Fast correction of B₀ field inhomogeneity for pH-specific magnetization transfer and relaxation normalized amide proton transfer imaging of acute ischemic stroke without Z-spectrum

Abstract

Purpose:

The magnetization transfer and relaxation normalized amide proton transfer (MRAPT) analysis is promising to provide a highly pH-specific mapping of tissue acidosis, complementing commonly used CEST asymmetry analysis. We aimed to develop a fast B₀ inhomogeneity correction algorithm for acute stroke MRAPT imaging without Z-spectral interpolation.

Methods:

The proposed fast field inhomogeneity correction describes B₀ artifacts with linear regression. We compared the new algorithm with the routine interpolation correction approach in CEST imaging of a dual-pH phantom. The fast B₀ correction was further evaluated in amide proton transfer (APT) imaging of normal and acute stroke rats.

Results:

Our phantom data showed that the proposed fast B₀ inhomogeneity correction significantly improved pH MRI contrast, recovering over 80% of the pH MRI contrast to noise ratio (CNR) difference between the raw MTR_asym and that using the routine interpolation-based B₀ correction approach. In normal rat brains, the proposed fast B₀ correction improved pH-specific MRI uniformity across the intact tissue, with the ratio of magnetization transfer and relaxation normalized APT ratio (ΔMRAPTR) being 10% of that without B₀ inhomogeneity correction. In acute stroke rats, fast B₀ inhomogeneity corrected pH MRI reveals substantially improved pH lesion conspicuity, particularly in regions with non-negligible B₀ inhomogeneity. The pH MRI CNR between the ipsilateral diffusion lesion and contralateral normal tissue improved significantly with fast B₀ correction (from 1.88 ± 0.48 to 2.20 ± 0.44, P<0.01).

Conclusions:

Our study established an expedient B₀ inhomogeneity correction algorithm for fast pH imaging of acute ischemia.

Introduction

Chemical exchange saturation transfer (CEST) MRI provides a uniquely sensitive means to characterize microenvironment properties such as intracellular pH (1–5) and total amide proton content (6,7), which has been postulated to provide critical information in acute stroke (8–13), tumor (14–18) and body MRI applications (19–22). Tremendous progress has been achieved in optimization and quantification of CEST MRI effect (23–26). Specifically, the widely used magnetization transfer (MT) ratio asymmetry (MTR_asym) analysis provides a simplistic yet effective correction of concomitant RF saturation effect, in particular, direct RF saturation effect. Early work has shown that pH-weighed amide proton transfer (APT) imaging, a specific form of CEST MRI that probes amide proton exchange phenomenon, identifies metabolically disrupted infarction tissue (1). Building on that work, it has been shown that the use of a moderate RF saturation power level (i.e., 0.75 μT at 4.7 Tesla) maximizes pH-weighted MRI sensitivity (27,28), which not only capturers pH change in the infarcted ischemic core but also identifies mild acidic region prior to diffusion changes (i.e., metabolic penumbra) (29). The Lorentzian-based decoupling analysis revealed that MTR_asym is dominated by pH-dependent APT effect and the nuclear overhauser enhancement (NOE) signal is not strongly pH sensitive in focal and global ischemia rodent models (30,31). Similar findings have been reported under weak-moderate CEST RF irradiation levels at multiple field strengths (32,33). Nevertheless, pH-weighted in vivo MTR_asym index is prone to field inhomogeneity artifacts, in particular, B₀ inhomogeneity. The commonly used field inhomogeneity correction algorithms developed thus far have been based on interpolation of either a complete or a segment of CEST Z-spectrum/spectra, which can be time-consuming (34–37). For applications such as acute stroke imaging, the scan time has to be minimized in order not to impede the time-critical thrombectomy and/or thrombolysis therapy. As such, it is necessary to develop a fast B₀ inhomogeneity correction algorithm for in vivo CEST imaging without Z-spectrum acquisition.

