sub-slices of 8 × 8 × 0.1 mm³; G_sat = 4, 6, 8, 10 mT/m, the corresponding BW_sat values were 21,400, 32,100, 42,800, and 53,500 rad/s (i.e., 26.6, 39.9, 53.2, and 66.7 ppm, respectively). TSE factor = 40, TE/TR = 21/5,873 ms, excitation/refocusing pulse angle = 90°/120°. The sensitivity encoding (SENSE)²⁶ factors were set to 1 in both in-plane directions. Encoding was done with Δω_c= 0 ppm with adjusted Δω_sat (Eq. [4]), resulting the Z-spectrum was centered at the slice; S₀ was acquired with Δω_c = 200 ppm. The total scanning time including both S_sat and S₀ was 2 mins and 15 s.

Conventional CEST imaging with FOV = 144 × 144 mm², slice thickness = 20 mm, voxel size = 8 × 8 mm², TSE factor = 9, TE/TR = 4.9/6,000 ms, excitation/refocusing pulse angle = 90°/120°. SENSE factor was set to 2 in the phase-encoding direction. A total of 58 saturated images (S_sat) were acquired from −5 to 12 ppm, and followed by 5 unsaturated images (S₀) where the saturation pulse was applied at 100 ppm. The acquisition time for one scan was 6 mins and 24 s.

Egg white phantom

The experiments on egg white phantom had the same FOV as the salicylic acid phantom but used slice thicknesses (L_s) = 10 mm. TS-UFZ imaging was conducted for G_sat = 4, 6, 8, 10 mT/m, the corresponding BW_sat values were 10,700, 16,050, 21,400, and 26,750 rad/s (i.e., 13.3, 20.0, 26.6, and 33.3 ppm, respectively). The number of N was 100. The scan at each G_sat was repeated three times to assess signal reproducibility. SENSE factors were set to 2 and 1 in two in-plane directions, respectively. And the total scan time of acquiring both S_sat and S₀ was 53 s. For conventional CEST, a total of 54 saturated images (S_sat) were acquired from −5 to 5 ppm with 0.2 ppm step increment, as well as 7, 10, and 15 ppm, and 5 unsaturated images (S₀). The acquisition time was 6 mins and 6 s.

2.2.2. In vivo experiments

Human studies were approved by the Johns Hopkins Medicine Institutional Review Board and performed on three healthy subjects, after informed consent was obtained.

The in vivo TS-UFZ experiments were performed similar to egg white with modified FOV parameters in both low and high in-plane resolutions, FOV = 150 × 206 × 10 mm³, L_s = 10 mm. Low spatial resolution: voxel size = 8 × 8 × 0.104 mm³, N was 96, TE/TR = 21/5,873 ms, the total scan time including both S_sat and S₀ was 1 min and 28 s. High spatial resolution: voxel size = 5 × 5 × 0.12 mm³, N was 84, TE/TR = 19/5,792 ms, the total scan time for both S_sat and S₀ was 3 mins and 58 s. G_sat = 6 mT/m, and the corresponding BW_sat was 16,050 rad/s (i.e., 20.0 ppm).

Conventional CEST imaging was also performed with both low-resolution and high-resolution image acquisitions with FOV = 150 × 205 mm², slice thickness = 10 mm. Low-resolution: voxel size = 8 × 8 mm², TSE factor = 13. High-resolution: voxel size = 5 × 5 mm², TSE factor = 21. Both TE/TR = 4.9/6,000 ms. The saturation frequency list and total scan time were same as egg white phantom.

2.3. Data processing

Both TS-UFZ and conventional CEST images were downsampled to achieve the same in-plane matrix size (32×32 for phantom, 33×33 for in vivo low-resolution, and 60×60 for in vivo high-resolution). For TS-UFZ, the Z-spectrum was constructed by normalizing the S_sat spectrum with the S₀ spectrum. In contrast to the conventional WASSR-based B₀ correction methods which shift the saturation spectrum, B₀ field inhomogeneities were corrected for the position of the direct water saturation peak, where the nominal saturation frequency list for each voxel (calculated from Eqs. [2, 3]) was shifted so the frequency offset of water peak was at 0 ppm. The approach was based on the fact that shifting the spectrum would also have resulted in shifts in the spatial position of the voxels. For conventional CEST, the images with saturation were normalized by the corresponding image without saturation, and B₀ correction was performed voxel-by-voxel by shifting the direct water saturation to 0 ppm²⁷. Then, the Z-spectrum was interpolated to 0.2 ppm step size and employed PCA-based denoising²⁸ to improve signal-to-noise (SNR) for both techniques.

