Quasi-steady-state amide proton transfer (QUASS APT) MRI enhances pH-weighted imaging of acute stroke

Abstract

Purpose:

Chemical exchange saturation transfer (CEST) imaging measurement depends not only on the labile proton concentration and pH-dependent exchange rate but also on experimental conditions, including the relaxation delay and radiofrequency (RF) saturation time. Our study aimed to extend a quasi-steady-state (QUASS) solution to a modified multi-slice CEST MRI sequence and test if it provides enhanced pH imaging after acute stroke.

Methods:

Our study derived the QUASS solution for a modified multislice CEST MRI sequence with an unevenly segmented RF saturation between image readout and signal averaging. Numerical simulation was performed to test if the generalized QUASS solution corrects the impact of not sufficiently long relaxation delay, primary and secondary saturation times, and multi-slice readout. In addition, multiparametric MRI scans were obtained after middle cerebral artery occlusion, including relaxation and CEST Z-spectrum, to evaluate the performance of QUASS CEST MRI in a rodent acute stroke model. We also performed Lorentzian fitting to isolate multi-pool CEST contributions.

Results:

The QUASS analysis enhanced pH-weighted magnetization transfer asymmetry contrast over the routine apparent CEST measurements in both contralateral normal (−3.46±0.62% (apparent) vs. −3.67±0.66% (QUASS), P<0.05) and ischemic tissue (−5.53±0.68% (apparent) vs. −5.94±0.73 % (QUASS), P<0.05). Lorentzian fitting also showed significant differences between routine and QUASS analysis of ischemia-induced changes in magnetization transfer, amide, amine, guanidyl CEST, and NOE (−1.6 ppm) effects.

Conclusion:

Our study demonstrated that generalized QUASS analysis enhanced pH MRI contrast and improved quantification of the underlying CEST contrast mechanism, promising for further in vivo applications.

Introduction

Chemical exchange saturation transfer (CEST) MRI provides a sensitive contrast mechanism for detecting the ‘invisible’ dilute labile groups via their exchange with the abundant bulk water protons (1). The CEST effect is measured as a loss of water signal, the magnitude of which confers image sensitivity to labile proton concentration and exchange rate. Because the chemical exchange is often pH-dependent (2), CEST MRI provides a non-invasive in vivo pH-weighted imaging approach (3). Specifically, the endogenous amide proton transfer-weighted (APTw) MRI is pH-dependent, promising to detect tissue acidification after acute stroke (4–10). In addition, pH MRI has been postulated to complement diffusion and perfusion MRI to stratify heterogeneous ischemic tissue insult, with some initial success in early clinical translation (11–18). Continued technical progress has been accomplished in improving the spatial coverage (19–22), specificity (23–26), and quantification (27–33) of CEST imaging.

Despite its increasing use, pH-weighted MRI is susceptible to multiple confounding factors. The CEST MRI effect reflects a balance of signal decrease due to saturation transfer and signal recovery via T₁ relaxation (34–38). As such, the CEST measurement has a complex dependence on the radiofrequency (RF) saturation time (Ts) and relaxation delay (Td), and flip angle (39,40). Although long Ts and Td are desired for modeling the CEST signal, they are often shortened experimentally to minimize the scan time. Therefore, it is not straightforward to isolate T₁-weighting from CEST MRI measurement, particularly at the non-steady-state (38,41). The situation is further complicated due to a non-negligible T₁ change following acute stroke, which, if not accounted for, confounds the origin of amide proton exchange with the nuclear overhauser enhancement (NOE) effect (42–44). As a result, the APTw MRI has a complex dependence on scan parameters and tissue properties (45–47). Recently, a quasi-steady-state (QUASS) CEST analysis has been established to derive the equilibrium CEST effect (i.e., long Ts and Td) from the experimental measurements, improving the quantification of the underlying CEST system (48–53). Kim et al. examined QUASS CEST MRI for a CEST sequence with a multi-slice readout after a single RF saturation pulse (54). Although multi-slice readout can be obtained after a single RF labeling pulse, each subsequent slice suffers an additional loss of CEST effect due to post-saturation delay (55). Hence, the current work generalized the QUASS CEST solution to a modified sequence that includes a long primary RF saturation, followed by interleaved multi-slice image readout and short secondary saturation pulses (19). The use of repeated secondary labeling RF has been shown advantageous to maintaining CEST contrast for multi-slice readout. Both numerical simulation (3-pool) and an animal model of acute stroke have been utilized to test the generalized QUASS CEST imaging, laying the groundwork for future clinical translation.