It has been shown that B₀ inhomogeneity can be corrected in phantoms using an empirical solution that accounts for the direct saturation effect (38). However, it is challenging to translate this approach in vivo. It is so because, under the optimal B₁ saturation experimental condition for in vivo pH imaging, the routine MTR_asym image is heterogeneous across the intact white and grey matter (WM and GM) despite their little pH difference (39). This is likely arising from concomitant NOE and the slightly asymmetric semisolid macromolecular MT effect. Recently, we have shown that the magnetization transfer and relaxation-normalized APT (MRAPT) MRI can minimize the non-pH contrast in intact tissue (40), correction of which improved pH lesion conspicuity, enabling semiautomatic lesion segmentation and quantitative pH mapping (41). Capitalizing on the uniform intensity of pH-specific MRAPT image in the intact tissue, we here aimed to develop an expedient B₀ inhomogeneity correction algorithm for in vivo pH MRI without Z-spectrum. By accounting for concomitant relaxation, MT and NOE contributions that are not highly pH sensitive under the chosen experimental condition, the MRAPT analysis simplifies the description of in vivo APT effect from a multi-pool model (e.g. APT, NOE and MT) to a 2-pool (APT) exchange model. Building on this simplification, we derived the mathematical description of CEST effect dependence of B₀ inhomogeneity by modeling it using a polynomial function. The regression-based B₀ inhomogeneity correction was first validated in a dual pH creatine-gel phantom as a proof of concept. The algorithm was then applied in normal adult rat brains, showing significantly reduced pH variation across the intact normal tissue. We further evaluated the performance of the fast B₀ inhomogeneity correction in acute stroke rats, demonstrating significantly improved pH contrast to noise ratio (CNR) between the contralateral normal and ipsilateral ischemic tissues, over that obtained without B₀ inhomogeneity correction.

Theory

The 2-pool CEST MRI effect can be empirically described by the multiplication of the simplistic CEST effect and experimental factors of labeling coefficient and spillover factor (28). The CEST ratio (CESTR) can be shown to be equal to:

$$\mathit{\text{CESTR}}\approx \frac{f_s\bullet k_{sw}}{R_{1w}}\bullet \alpha \bullet \left(1-\sigma \right)$$ (1)

where f_s and k_sw are labile proton concentration and exchange rate, respectively, while α is the labeling coefficient and $\sigma$ is the spillover factor. It has been shown that in the presence of field inhomogeneity, the labeling coefficient is given by $\alpha =\frac{{\omega }_1^2}{{\omega }_1^2+p\bullet q\bullet \left(1+{\left(\frac{\Delta {\omega }_s}{p}\right)}^2\right)}\approx \frac{1}{1+\frac{p\bullet q}{{\omega }_1^2}\bullet \left(1+{\left(\frac{\Delta {\omega }_s}{p}\right)}^2\right)}$, where ${\omega }_1=2\pi \gamma B_1,p=r_{2s}-\frac{k_{sw}\bullet k_{ws}}{r_{2w}}$ , and $q=r_{1s}-\frac{k_{sw}\bullet k_{ws}}{r_{1w}}$, in which k_ws is the reverse exchange rate from the bulk water to labile protons, and Δω_s is field inhomogeneity (38). For field inhomogeneity less than r_2s, we have $\alpha \approx 1-\frac{p\bullet q}{{\omega }_1^2}\bullet \left(1+{\left(\frac{\Delta {\omega }_s}{p}\right)}^2\right)$. In addition, the B₀ inhomogeneity effect on the direct RF saturation can be approximated by (42)

$$\Delta I\approx \frac{4{\bullet \delta \omega }_s\bullet {\omega }_1^2{\bullet T}_{1w}\bullet T_{2w}^3}{{\left(1+{\delta \omega }_s^2{\bullet T}_{2w}^2+{\omega }_1^2{\bullet T}_{1w}{\bullet T}_{2w}\right)}^2}\bullet \Delta {\omega }_s$$ (2)