The APT signal was extract using a modified PLOF method^29,30. In detail, a partial Z-spectrum (Z_acq) ranging from 1 to 6 ppm was used for fitting. First, the background spectrum (Z_back, excluding the APT signal between [2.4, 4.4] ppm, including DS, MTC, and other CEST effects, was fitted with a mixed Lorentzian and polynomial function. Second, by subtracting Z_back from the Z_acq, a residual spectrum (Z_resd) was obtained and integrated over the 3 to 4 ppm range to generate an APT contrast map. In addition, APT_w images were obtained using the averaged MTR asymmetry (MTR_asym) between 3 and 4 ppm, MTR_asym (Δω) = S_sat (-Δω)/S₀ - S_sat (+Δω)/S₀. Finally, the in vivo APT contrast and APT_W maps were smoothed by 3 × 3 median filter.

All code was written in-house with Python.

3. Results

Salicylic acid phantom

To assess the implementation of the TS-UFZ pulse sequence, experiments were performed on a salicylic acid phantom (exchangeable proton resonance around 9.3 ppm from water). The profile of the S_sat and S₀ TS-UFZ images across the slice direction are shown in Figure 2. On the S_sat image, the black band at the center of the slice shows that the frequency of the applied saturation pulse Δω_sat was adjusted thus the DS was at the center of the slice. A faint dark band labeled by the red arrow indicates the CEST effect corresponding to 9.3 ppm from water. From the saturated image, the Z-spectrum was recovered by normalizing S_sat with S₀. As shown in Figure 2C, the Z-spectra averaged over four acquisitions with varied G_sat shows a small standard deviation (SD) across the spectrum and clearly shows the salicylic acid CEST peak, illustrating that this implementation of the TS-UFZ method can provide clear CEST signals and has good reproducibility.

Figure 2.

TS-UFZ for a salicylic acid phantom (50 mM, pH = 7.5). (A) The illustration of slice chosen for TS-UFZ imaging. The salicylic acid phantom is labelled by yellow circle. (B) The profile of UFZ images across the slice direction with saturation (S_sat) and without saturation (S₀) for the salicylic acid phantom. The red arrow labels the CEST effect on the S_sat image. (C) Comparison of Z-spectra from TS-UFZ imaging (mean ± SD) and conventional CEST MRI. The standard deviation (SD) is over 4 acquisitions using G_sat = 4, 6, 8, 10 mT/m.

Egg white phantom

Experiments were performed on egg white to investigate the ability of the TS-UFZ technique to map the APT effect. As shown in Figure 3A–D, TS-UFZ S_sat and S₀ spectra of egg white displayed negligible deviations over three repeat scans for each G_sat. The resulting Z-spectra with small SDs (n = 3) were in good agreement with conventional CEST and successfully distinguished the APT effect under various G_sat (Figures 3E–H). However, a slight broadening of the direct water saturation peak was observed with increasing G_sat. The APT contrast was quantified from the residual spectrum by removing background signals from the acquired Z-spectrum, as illustrated in Figures 3I–L.

Figure 3.

TS-UFZ Z-spectra from an ROI in an egg white phantom for different saturation gradient strengths (G_sat = 4, 6, 8, 10 mT/m). (A-D) Mean intensities of S_sat and S₀ signals with standard deviation (SD) of TS-UFZ. The SD is over 3 repeat acquisitions for each G_sat. (E-H) The comparison between Z-spectra (mean ± SD) of TS-UFZ (blue line) and conventional CEST (red circles). (I-L) The extraction of APT signals for TS-UFZ (Z-spectrum from one scan for each G_sat). The acquired Z-spectrum (black open circles), background data points included in the fit (blue crosses inside the black circles), and fitted background (black line) are scaled on the left vertical axis. The APT contrast is quantified by integrating the residual spectrum (red dotted line, scale on the right vertical axis) between 3 to 4 ppm (gray shadow).