Theory

The QUASS calculation models the impacts of experimental parameters on the CEST signal. The equilibrium CEST effect can be calculated from the experimental measurements by solving the QUASS spin-lock relaxation rate, despite not-sufficiently long Ts and Td (56).

The original QUASS algorithm was developed for a CEST sequence with a single RF saturation followed by a fast image readout. Our study aims to generalize the derivation to a modified multi-slice CEST MRI sequence that includes a long primary RF saturation, followed by interleaved multi-slice image readout, signal average, and short secondary saturation pulses. For the modified CEST MRI sequence, the first round of MRI signal relaxes under the Td, Ts1, and slice-dependent Ts2, while the rest of the signal averages have identical effective relaxation for all slices. Therefore, the control scan without RF saturation can be shown to be

$$I_0^{app}(i,N,NA)\approx \left(1-e^{-R_{1w}\bullet [(\text{Td}+\text{Ts}1)+\text{Ts}2*(\text{i}-1)]}\right)+{\sum }_2^{NA}\left(1-e^{-R_{1w}\bullet [\text{N}\bullet \text{Ts}2]}\right)$$ (1)

where i is the ith slice, N is the total number of slices, and NA is the number of signal averages. The signal was also derived for the saturated scans by summing the first acquisition and the rest of the signal averaging readout. We have

$$I_{sat}^{app}(i,N,NA)\approx \left(1-e^{-R_{1w}\bullet Td}-\frac{R_{1w}co{s}^2\theta }{R_{1\rho }}\right)e^{-R_{1\rho }\bullet [Ts1+Ts2*(i-1)]}+\frac{R_{1w}co{s}^2\theta }{R_{1\rho }}\{1+{\sum }_2^{NA}(1-e^{-R_{1\rho }\bullet [N\bullet Ts2]})\}$$ (2)

in which $\theta =\mathrm{atan}\left(\frac{\gamma B_1}{∆\omega }\right)$, γ is the gyromagnetic ratio, and R₁ and R₂ are the bulk tissue water longitudinal and transverse relaxation rates. In addition, B₁ and Δω are the amplitude and offset of the RF irradiation. Altogether, the apparent Z spectrum for the modified CEST sequence is given as,

$$Z^{app}=\frac{\left(1-e^{-R_{1w}\bullet Td}-\frac{R_{1w}co{s}^2\theta }{R_{1\rho }}\right)e^{-R_{1\rho }\bullet [Ts1+Ts2*(i-1)]}+\frac{R_{1w}co{s}^2\theta }{R_{1\rho }}\left\{1+{\sum }_2^{NA}\left(1-e^{-R_{1\rho }\bullet [N\bullet Ts2]}\right)\right\}}{(1-e^{-R_{1w}\bullet [(Td+Ts1)+Ts2*(i-1)]})+{\sum }_2^{NA}(1-e^{-R_{1w}\bullet [N\bullet Ts2]})}$$ (3)

The QUASS CEST MRI solves the spinlock relaxation rate ($R_{1\rho }^{QUASS}$) from the apparent Z-spectrum, from which the equilibrium CEST effect is calculated as

$$Z^{QUASS}=\frac{R_1\bullet co{s}^2\theta }{R_{1\rho }^{QUASS}}$$ (4)

Methods

Simulation

We simulated the CEST effect using classical 3-pool Bloch McConnell (BM) equations (57) in MATLAB 2021a (Mathworks, Natick MA) for a modified multislice CEST MRI sequence, as shown in Supplementary Figure 1 (19). Specifically, the sequence includes a relatively long primary saturation module (Ts1), so the CEST signal reasonably approaches its equilibrium state, followed by a secondary RF saturation module of moderate duration (Ts2) to maintain the CEST effect between multislice readout and signal averaging. We assumed representative T₁ of 1.5 s, 1 s, and 1 s for bulk water, amide proton, and semisolid macromolecules, with their T₂ being 50 ms, 10 ms, and 10 μs, respectively. The exchange rates were assumed to be 100 s⁻¹ for amide protons (3.5 ppm) and 23 s⁻¹ for semisolid magnetization transfer (MT) at 0 ppm, with their fraction ratio being 0.125% and 13.9%, respectively (58). We simulated the parameters used in the experiment, being 5 slices and 2 averages (B₁=0.75 μT at 4.7 Tesla), under two conditions: 1) Td=3.5 s, Ts1=3 s, and Ts2=0.5 s (same parameters as used in the in vivo scans); 2) Td=10 s, Ts1=10 s, and Ts2=10 s (serving as the equilibrium reference.)