Therefore, the experimentally measured CEST effect, in the presence of field inhomogeneity, can be described as

$$\mathit{\text{CESTR}}\approx \frac{f_s\bullet k_{sw}\bullet \left(1-\sigma \right)}{R_{1w}}\bullet \left(1-\frac{p\bullet q}{{\omega }_1^2}\right)+\frac{4{\bullet \delta \omega }_s\bullet {\omega }_1^2{\bullet T}_{1w}\bullet T_{2w}^3}{{\left(1+{\delta \omega }_s^2{\bullet T}_{2w}^2+{\omega }_1^2{\bullet T}_{1w}{\bullet T}_{2w}\right)}^2}\bullet \Delta {\omega }_s-\frac{f_s\bullet k_{sw}\bullet \left(1-\sigma \right)}{R_{1w}}\bullet \frac{q}{{p\bullet \omega }_1^2}\bullet {\Delta {\omega }_s}^2$$ (3)

The B₀ inhomogeneity effect on spillover effect is described by the first-order correction term while the loss of saturation efficiency was corrected as the second-order term.

Whereas CESTR is equivalent to MTR_asym for a simple 2-pool exchange model, in vivo MTR_asym is complex and susceptible to non-pH dependent concomitant semisolid MT and NOE effects. The pH-specific MRAPT analysis accounts for such baseline heterogeneity, which effectively reduces the multi-pool in vivo exchange model (APT, NOE, and MR) to a 2-pool (APT) model (40,41). Briefly, we have

$$R_{1w}\bullet {MTR}_{\mathit{\text{asym}}}=f_s\bullet k_{sw}\bullet \alpha \bullet \left(1-\sigma \right)+{R_{1w}\bullet MTR}_{\mathit{\text{asym}}}^{\text{'}}$$ (4)

where ${\text{MTR}}_{\text{asym}}^{\text{'}}$ is non-pH dependent concomitant RF irradiation effects. The ${\text{R}}_{1\text{w}}\bullet {\text{MTR}}_{\text{asym}}$ heterogeneity in the intact tissue can be described by a regression function F(R_1w, MMTR). The ΔMRAPTR takes the difference between experimentally measured R_1w-scaled MTR_asym and the baseline heterogeneity map estimated from the intact tissue to improve its pH specificity, being

$$\Delta \mathit{\text{MRAPTR}}=R_{1w}{\bullet MTR}_{\mathit{\text{asym}}}-F\left(R_{1w},\mathit{\text{MMTR}}\right)$$ (5)

where MMTR is the mean MTR at ±3.5 ppm. Because the semisolid macromolecular MT effect has a relatively broad linewidth, the field inhomogeneity effect on MT can be reasonably described by a first-order correction term as well. Note that under the chosen experimental conditions, the experimental factor (i.e., $\alpha \bullet \left(1-\sigma \right))$ has been shown to be fairly uniform and nearly 100%, without B₀ field inhomogeneity (5). Altogether, the B₀ inhomogeneity artifact can be generally described by a polynomial function that accounts for the modulation of experimental factors and a baseline shift due to misaligned label and reference scans, as

$$\Delta \text{MRAPTR}\approx {\text{f}}_{\text{s}}\bullet \Delta {\text{k}}_{\text{sw}}+\left[C_1\bullet \Delta {\omega }_s+C_2\bullet {\Delta \omega }_s^2\right]$$ (6)

where Δk_sw is the exchange rate change from that of the intact tissue, C₁ and C₂ are coefficients determined from polynomial regression, in the units of inverse $\Delta {\omega }_s$ and ${\Delta \omega }_s^2$, respectively.

Methods

Phantom

We prepared a dual-pH creatine agarose gel phantom, following the protocol published previously (43). Briefly, 1% (w/w) low-melt temperature agarose was added to deionized water, heated to boiling and then immersed in a water bath set at 50°C. After the temperature settles, creatine and gadoteridol (Bracco Diagnostics, Monroe Township, NJ) were added into the solution and reach concentrations of 50 mM and 30 μM, respectively. The solution pH was titrated to 6.5 and 6.0 and transferred to two separate compartments of a dual pH phantom holder. The phantom solidified at room temperature before MRI.