APT contrast maps from the egg white phantom obtained using TS-UFZ and conventional CEST MRI are shown in Figures 4A and 4B, respectively, demonstrating that the TS-UFZ contrast is comparable to conventional CEST. Interestingly, TS-UFZ water shift maps displayed a clear dependence on G_sat and differed from water shift maps determined using conventional CEST (Figures 4C, D). The water peak shift of conventional CEST is caused by B₀ inhomogeneity, whereas the TS-UFZ shift is due to a combination of B₀ inhomogeneity and an imprecise calculation of the off-center slice position (Δs). Consequently, Δω_c was not completely set to the water resonance frequency thus the water peak shifted depending on G_sat. This shift was corrected by shifting the nominal saturation frequency list in each voxel so that the frequency offset of water peak was assigned to 0 ppm.

Figure 4.

Comparison of APT contrast (A, B) and water peak shift (C, D) between TS-UFZ and conventional CEST MRI for egg white phantom. APT contrast maps for (A) TS-UFZ under G_sat = 4, 6, 8, 10 mT/m and (B) conventional CEST. (C, D) The water peak shift maps for both acquisition techniques.

Human brain

The feasibility of TS-UFZ imaging was further tested on healthy volunteers, by imaging the coronal orientation of the brain. The S_sat and S₀ TS-UFZ profiles (high in-plane resolution scan) across the slice direction are shown in Figures 5A, B. DS was centered at the middle sub-slice, which is clearly visible as a broad black band on the S_sat profile. TS-UFZ S_sat and S₀ for a selected region in white matter are shown in Figure 5C. After normalization, the TS-UFZ Z-spectral baseline shows slightly lower saturation when compared to conventional CEST (Figure 5D). We tentatively attribute this to the saturation gradient (G_sat) making the sequence sensitive to flow through the slice. To account for this baseline difference, the APT signal was extracted by removing the background from the acquired Z-spectra, as shown in Figures 5E, F.

Figure 5.

In vivo TS-UFZ results for subject 1 from the high in-plane resolution scan. (A, B) Profiles of TS-UFZ images in the slice direction with saturation (A, S_sat) and without saturation (B, S₀). (C) Intensities of S_sat and S₀ signals for the white matter ROI shown on a TS-UFZ image. (D) The Z-spectra of TS-UFZ and conventional CEST. (E, F) Extraction of APT signals for TS-UFZ and conventional CEST. The acquired Z-spectral data are shown as black open circles (scale on the left vertical axis), with selected background data points (circles filled with blue crosses) and fitted background (black line) also indicated. The APT contrast is quantified by integrating the residual spectrum (red line, scale on the right vertical axis) over the 3 to 4 ppm (gray shadow).

APT contrast maps were then obtained by integrating the residual curve, as shown in Figure 6. The TS-UFZ results show signal intensities in the Z-spectra that are on the same order of magnitude as conventional CEST results, but APT contrast across the slice differed between methods. The difference is likely due to the TS-UFZ image (1 ppm integral) representing a 0.5 mm sub-slice while the conventional CEST signal is the average over a 10 mm slice. In Figures 6D, G, there are clear differences in the APT_W maps generated from the two methods. The TS-UFZ maps showed unexpected heterogeneity, possibly due to Z-spectral baseline shifts and tissue heterogeneity between the images acquired at each frequency about the water peak.

Figure 6.

Comparison of APT contrast maps for a coronal slice through the midbrain of subject 1. (A) T_2w image. (B, C) TS-UFZ and (E, F) conventional CEST for both low (voxel size = 8 × 8 mm²) and high in-plane resolution (voxel size = 5 × 5 mm²). (D, G) APT_W maps obtained by using MTR_asym analysis for the two techniques (high-resolution image).