Animals

In vivo studies have been approved by the Institutional Animal Care and Use Committee. We performed multiparametric MRI scans in 7 adult male Wistar rats following acute stroke, with anesthesia maintained between 1.5-2% isoflurane and air mixture (1-1.5 liter/min) for the duration of the stroke surgery and imaging. The heart rate and blood oxygen saturation were monitored (Nonin Pulse Oximeter 8600, Plymouth, MN), with their core temperature maintained with a warm water jacket surrounding the torso. The middle cerebral artery (MCA) occlusion was induced by gently inserting a 4-0 silicone-coated nylon filament through the internal carotid artery to the origin of the MCA. Stroke rats were imaged between 1-3 hours after acute ischemia.

MRI

All MRI scans were performed at a 4.7 Tesla small-bore MRI system (Bruker Biospec, Ettlingen, Germany). We used single-shot spin echo echo planar imaging (EPI) with a field of view (FOV) of 25 x 25 mm² (matrix = 64 x 64, slice thickness/slice gap = 1.8/0.2 mm, 5 slices). We acquired multiparametric MRI scans, including diffusion-weighted, T₁-weighted, T₂-weighted, and CEST Z-spectral imaging. Specifically, T₁–weighted inversion recovery images were acquired with inversion times ranging from 250 to 2750 ms (relaxation delay (Td)/echo time (TE)=6000/30 ms, 4 averages, scan time= 3 min 25 s). T₂–weighted spin echo images were acquired with separate spin echo times, with two echo times being 30 and 100 ms (repetitio time (TR)=3000 ms, 8 averages, scan time= 48 s). In addition, diffusion imaging was obtained with a single-shot isotropic diffusion-weighted MRI (two b-values of 250 and 1000 s/mm², TR/TE=3250/42 ms, 16 averages, scan time =1 min 44 s) (59). We used a CEST MRI sequence with an unevenly segmented RF irradiation (19). We collected Z-spectrum from −6 to 6 ppm with intervals of 0.05 ppm (B₁ = 0.75 μT, Td = 3500 ms, Ts1 = 3000 ms, Ts2= 500 ms, 2 averages, scan time = 42 min 36s). We chose 0.75 μT, a relatively weak RF saturation amplitude, to maximize the pH-weighted APT contrast (60). The CET scan time is relatively long due to the acquisition of densely sampled CEST images (i.e., 241 offsets). The total scan time is provided (scan time = [Ts1+Td+(Ts2 * the number of slices) * the number of averages] * the number of offsets). The use of repeated secondary labeling RF is advantageous to maintaining CEST contrast for multi-slice readout.

Data Analysis

Images were processed in Matlab R2021a (Mathworks, Natick, MA). Parametric T₁, T₂, and apparent diffusion coefficient (ADC) maps were obtained using least-squares mono-exponential fitting of the signal intensities as functions of the inversion time, echo time, and diffusion b values, as described before (61). The B₀ field inhomogeneity was determined from the Z-spectrum scan due to its fine frequency density. The Z spectrum was analyzed using the unsaturated scan denominated normalization (i.e., $Z=\frac{\mathrm{I}\left(∆\omega \right)}{{\mathrm{I}}_0}$), in which I(Δω) is the signal intensity with RF saturation applied at Δω, and I₀ is the control scans without RF saturation. The Z-spectra were also analyzed with the inverse Z-spectral analysis (i.e., $Z_{inv}=\frac{I_0}{I\left(∆\omega \right)})$ (62,63). The direct water saturation (DWS) effect was calculated as

$$Z^{DWS}=\frac{R_{1}co{s}^2\theta }{R_{1}co{s}^2\theta +R_{2}si{n}^2\theta },$$ (5)

The DWS exchange spectrum was calculated as,

$$R_{ex}^{DWS}(\text{Δ}\omega )=R_{1}co{s}^2\theta +R_{2}si{n}^2\theta$$ (6)

The CEST exchange spectrum was calculated as

$$R_{ex}(\text{Δ}\omega )=R_1\bullet co{s}^2\theta \bullet Z_{inv}-R_{ex}^{DWS}(\text{Δ}\omega )$$ (7)

Diffusion lesion was segmented using a K-means clustering algorithm (64). The ipsilateral ischemic lesion was mirrored along the midline to the contralateral brain as the reference region of interest. The change in R_ex was calculated as the difference between R_ex of the ischemic lesion and the contralateral normal area (${\text{∆}R}_{ex}=R_{ex}^{ischemic}-R_{ex}^{normal}$).