Animals

Animal experiments were approved by the institutional animal care and use committee, Massachusetts General Hospital (IACUC, MGH). Anesthesia was initially induced with 5% isoflurane in air and then animals were maintained under anesthesia with 1.5–2.0% isoflurane in air. Heart rate and peripheral capillary oxygen saturation were monitored online (Nonin Pulse Oximeter 8600, Plymouth, MN), and body temperature was maintained by a circulating warm water jacket. Five normal adult male Wistar rats were imaged to assess the typical field inhomogeneity and evaluate the proposed fast B₀ correction algorithm following the recently developed pH specific MRAPT MRI. In addition, ten adult male Wistar rats underwent MRI one hour after middle cerebral artery occlusion (MCAO) surgery. Briefly, permanent MCAO was induced in rats with a silicone-coated 4–0 nylon filament, and rats underwent MRI about 1 hour after stroke induction. One stroke rat showed grossly low whole-brain apparent diffusion coefficient (ADC), likely due to a malfunction of the diffusion MRI sequence. This animal was excluded from data analysis.

MRI

Phantom experiments were performed using a 7 Tesla small-bore MRI scanner (Bruker Biospin, Billerica, MA). We collected a single slice, single shot, spin echo (SE), echo planner imaging (EPI, bandwidth = 265 kHz). The slice thickness was 10 mm and the field of view (FOV) was 52× 52 mm (matrix size = 96× 96). We chose an off-centered slice (−10 mm) to subject the CEST experiments to a condition of non-negligible B₀ inhomogeneity. We centered the bulk water resonance frequency without field mapping and shimming adjustment. The water saturation shift referencing (WASSR) experiment was conducted to assess the field inhomogeneity (B₁=0.5 μT, ±0.5 ppm with intervals of 0.025 ppm, repetition time (TR)/saturation time (TS)/echo time (TE)= 2,000/500/40 ms, one average). We acquired CEST Z-spectrum (B₁=1 μT, ±6 ppm with intervals of 0.05 ppm, TR/TS/TE=6,000/3,000/40 ms, one average).

In vivo scans were performed using a 4.7 Tesla small-bore MRI scanner (Bruker Biospin, Billerica, MA). Multi-slice MRI (5 slices, slice thickness/gap=1.8/0.2 mm, the field of view=25 × 25 mm², image matrix=64 × 64) was acquired with single-shot SE EPI (bandwidth= 179 kHz). We collected multiparametric T₁, T₂, perfusion, APT and diffusion MRI. Specifically, T₁-weighted MRI was acquired with inversion recovery EPI of seven inversion delays from 250 to 2,750 ms (recover time/TE = 6,000/30 ms, 4 averages, scan time = 3.5 min), and T₂-weighted EPI was obtained with two separate spin echo EPI images with TE of 30 and 100 ms (TR = 3,000 ms, 8 averages; scan time = 2 min). We used the amplitude-modulated continuous arterial spin labeling (AM-CASL) MRI (TR/TE = 5,000/30 ms). The ASL tagging pulse duration and amplitude were 2500 ms and 4.7 μT, respectively (16 averages, scan time ~ 4.5 min) (44). pH-sensitive APT scans were acquired with fast unevenly segmented RF irradiated CEST MRI (45). We used a recovery time of 3,000 ms, the primary RF saturation duration of 3,000 ms, and secondary RF saturation duration of 500 ms between slices for an RF irradiation amplitude of 0.75 μT applied at ±3.5 ppm. The unsaturated control scan was averaged 8 times, while the saturated images were averaged 32 times (total scan time = 4 min). Moreover, a single-shot isotropic diffusion-weighted EPI was performed (b-values = 250/1,000 s/mm², TR/TE = 3,000/79 ms, 16 averages, scan time = 1.5 min) (46). _ENREF_12