4. Discussion

In this work, we proposed an UFZ imaging approach in which a saturation gradient and the readout gradient were applied in the slice direction and phase-encoding in both in-plane directions, named TS-UFZ. Previous in vivo UFZ implementations, such as ultrafast localized CEST-spectroscopy with PRESS (UCEPR) and gradient-reversed ultrafast Z-spectroscopy (GRUFZS)^23,24, improved Z-spectral SNR by increasing the number of signal averages (NSA). The acquisition time was still significantly reduced compared to conventional CEST when the Z-spectroscopy of a single voxel was studied. However, increasing NSA is not practical for imaging a whole slice due to excessive scan times. Here, using TS-UFZ imaging, sufficient SNR could be achieved with NSA=1, the obtained APT contrast map resembled a conventional CEST approach for phantom experiments and on the same order of magnitude as conventional CEST for human brain.

The TS-UFZ Z-spectral resolution is determined by the resolution across the slice (L_s/N) at specific values of G_sat, which in this study (L_s = 10 mm, N = 100) resulted in a very high Z-spectral frequency resolution of 0.13 ppm when using G_sat = 4 mT/m, without requiring additional scan time. This frequency resolution ensures minimal chance of missing important Z-spectral features during CEST experiments without additional scan time as would be the case with conventional CEST where separate images need to be acquired per frequency leading to reduced spectral resolution for the same scan time. However, conventional CEST acquisition results in averaging over the whole slice thickness, while TS-UFZ acquisition presents signals from a sub-slice for each frequency offset and the thickness of each sub-slice is determined by N. The significant volume reduction for each frequency offset in TS-UFZ will result in the loss of in-plane SNR. Even though an exact comparison of TS-UFZ and conventional CEST is difficult due to different readout parameters, we compared the SNR of the in-plane S₀ images using two repeat acquisitions, where $\mathit{\text{SNR}}=\left(S_{0,1}+S_{0,2}\right)/\sqrt{2}\left(S_{0,1}-S_{0,2}\right)$.³¹ Table S1 shows the SNR ratios between conventional CEST and TS-UFZ, illustrating that conventional CEST does provide much higher in-plane SNR than TS-UFZ for both phantom and in vivo experiments.

To recover Z-spectral information, normalization with reference scan, S₀, was required to remove proton density contrast and to compensate for possible gradient non-linearity. As was discussed by Xu et al²⁰, UFZ imaging objects can have an irregular shape or even under severe B₀ inhomogeneity but these irregularities needs to be accounted for when determining the center frequency of the Z-spectrum, Δω_c using Eq. [4] above. In this work, Eq. [4] was used in addition to second-order shimming to minimize errors in placing the direct water saturation at the gradient isocenter. However, small B₀ shifts were still observed and became severe at high G_sat values (Figure 4C), which might be due to the imperfect correction of Δω_sat shifted Δω_c. Along the slice direction, each sub-slice was approximately 0.1 mm and corresponded to a single frequency offset. Δω_sat was corrected according to the object isocenter, and Δω_c set at 0 ppm ideally. However, the mathematical calculation for object isocenter might not be perfectly accurate due to field inhomogeneities. For example, when the object isocenter was calculated with an error of 0.1 mm, Δω_c would shift about 0.13 ppm at G_sat = 4 mT/m. Additionally, the in vivo situation is further complicated by subject motion where 0.1 mm movements can affect the Δω_sat calculation.

We also noticed that the DS signal width for TS-UFZ imaging was increased for larger G_sat (Figures 3E–H). The Eq. [2] illustrates that higher G_sat will expand BW_sat for specific values of L_s. Also, the frequency step size becomes correspondingly larger when the total number of frequency offsets (N) is fixed. One possible reason of the DS deviation in Figure 3 might be that the spectral dispersion at higher G_sat results in less points around 0 ppm in the Z-spectrum and this in-turn reduces in the Z-spectrum intensity around 0 ppm. Additionally, while the saturation gradient was segmented to account for droop from the gradient amplifier, there might still be a small droop for G_sat. For example, for nominal G_sat = 10 mT/m and L_s = 10 mm, if the effective G_sat is 2% less than nominal G_sat, the effective BW_sat will be 535 rad/s (0.67 ppm) less than nominal BW_sat, which will be reflect as stretching of the frequency axis and become apparent around DS. Such BW_sat decrease would show a dependence with increasing G_sat. In terms of possible droop and spectral dispersion, some signal variation around DS was observed when using a G_sat = 8, 10 mT/m in the egg white phantom and therefore a lower G_sat (6 mT/m) was selected for in vivo experiments. The gradient strength used here was higher than previous in vivo studies which used G_sat = 3 mT/m^23,24. This was to ensure that the saturation gradient was uniform across the whole slice for the Z-spectral region of interest (we noticed non-uniform edges of the slice profile when using 3 mT/m).