We applied Lorentzian fitting to resolve multi-pool contribution to Z-spectra using the equation

$$Z=\frac{R_{1}co{s}^2\theta }{R_{1}co{s}^2\theta +R_{2}si{n}^2\theta +{\sum }_{i=1}^N\frac{A_i}{(1+4{\left(\frac{\text{Δ}\omega -{\delta }_i}{{\sigma }_i}\right)}^2)}}$$ (8)

where $A_i,{\delta }_i$, and ${\sigma }_i$ are the amplitude, chemical shift, and full width half maximum (FWHM) of i^th labile proton group. We used 7 labile groups, including semisolid MT (0 ppm), amide (3.5 ppm), amine (3 ppm), guanidyl (2 ppm), NOE (−1.6 ppm), NOE (−2.75 ppm), and NOE (−3.5 ppm) pools (65). Also, we used experimentally measured R₁ and R₂ relaxation rates to minimize the number of free parameters. The upper and lower boundaries are from 0 to 150% of the initial guesses for the amplitudes, ±25% for the FWHM, with the range for the chemical shifts being ±0.25 ppm, with an extended range of ± 2 ppm for the MT pool from the initial values. We used a two-tailed paired t-test with Bonferroni correction, and the results were regarded as statistically significant for adjusted P values less than 0.05.

Results

The generalized QUASS algorithm was tested with numerical simulation for the fast multislice CEST MRI sequence. The effective Td and Ts were derived per slice because they vary with slices. Figure 1a shows the apparent Z-spectra for each of the five slices were relatively close to each other, with slices 1 to 5 represented in colors of blue, red, yellow, purple, and green, respectively. The Z-spectrum under the condition of long Ts and Td was shown in solid red, which was noticeably more attenuated from the apparent Z-spectra. This difference is expected because the equilibrium state experiences more saturation transfer than under the condition of insufficiently long Ts. The corresponding QUASS reconstructed Z-spectra were shown in colored dashed lines, overlapping the equilibrium Z-spectral simulation. Figure 1b shows the Z-asymmetry spectra, revealing slice-dependent magnitude in the CEST effect. Again, the reconstructed QUASS asymmetry spectra and equilibrium asymmetry spectra (long Td, Ts1, and Ts2) overlapped, validating the generalized QUASS solution.

Figure 1.

Simulation of QUASS algorithm for the modified fast multislice CEST MRI sequence. a) Simulated Z-spectra for slices 1 to 5 were represented in colors of blue, red, yellow, purple, and green, respectively. They showed less attenuation than the reconstructed QUASS Z-spectra (dashed lines), which overlays nicely with the simulated Z-spectra assuming long Td and Ts (solid red). b) The corresponding asymmetry spectra, revealing slice-dependent magnitude in the CEST effect. Again, the reconstructed QUASS asymmetry spectra and equilibrium asymmetry spectra (long Td, Ts1, and Ts2) overlapped.

Multiparametric images of a representative acute stroke image are shown in Supplementary Figure 2. T₁ map (Supplementary Figure 2a) showed a small yet significant T₁ increase in the ipsilateral ischemic lesion from the contralateral normal area (1.56 ± 0.04 s vs. 1.72 ± 0.07 s, P<0.05). The T₂ map (Supplementary Figure 2b) also showed a small yet significant increase, from 54.77 ± 0.95 ms to 56.17 ± 1.13 ms (P<0.05). In addition, the diffusion map (Supplementary Figure 2c) shows ADC drop in the ischemic lesion, from 0.85 ± 0.04 μm²/ms to 0.53 ± 0.04 μm²/ms. Figure 2 compares the Z-, inverse Z-, R_ex-, and ΔR_ex-spectra from the routine CEST analysis. Figure 2a shows the apparent Z-spectra from the contralateral normal (red) and ipsilateral ischemic area (blue). The DWS spectra were simulated using Eq. 5, which showed a very small drop in acute ischemic tissue, suggesting DWS is not sensitive to a small T₁ change due to the normalization ($\frac{R_1\bullet {cos}^2\theta }{R_1\bullet {cos}^2\theta +R_2\bullet {sin}^2\theta }=\frac{1}{1+R_{2/}{R}_1\bullet {tan}^2\theta }$). The T₁-normalized inverse Z spectrum calculated from Eq. 7 showed a broad baseline drop in the ischemic tissue (Figure 2c) from the contralateral normal area. The simulated $R_{ex}^{DWS}$ (i.e., $R_1\bullet {cos}^2\theta +R_2\bullet {sin}^2\theta$) spectra showed that the baseline drop was primarily attributed to decreases in ischemic tissue relaxation rates. The CEST/MT exchange spectra excluding the DWS effect were shown in Figure 2e, which showed that most of the differences between the normal and ischemic tissues are around 3.5 ppm, amide proton offset instead of NOE effect −3.5 ppm. Figure 2f shows CEST/MT R_ex difference (ΔR_ex) between the normal area and ischemic lesion, confirming that APT signal change dominates that of NOE.