Data Analysis

Quantitative T₁ map was derived by fitting the T₁-weighted MRI signal as a function of the inversion time ($I_i=I_0\left|1-\left(1-\eta \right)e^{-{TI}_i/T_1}\right|$) per pixel, where η is the inversion efficiency and TI_i is the i^th inversion time. The parametric T₂ and apparent diffusion coefficient (ADC) maps were calculated using $T_2=\frac{\Delta TE}{ln\left(I\left({TE}_1\right)/I\left({TE}_2\right)\right)}$ and $ADC=\frac{ln\left(I\left(b_1\right)/I\left(b_2\right)\right)}{\Delta \text{b}}$, where TE_1,2 and b_1,2 are two TEs and diffusion b values, with ΔTE and Δb being their differences, respectively. The cerebral blood flow (CBF) was quantified using $CBF=\frac{\lambda \bullet \left(I_{ref}-I_{tag}\right)}{2\alpha \bullet I_{ref}}\bullet \frac{e^{w/T_{1a}}}{T_{1w}}$, where I_tag and I_ref are the label and the reference image, respectively, λ is the brain-blood partition coefficient for water, α is the degree of inversion with transient time correction, w is the post-labeling delay, and T_1a is the arterial blood longitudinal relaxation time. The relative CBF (rCBF) was calculated by normalizing the CBF map with the mean CBF calculated from the contralateral normal tissue. The magnetization transfer ratio (MTR) map was solved using $\text{MTR}\left(\pm 3.5\text{ppm}\right)=1-\frac{\text{I}\left(\pm 3.5\text{ppm}\right)}{{\text{I}}_0}$, where I₀ is the control image without RF irradiation and I(±3.5 ppm) are the label and reference images with RF irradiation applied at ±3.5 ppm, respectively. In addition, the mean MTR (MMTR) was the average of MTRs at ±3.5 ppm. Moreover, pH-weighted MTR_asym image was calculated as ${\text{MTR}}_{\mathit{\text{asym}}}=\frac{\text{I}\left(-3.5\text{ppm}\right)-\text{I}\left(+3.5\text{ppm}\right)}{{\text{I}}_0}$. The pH-specific ΔMRAPTR image was calculated as the difference (i.e. ΔMRAPTR= R_1w*MTR_asym – MRAPTR) between the experimentally measured R_1w-normalized MTR_asym and that estimated from the regression analysis of the contralateral normal tissue (40). Because ΔMRAPTR removes the baseline pH-sensitive MRI heterogeneity in the intact tissue not related to pH, it is of substantially improved pH specificity. In addition, pH MRI was derived from ΔMRAPTR using the calibration published previously (i.e., $\text{pH}={\text{pH}}_{\text{norm}}+\text{log}10\left(1+\frac{\Delta \text{MRAPTR}}{{\text{C}}_0}\right)/{\text{C}}_1$, where pH_norm=7.05 with C₀ and C₁ being 5.04 and 0.25, respectively) (41). Moreover, perfusion, pH, and diffusion ischemic lesions were segmented using a K-means clustering-based algorithm, as shown in our prior study (47). All values were reported in mean ± standard deviations (SD), and P values less than 0.05 were regarded as statistically significant.

Results

Fig. 1 evaluated the proposed fast B₀ inhomogeneity correction using a dual-pH creatine gel phantom. The raw MTR_asym image was calculated from the CEST Z spectrum (Fig. 1a), which shows prominent B₀ inhomogeneity artifacts. This is because we only centered the resonance frequency without shimming the magnetic field, resulting in gross B₀ field variation across the slice (Fig. 1b). Fig. 1c shows the correlation between field inhomogeneity and the apparent MTR_asym per pixel (markers + for pH=6.5 and × for pH=6.0). The regression model shows a satisfactory description of the B₀ inhomogeneity, with ${\text{MTR}}_{\text{asym}}\left(\%\right)=5.5+0.636\bullet {\text{B}}_0-1.3{\text{e}}^{-3}\bullet {\text{B}}_0^2$ (R²=0.90, P<0.001) and ${\text{MTR}}_{\text{asym}}\left(\%\right)=2.3+0.367\bullet {\text{B}}_0-6.7{\text{e}}^{-5}\bullet {\text{B}}_0^2$ (R²=0.95, P<0.001) for pH of 6.5 and 6.0, respectively. The fast B₀ corrected MTR_asym map was reconstructed by adjusting for numerically determined B₀ inhomogeneity regression shift, per pixel (i.e., MTR_asym(fast), Fig. 1d). In addition, the routine interpolation-corrected MTR_asym was shown in Fig. 1e (MTR_asym(intrpl)). Fig. 1f shows the correlation between interpolation-corrected and regression-corrected MTR_asym, per pixel (R²=0.956, P<0.001). The CNR between the two pH compartments was found to be 3.59, 14.45 and 12.56 for MTR_asym without B₀ correction, interpolation-based Z-spectral correction, and fast regression-based correction, respectively (Table 1). This is equivalent to recovering over 80% of the pH MRI CNR lost in the raw MTR_asym due to B₀ field inhomogeneity. The phantom study demonstrates that for slow and intermediate chemical exchange, the typical field inhomogeneity artifact can be reasonably modeled by polynomial regression, and therefore, corrected without Z-spectrum-based interpolation.