Another important consideration for TS-UFZ in vivo is tissue heterogeneity across the saturation gradient direction. Each saturation offset in the Z-spectrum corresponds to a different spatial location and thus different Z-spectral regions may contain signals from different tissue types. Consequently, MTR_asym-based quantification approach may provide erroneous results since the reference frequency offset does not correspond to the same tissue type (Figures 6D, S2D, S4D). Therefore, in this work, the APT signal was extracted using a fitting approach, and APT image contrast was quantified by integrating the signal curve over a narrow saturation frequency range of 1 ppm (~ 0.5 mm), over which voxels are assumed to have reasonably similar tissue properties. Another potential source for the in vivo error is that, compared to conventional CEST methods which saturate the whole brain, the TS-UFZ imaging saturation is very position dependent and bulk motion (e.g., flow) through the slice can lead to Z-spectral signal fluctuations. Unsaturated CSF and blood flowing into the slice region will affect signal intensities for S_sat and S₀, which we think caused the difference in baseline with the conventional approach in Figure 5. The coronal orientation of brain was studied in this work rather than transverse orientation, as it had less flow interference. To eliminate the baseline difference, APT signal was extracted by removing the fitted background from the acquired Z-spectra, using a modified PLOF method^29,30. The in vivo TS-UFZ approach was tested on three healthy human volunteers and provided APT signal intensities on the same order of magnitude in the Z-spectra as conventional CEST but with different contrast across the slice which likely is a result of different tissue regions contributing to the signal in each method (Figures 5, 6, and S1 – S4).

Additionally, the sub-slice thickness for each frequency offset is constant for a fixed N and a given L_s at different G_sat. However, higher G_sat increases BW_sat (Eq. [2]). So the measured sub-slice thickness over the same frequency range will decrease with increasing G_sat and the sub-slice positions will change accordingly. For example, in the egg white phantom experiments, the 10 mm slice thickness at G_sat = 4, 6, 8, 10 mT/m corresponded to BW_sat values of 13.3, 20.0, 26.6, and 33.3 ppm, respectively. In turn, the Z-spectral signal from 3 to 4 ppm corresponds to the measured slice thicknesses of 0.75, 0.5, 0.38 and 0.3 mm, respectively. Hence, different saturation gradient strengths can change the resulting contrast since the measured slice thickness and sub-slice positions contributing to the signal over a specific frequency range will be different. The APT contrasts from the egg white phantom under different G_sat resembled each other well (Figures 3, 4), because the sample was homogeneous. The in vivo situation is more complex due to the tissue heterogeneity across the saturation gradient direction and contrast will change with different G_sat.

Theoretically, the TS-UFZ method should speed up acquisition because saturation pulses at only two frequency offsets (S_sat and S₀) are needed for the entire Z-spectral image series. However, TS-UFZ imaging using a traditional 3D SE imaging technique resulted in an acquisition time dependent on the acquired in-plane voxel size. Compared to conventional CEST, in vivo TS-UFZ imaging could save about three-quarters of the scan time for low image resolution (voxel size of 8 × 8 mm²). However, savings were not significant for high-resolution TS-UFZ imaging (voxel size of 5 × 5 mm²). Still, several acceleration methods can be applied to the phase-encoding steps to significantly speed up acquisition time, for instance compressed sensing and parallel imaging^32–35. This would allow for high spatiotemporal resolution dynamic studies. In this work, we first used SENSE factors of 1 (no acceleration in k-space) for both in-plane directions to guarantee good image quality for the salicylic acid phantom. However, there was still a temporal acceleration in the acquisition of TS-UFZ. Then the egg white phantom which with same in-plane FOV and voxel size as the salicylic acid phantom used an increased in-plane SENSE factor (2 and 1 in Figures 3 and 4; 2 and 2 in Figures S5 and S6), allowing a scan time saving by a factor of two and four with respect to the salicylic acid phantom, respectively (2 mins 15 s for 1 and 1, 53 s for 2 and 1, and 30 s for 2 and 2). Further, this method has the potential to be extended to whole brain coverage by incorporating multi-slice readout methods.