Figure 2.

The analysis of apparent CEST Z-spectra. a) Z-spectra from the contralateral normal area (red) and ischemic lesion (blue). b) Simulated Z-spectra from T₁ and T₂ maps for the normal and ischemic ROIs. c) $R_{ex}$ spectra from the normal area (red) and ischemic lesion (blue). d) Simulated $R_{ex}^{DWS}$ spectra from the normal and ischemic ROIs. e) CEST/MT spectra (i.e., $R_{ex}-R_{ex}^{DWS}$) from the normal area (red) and ischemic lesion (blue). f) The CEST/MT spectral difference between the normal and ischemic ROIs.

We also analyzed the QUASS CEST results. Figure 3a shows the QUASS Z-spectra for the contralateral normal (red) and ischemic tissue (blue). The supplementary Figure 3 compares routine and QUASS Z-spectra, showing further QUASS Z-spectral attenuation in the normal (Supplementary Figure 3a) and ischemic tissue (Supplementary Figure 3b) from the corresponding routine Z spectra. This difference is because routine Z spectra were acquired under not sufficiently long Ts and Td, while the QUASS processing restores the equilibrium CEST effects, similar to the simulated results (Figure 1a). Figure 3b displayed R_ex spectra, removing $R_{ex}^{DWS}$ calculated from measured relaxation times to account for CEST/MT effects changes. In contrast to Z-spectra, R_ex-spectra show little intensity change in the high field shifts, with most of the differences originating from around the amide chemical shift (Table 1). This result agreed with prior findings that the relaxation change causes a baseline drop, introducing an apparent Z-spectral signal decrease at aliphatic chemical shifts (44). In addition, due to the correction of the incomplete saturation transfer effect, QUASS R_ex spectra are consistently higher than the apparent R_ex spectra, particularly close to the bulk water resonance. The ΔR_ex spectra between the contralateral normal and ischemic lesion were shown in Figure 3c. Indeed, ${\text{∆}R}_{ex}^{QUASS}(-3.5ppm)$ is not statistically different from 0 (−0.00 ± 0.01).

Figure 3.

The analysis of QUASS CEST Z-spectra. a) Z-spectra from the contralateral normal area (red) and ischemic lesion (blue). b) CEST/MT QUASS spectra (i.e., $R_{ex}-R_{ex}^{DWS}$) from the normal area (red) and ischemic lesion (blue). c) The CEST/MT QUASS spectral difference between the normal and ischemic ROIs.

Table 1.

Table 1 compares multiparametric pH-weighted indices after acute stroke with and without QUASS analysis.

	Contralateral Normal Area		Ischemic Lesion		Difference (Ischemic – Normal ROIs)
	Apparent	QUASS	Apparent	QUASS	Apparent	QUASS
MTRasym @3.5 ppm (%)	−3.46±0.62#*	−3.67±0.66#*	−5.53±0.68#*	−5.94±0.73#*	−2.07±0.52*	−2.27±0.57*
MTRasym @2 ppm (%)	−0.32±0.60	−0.33±0.61	−0.85±0.37	−0.87±0.39	−0.53±0.53	−0.55±0.54
Rex @ 3.5 ppm (1/s)	0.19±0.02#*	0.25±0.02#*	0.16±0.02#*	0.22±0.03#*	−0.03±0.01	−0.03±0.01
Rex @ −3.5 ppm (1/s)	0.23±0.03*	0.30±0.02*	0.22±0.03*	0.30±0.03*	−0.01±0.01*	−0.00±0.01*
Rex @ 2 ppm (1/s)	0.20±0.03#*	0.28±0.04*	0.18±0.04#*	0.27±0.04*	−0.02±0.01*	−0.01±0.01*
Rex @ −2 ppm (1/s)	0.21±0.03*	0.28±0.03*	0.18±0.04*	0.28±0.04*	−0.01±0.01*	−0.00±0.01*

Paired t-test was performed to test ischemia and normal values for apparent and QUASS CEST measurements (# P< 0.05), as well as between the apparent and QUASS values (* P<0.05).