Fig. 1.

Evaluation of fast B₀ inhomogeneity correction in a dual pH phantom. a) The raw MTR_asym calculated from asymmetry analysis. b) B₀ inhomogeneity map determined from the WASSR scan.c) Regression analysis between B₀ inhomogeneity and raw MTR_asym, per pixel. d) Fast B₀ inhomogeneity-corrected MTR_asym(fast) map. e) Interpolation-based B₀ inhomogeneity-corrected MTR_asym(intrpl) map. f) Regression analysis between MTR_asym(fast) and MTR_asym(intrpl), per pixel.

Table 1.

Comparison of CEST MRI of two pH compartments and their CNR determined from the raw MTR_asym, interpolation-based correction of B₀ inhomogeneity using Z-spectrum (MTR_asym(intrpl)), and the fast regression correction algorithm of MTR_asym without Z-spectrum ((MTR_asym(fast)).

	Raw MTRasym	MTRasym(intrpl)	MTRasym(fast)
CESTRint (%)	5.42 ± 0.81	5.49 ± 0.22	5.50 ± 0.26
CESTRext (%)	1.92 ± 1.12	2.48 ± 0.20	2.28 ± 0.26
CNR	3.59	14.45	12.56

We evaluated the fast B₀ inhomogeneity correction in normal adult rats. Fig. 2a shows the MMTR(±3.5ppm) map. The routine MTR_asym map is shown in Fig. 2b, which displays typical WM/GM heterogeneity. The pH-specific MRAPTR map (Fig. 2c) and absolute tissue pH (Fig. 2d) maps show the variation that closely resembles B₀ field inhomogeneity (Fig. 2e), determined from the WASSR scan. The superior brain has noticeably negative B₀ shift while the inferior brain displays a slightly positive B₀ shift. Across all five slices for 5 normal animals, we found that B₀ field inhomogeneity was −32.3 ± 26.2 Hz. The relationship between pH-specific ΔMRAPTR and B₀ field inhomogeneity per pixel can be described by linear regression (R²=0.55, P<0.001, Fig. 2f), correction of which reduces ΔMRAPTR variation due to B₀ field inhomogeneity (Fig. 2g). Specifically, ΔMRAPTR was found to be −0.026 ± 0.008 %/s without B₀ correction while the fast B₀ inhomogeneity correction reduced it to 0.001 ± 0.003 %/s. The absolute magnitude of ΔMRAPTR significantly reduced (0.003 ± 0.002 vs. 0.026 ± 0.007 %/s, P<0.005, paired t-test) with fast B₀ inhomogeneity correction. The ratio of ΔMRAPTR with and without B₀ inhomogeneity correction was 10.3 ± 6.5 % (P<0.001, one-sample t-test). Similarly, pH map with B₀ inhomogeneity correction (Fig. 2h) showed significantly improved uniformity across the brain (Fig. 2d vs. Fig. 2h). The tissue pH was found to be 7.04 ± 0.04 and 7.05 ± 0.01, without and with B₀ field inhomogeneity correction, respectively. Notably, B₀ field inhomogeneity correction reduced pH variation with a significantly improved coefficient of variation (COV, 2.4 ± 0.16% vs. 1.9 ± 0.12%, P<0.001, paired t-test).