5. Conclusion

In this work, we propose a TS-UFZ imaging method which is an extension of gradient-encoded ultrafast Z-spectroscopy but has spectroscopic imaging capabilities. It was demonstrated on a 3T clinical scanner for both phantom and in vivo applications. TS-UFZ imaging was able to distinguish CEST effects as a function of frequency in both cases. In phantoms, APT contrast maps were in excellent agreement with those for conventional CEST MRI. In vivo, APT contrast intensities in Z-spectra were of the same order of magnitude as for conventional CEST, but maps showed differences that are likely due to the large difference in slice thickness for the integrated APT region.

Supplementary Material

SUPINFO

Figure S1. In vivo TS-UFZ results for subject 2 from the high in-plane resolution scan. (A, B) Profiles of TS-UFZ images in the slice direction with saturation (A, S_sat) and without saturation (B, S₀). (C) Intensities of S_sat and S₀ signals for the white matter ROI shown on a TS-UFZ image. (D) The Z-spectra of TS-UFZ and conventional CEST. (E, F) Extraction of APT signals for TS-UFZ and conventional CEST.

Figure S2. Comparison of APT contrast maps for a coronal slice through the midbrain of subject 2. (A) T_2w image. (B, C) TS-UFZ and (E, F) conventional CEST for both low (voxel size = 8 × 8 mm²) and high in-plane resolution (voxel size = 5 × 5 mm²). (D, G) APT_W maps obtained by using MTR_asym analysis for the two techniques (high-resolution image).

Figure S3. In vivo TS-UFZ results for subject 3 from the high in-plane resolution scan. (A, B) Profiles of TS-UFZ images in the slice direction with saturation (A, S_sat) and without saturation (B, S₀). (C) Intensities of S_sat and S₀ signals for the white matter ROI shown on a TS-UFZ image. (D) The Z-spectra of TS-UFZ and conventional CEST. (E, F) Extraction of APT signals for TS-UFZ and conventional CEST.

Figure S4. Comparison of APT contrast maps for a coronal slice through the midbrain of subject 3. (A) T_2w image. (B, C) TS-UFZ and (E, F) conventional CEST for both low (voxel size = 8 × 8 mm²) and high in-plane resolution (voxel size = 5 × 5 mm²). (D, G) APT_W maps obtained by using MTR_asym analysis for the two techniques (high-resolution image).

Figure S5. TS-UFZ Z-spectra from an ROI in an egg white phantom for different saturation gradient strengths (G_sat = 4, 6, 8, 10 mT/m) with SENSE factor of 2 and 2 in both in-plane (the total scan time of acquiring both S_sat and S₀ was 45 s). (A-D) Mean intensities of S_sat and S₀ signals with standard deviation (SD) of TS-UFZ. The SD is over 3 repeat acquisitions for each G_sat. (E-H) The comparison between Z-spectra (mean ± SD) of TS-UFZ (blue line) and conventional CEST (red circles). (I-L) The extraction of APT signals for TS-UFZ (Z-spectrum from one scan for each G_sat).

Figure S6. APT contrast of TS-UFZ for egg white phantom under G_sat = 4, 6, 8, 10 mT/m and SENSE factor of 2 and 2 in both in-plane (the total scan time of acquiring both S_sat and S₀ was 45 s).

Table S1. SNR ratios between TS-UFZ and conventional CEST calculated for in-plane S₀ images using two repeat acquisitions, $\mathit{\text{SNR}}=\left(S_{0,1}+S_{0,2}\right)/\sqrt{2}\left(S_{0,1}-S_{0,2}\right)$.