The apparent and QUASS MTR_asym and CEST/MT $R_{ex}$ images were compared. Figs. 4a and 4d display MTR_asym images calculated from the apparent and QUASS Z-spectra, respectively. The QUASS processing resulted in a more ischemia-induced signal drop (red arrow, Figure 4d) from the routine image (Figure 4a). The apparent MTR_asym was −3.46±0.62% and −5.53±0.68% for the contralateral normal and ischemic areas, respectively, with a difference of −2.07 ± 0.52%. The corresponding QUASS MTR_asym was −3.67±0.66% and −5.94±0.73%, respectively, with a difference of −2.27±0.57%. In addition, the apparent and QUASS $R_{ex}(+3.5ppm)$ images were shown in Figs. 4b and 4e, respectively, with corresponding $R_{ex}(-3.5ppm)$ images in Figs. 4c and 4f. It is worth noting that QUASS $R_{ex}$ images were consistently higher than the apparent $R_{ex}$ images, with the ischemic lesion showing notable $R_{ex}\left(+3.5ppm\right)$ hypointensity.

Figure 4.

Comparison of apparent and QUASS CEST images. a) The apparent MTR_asym (3.5 ppm) map. b) The apparent CEST/MT exchange spectral at +3.5 ppm (i.e., APT). c) The apparent CEST/MT exchange spectral at −3.5 ppm (i.e., NOE). d) The QUASS MTR_asym (3.5 ppm) map. e) The QUASS CEST/MT exchange spectral at +3.5 ppm (i.e., APT). f) The QUASS CEST/MT exchange spectral at −3.5 ppm (i.e., NOE).

Because R_ex reflects the combined changes of the baseline and labile proton signals, we further applied Lorentzian fitting to isolate multi-pool CEST contribution from the ischemic lesion and the contralateral normal area (Table 2). The mean and standard deviation of Lorentzian fitting from the apparent and QUASS Z-spectra (normal (solid lines) and ischemic (dash-dotted lines)) were shown in Figures 5a and 6a, with isolated peaks (Figs. 5b and 6b) and the differences between fitting and data (Figs. 5c and 6c). The root-mean-square deviation (RMSD) for all fitting was under 0.25%, suggesting good fitting. It is worth noting that QUASS analysis detected significant signal changes in MT, amide, and guanidyl pools. Briefly, the amplitude of the MT pool was (20.83±2.41 %/s (normal) vs. 18.34±2.75 %/s (ischemia), P<0.05) from the apparent CEST solution, while they were 27.86±2.57 %/s and 26.12±2.97 %/s (P<0.05) from QUASS analysis. In addition, the apparent amide peaks decreased from 2.14±0.29 %/s to 1.45±0.20 %/s (P<0.05), following ischemia, with the corresponding QUASS amide peaks being 2.61±0.34 %/s and 1.78±0.21 %/s (P<0.05), respectively. Also, the apparent guanidyl signal increased from 1.84±0.28 %/s to 2.80±0.39 %/s (P<0.05), following ischemia, while the QUASS guanidyl signal increased from 2.09± 0.38 %/s and 3.28±0.44 %/s (P<0.05).

Table 2.

Table 2, Lorentzian fitting resolves the multi-pool signal changes following acute stroke.

Amplitude of isolated CEST effects (multi-pool Lorentzian fitting)	Contralateral Normal Area		Ischemic Lesion		Difference (Ischemic – Normal ROIs)
	Apparent	QUASS	Apparent	QUASS	Apparent	QUASS
MT (%/s)	20.83±2.41#*	27.86±2.57#*	18.34±2.75#*	26.12±2.97 #*	−2.49±0.70*	−1.73±0.86*
Amide @ 3.5 ppm (%/s)	2.14±0.29#*	2.61±0.34#*	1.45±0.20#*	1.78±0.21#*	−0.69±0.25*	−0.83±0.30*
Amine @ 3 ppm (%/s)	2.61±0.47#*	2.78±0.59#*	1.56±0.37#*	1.39±0.55#*	−1.05±0.41*	−1.39±0.48*
Guanidyl @ 2 ppm (%/s)	1.84±0.28#*	2.09±0.38#*	2.80±0.39#*	3.28±0.44#*	0.96±0.26*	1.19±0.25*
NOE @ −1.6 ppm (%/s)	0.68±0.50	0.58±0.45	0.58±0.34*	0.28±0.21*	−0.09±0.38*	−0.30±0.40*
NOE @ −2.75 ppm (%/s)	0.91±0.36	1.19±0.57	1.26±0.43	1.21±0.46	0.36±0.52	0.02±0.90
NOE @ −3.5 ppm (%/s)	5.18±0.29*	5.44±0.27*	5.16±0.47*	5.64±0.58*	−0.02±0.23	−0.20±0.45

Paired t-test was performed to test ischemia and normal values from apparent and QUASS CEST measurements (# P< 0.05), as well as between the apparent and QUASS values (* P<0.05).

Figure 5.