Fig. 2.

Evaluation of fast B₀ inhomogeneity correction in a representative normal adult rat brain. a) Mean MTR (MMTR) at ±3.5 ppm. b) MTR_asym map. c) ΔMRAPTR map without B₀ inhomogeneity correction. d) pH map determined from ΔMRAPTR map without B₀ inhomogeneity correction. e) B₀ inhomogeneity map determined from the WASSR scan. f) Regression analysis between B₀ inhomogeneity and raw ΔMRAPTR, per pixel. g) ΔMRAPTR map with the proposed fast B₀ inhomogeneity correction. h) pH map determined from ΔMRAPTR map with fast B₀ inhomogeneity correction.

We further tested the fast B₀ correction algorithm in acute stroke rats. Fig. 3a shows the raw pH-specific ΔMRAPTR image without B₀ field correction. The WASSR field map revealed moderate field inhomogeneity (Fig. 3b), with bilateral positive B₀ field drift while the caudate putamen region displayed a slight negative B₀ field shift. Fig. 3c shows the regression between the raw ΔMRAPTR and B₀ field inhomogeneity, per pixel (R²=0.36, P<0.001). The ΔMRAPTR shift increases with B₀ shift. As such, regions with positive B₀ field shift is subject to an inflation of ΔMRAPTR, correction of which results in more uniform ΔMRAPTR in the intact tissue (Fig. 3d). In addition, the ipsilateral ischemic region had a noticeable positive B₀ field shift, correction of which improved pH lesion conspicuity in the primary cortices and amygdala. Indeed, pH CNR between the diffusion lesion and the contralateral normal tissue was 1.88 ± 0.48 and 2.20 ± 0.44, without and with the proposed B₀ inhomogeneity correction, respectively (P<0.01, paired t-test). The significant improvement in CNR and conspicuity of pH-specific ΔMRAPTR demonstrates the effectiveness of the proposed fast B₀ inhomogeneity correction in vivo.

Fig. 3.

Evaluation of fast B₀ inhomogeneity correction in a representative acute stroke rat. a) ΔMRAPTR map without B₀ inhomogeneity correction. b) B₀ inhomogeneity map determined from the WASSR scan. c) Regression analysis between B₀ inhomogeneity and raw ΔMRAPTR, per pixel. d) ΔMRAPTR map with fast B₀ inhomogeneity correction.

The multi-parametric perfusion, pH (with B₀ field inhomogeneity correction) and diffusion images were shown in Figs. 4a, 4b, and 4c, respectively, with their lesion volumes being 334 ± 40, 211 ± 47, and 140 ± 75 mm³, respectively. The multi-parametric perfusion, pH and diffusion lesions were superimposed in Fig. 4d, with significant mismatches among lesion (P<0.05, one-way ANOVA with Tukey’s multiple comparison test). The contralateral normal regions showed no significant difference in rCBF, pH, and ADC, as expected (Table 2). All three ischemic lesions (i.e., perfusion, pH, and diffusion lesions) showed a significant rCBF drop from the contralateral normal tissue while there was no significant rCBF difference among perfusion, pH and diffusion lesions. All ischemic lesions had a significant ADC drop from that of the contralateral normal tissue. Perfusion and pH lesions had significantly higher ADC than the diffusion lesion without significant difference between perfusion and pH lesions. For pH, the ischemic lesions showed a significant drop from the contralateral normal tissue. Due to the noticeable overlap between diffusion and pH lesions, there was no significant difference in the magnitude of pH change between them.

Fig. 4.

Multi-parametric MRI of a representative acute stroke rat. a) CBF map. b) pH map determined from ΔMRAPTR image with the proposed fast B₀ inhomogeneity correction. c) ADC map. d) Multi-parametric image lesion mismatch overlaid on a diffusion-weighted image (DWI).

Table 2.