Lorentzian fitting of the apparent Z-spectra. a) Apparent Z-spectra from the contralateral normal (red) and ipsilateral ischemic lesion (blue). b) Isolated multi-pool apparent spectral peaks, with the normal area (solid lines) and ischemic lesion (dash-dotted lines). c) The residual errors between the apparent Z-spectral measurement and fitting for the contralateral normal (red) and ipsilateral ischemic lesion (blue).

Figure 6.

Lorentzian fitting of the QUASS Z-spectra. a) QUASS Z-spectra from the contralateral normal (red) and ipsilateral ischemic lesion (blue). b) Isolated multi-pool QUASS spectral peaks, with the normal area (solid lines) and ischemic lesion (dash-dotted lines). c) The residual errors between the reconstructed QUASS spectra and fitting for the contralateral normal (red) and ipsilateral ischemic lesion (blue).

Discussion

Our study extended the QUASS algorithm to describe a modified fast multislice CEST MRI sequence. Capitalizing on a densely sampled Z-spectrum from an acute ischemic rodent model, we found that the QUASS reconstruction enhances pH-weighted MTR_asym contrast to depict the ischemic insult over the routine Z-spectral analysis. Consistent with prior studies (43,44), the QUASS $R_{ex}$ spectrum confirmed that APT signal drop dominates NOE signal change, which was negligible after acute ischemia.

The modified sequence has an effective saturation time ranging from 5 s (slice 5, 1^st average) to 2.5 s (2^nd average and beyond), comparable with the typical experimental choice of saturation parameters. We used a numerical simulation of a 3-pool model (amide, MT, and bulk water) to test the generalized QUASS solution. Once the derivation is confirmed with simulation, it is safe to assume the generalized QUASS solution is valid. This is because of prior QUASS experiments, in which we experimentally obtained Z-spectra (phantoms, brain tumor models, and human subjects) with multiple Ts and Td conditions and directly tested if QUASS calculation generated consistent results. We could not directly simulate in vivo CEST effects because there have been no agreed-upon gold standards for the number of labile groups and their exchange rates. It is foreseeable that the development of QUASS analysis could facilitate consensus of quantitative CEST analysis. The overt difference between the apparent and QUASS Z-spectra shows that the use of moderate Ts and Td impacts the measurement (e.g., MTR_asym) as well as $R_{ex}$ results. By minimizing the confounding effect of not sufficiently long saturation time and relaxation delay on the CEST measurement, our study provided additional insight into the contrast mechanism of APT MRI in acute stroke. This finding also suggests that T₁ contribution to the CEST measurement is not due to DWS Z-spectral difference (Figure 2b), but because of $R_{ex}^{DWS}$ and T₁ scaling (Figure 2d). The QUASS solution enables the use of well-established spin lock theory to analyze $R_{ex}$ spectra. Also, it is helpful to discuss the implication of QUASS calculation on the signal-to-noise ratio (SNR). The QUASS reconstruction recovers the magnitude of the CEST effect, but SNR is primarily determined by the original CEST signals. Fortunately, the QUASS reconstruction aids experimental optimization and quantification. This is because experimental conditions such as Ts, Td, and flip angle can be optimized for CNR per unit time without concerning the magnitude, hence the accuracy of the measurement, which can be reinstated from QUASS post-processing.

We utilized a 7-pool Lorentzian fitting to describe in vivo Z-spectrum reasonably well. The densely sampled Z-spectra revealed contributions from amine (3ppm) and NOE (−1.6 ppm) groups, often not observed below 7T. It is worth pointing out that multi-pool Lorentzian fitting depends on the Z-spectral resolution, SNR, RF saturation amplitude, and duration. Our study chose a weak B₁ saturation field of 0.75 μT, which is advantageous because it provides a relatively narrow Z-spectrum to resolve multiple labile proton groups that are somewhat overlapping. Because non-linear fitting is prone to interactions and hence, inaccuracies, we used experimentally measured relaxation rates to alleviate such interaction, so the relaxation changes in the ischemic lesion can be directly accounted for. It is helpful to discuss the contribution of multiple components in pH-weighted MRI contrast in acute stroke. There are significant decreases in semisolid MT, amide, and amine CEST effects with an increase in guanidyl CEST effect following acute stroke. Because MT is not sensitive to pH (66), MT change is likely due to edema, as evidenced by the T₁ increase. Both amide signal decrease (47) and guanidyl signal increase (67) have been associated with pH drop. Whereas neither MTR_asym nor $R_{ex}$ spectra show significant changes at 2 ppm, the Lorentzian fitting documented a clear guanidyl signal increase. This discrepancy was likely due to the amine signal drop, which has been postulated to be caused by a decrease in metabolite concentration (68). In addition, because the guanidyl CEST effect is relatively close to the bulk water resonance, it is susceptible to direct water saturation, which is complex in the presence of T₁ and T₂ changes. Fortunately, the bandwidth of the guanidyl CEST effect is moderate, which does not affect amide exchange at 3.5 ppm. Because amine protons have a much broader linewidth than the amide exchange, the amide signal benefits from a synergistic decrease in the amine CEST signal. This finding is consistent with ${\text{∆}R}_{ex}$ spectral change between ischemic and contralateral normal tissue, which peaked at 3.5 ppm. Future studies at ultrahigh fields will be needed to determine the origin of the observed amine signal in acute stroke.

It is helpful to discuss asymmetry, R_ex, and Lorentzian fitting analyses. Under the condition of long relaxation recovery and RF saturation, the commonly used asymmetry calculation effectively corrects symmetric signal changes, such as direct RF saturation when the RF saturation is of lower amplitude. It is not straightforward to correct T₁ contribution to the routine MT asymmetry calculation due to the relaxation cross terms ($Z\left(-\right)-Z\left(+\right)=\frac{R_1{cos}^2\theta \bullet \left(R_{ex}\left(+\right)-R_{ex}(-)\right)}{\left[R_1{cos}^2\theta +R_2{sin}^2\theta +R_{ex}(-)\right]\left[R_1{cos}^2\theta +R_2{sin}^2\theta +R_{ex}(+)\right]}$). In comparison, the inverse Z asymmetry calculation simplifies the T₁ correction ($\frac{1}{Z\left(+\right)}-\frac{1}{Z\left(-\right)}=\frac{R_{ex}\left(+\right)-R_{ex}(-)}{R_1{cos}^2\theta }$). Because the R_ex spectrum accounts for relaxation rate changes, it should be more reflective of the underlying CEST changes than the asymmetry calculation. However, when the relaxation time and saturation time are not long enough, the CEST signal has complex dependencies on experimental parameters and T₁. Under such conditions, the use of equilibrium or QUASS CEST effects should be considered. In addition, when multiple exchange groups are somewhat overlapping, the asymmetry calculation yields a mixed result. Therefore, although the asymmetry analysis provides a quick analysis, the Lorentzian fitting of the equilibrium/QUASS CEST Z-spectrum is recommended to provide a more accurate quantification of the underlying CEST system. For example, Lorentzian fitting revealed the guanidyl CEST signal change that was not pronounced in the asymmetry and R_ex spectral calculations due to contamination from multi-pool CEST effects. It is helpful to point out that a prior study by Zaiss et al. adopted a very long saturation time of 5s, which is likely close to the equilibrium CEST effects. They reported a signal decrease of about 1% between the ischemic and normal tissue. Note that $AREX\approx R_{ex}\left(3.5ppm\right)-R_{ex}\left(-3.5ppm\right)$. However, using three-point baseline estimation (43) is problematic at non-high fields because R_ex is prone to multi-pool contributions when the spectral resolution is not sufficiently high. Although it is not straightforward to compare two studies with different onset times after MCAO (hours vs. days), field strengths, and RF saturation schemes, the Lorentzian fitting showed that ischemia-induced APT signal change was −0.69±0.25 % (apparent) and −0.83±0.30% (QUASS), comparable to AREX calculation at high field with a long Ts.

Conclusion

Our study extended the QUASS algorithm to a modified fast multislice CEST MRI sequence with unevenly segmented RF saturation modules. QUASS CEST MRI showed enhanced pH-weighted MTR_asym contrast than the apparent CEST measurement after acute stroke. In addition, Lorentzian fitting documented significant decreases in MT and amide signals and an increase in guanidyl signal, with significantly elevated changes using QUASS analysis than the apparent CEST measurement.

Supplementary Material

supinfo

Figure S1. The fast multislice CEST MRI sequence. It includes a relaxation delay time (Td), a long primary saturation time (Ts1), followed by short secondary RF saturation times (Ts2) that is repeated between multislice spin echo EPI readout and signal averaging loop.

Figure S2. Multiparametric images from a representative acute stroke rat. a) Parametric T₁ map. b) Parametric T₂ map. c) Parametric ADC map. The ischemic lesion and the contralateral normal areas were outlined in black and green lines, respectively.

Figure S3. Comparison of apparent and QUASS Z-spectra. a) The apparent (red) and QUASS (black) Z-spectra from the contralateral normal area. b) The apparent (blue) and QUASS (black) Z-spectra from the ischemic lesion.