Multi-parametric MRI of acute stroke rats showing heterogeneous perfusion, pH, and diffusion MRI. There is a significant difference among perfusion, pH and diffusion lesion volumes. The rCBF, pH and ADC values (mean ± standard deviations) were reported in the ipsilateral ischemic lesions and the contralateral normal areas.

	Lesion volume (mm3)	rCBF		pH		ADC
		Contra normal	Ipsi ischemic	Contra normal	Ipsi ischemic	Contra normal	Ipsi ischemic
rCBF	334 ± 40	0.92 ± 0.04	0.24 ± 0.09	0.99 ± 0.09	0.28 ± 0.14	1.08 ± 0.10	0.35 ± 0.19
pH	211 ± 47	7.03 ± 0.01	6.83 ± 0.06	7.04 ± 0.01	6.68 ± 0.08	7.07 ± 0.03	6.71 ± 0.08
ADC (μm2/ms)	140 ± 75	0.85 ± 0.02	0.76 ± 0.06	0.85 ± 0.02	0.71 ± 0.07	0.85 ± 0.02	0.63 ± 0.04

Discussion

Our study here developed a simplified B₀ inhomogeneity correction algorithm for pH-specific CEST effect without Z-spectrum and demonstrated its effectiveness both in a phantom and in vivo. Because the fast B₀ inhomogeneity correction only requires CEST scans at the label and reference frequency offsets (i.e., ±3.5 ppm), we were able to average them extensively in vivo (i.e., 32 averages), which boosted its SNR. For the phantom experiments at 7 Tesla, we centered the resonance frequency without shimming. For our in vivo scans at 4.7 Tesla, we used a gradient coil insert that does not have 2^nd order shimming adjustment capability (B-GA12, Bruker Biospin, Billerica, MA). Under both difficult shimming conditions, we demonstrated that the B₀ inhomogeneity can be reasonably corrected in the absence of Z-spectrum. Although amide T_2s was not measured in our study, the typical brain metabolite T_2s has been reported to range from 7 to 23 ms (48). This translates to an estimated r_2s of about 66 Hz. It is worth noting that typical field inhomogeneity, after shimming, shall be well within this range. In addition, our study provided compelling evidence that it is feasible to achieve high-quality CEST images utilizing post-processing B₀ correction without resorting to high order gradient shimming. Because the field mapping and high order gradient shimming adjustment typically take a few minutes, the proposed post-processing B₀ correction is of advantage if the tedious high order shimming procedure can be omitted, which needs to be investigated upon clinical translation.

The proposed fast B₀ inhomogeneity correction approach builds on pH-specific MRAPT MRI that nulls the intrinsic MTR_asym heterogeneity in the intact tissue not related to tissue pH change so that the deviation in ΔMRAPTR in the intact tissue can be attributed to B₀ inhomogeneity (40). It has been shown that the MTR_asym contrast between the intact WM and GM can also be minimized by adjusting RF saturation power (49). We posit that under such conditions, the proposed fast B₀ inhomogeneity correction may be applicable. Alternatively, the brain tissue can be segmented (e.g., WM and GM) and regression-based B₀ inhomogeneity correction is performed for WM and GM independently. Because we used a volume transmit coil, which has a homogeneous B₁ profile within 5% for the field of view of the rat brain and phantom, our study here did not consider B₁ inhomogeneity. The amide proton exchange has been estimated to be 30 and 10 s⁻¹, in normal and ischemic tissue, respectively, and ischemia-induced reduction in exchange rate shall be no more than 20 s⁻¹ (1). Therefore, there shall be very little B₁ inhomogeneity artifacts (50). Moreover, it has been shown that the MTR_asym remains relatively constant around the optimal RF power level (27). Nevertheless, we postulate that the regression-based field inhomogeneity correction may be extended to correct for minor B₁ field inhomogeneity without requiring CEST scans obtained under different B₁ saturation levels.

Our current work used the standard WASSR approach to determine the B₀ inhomogeneity map. The field map can also be obtained from dual-echo GRE method (51). The conventional B₀ inhomogeneity correction algorithm requires either a complete or a segment of Z-spectrum, based on which the signal intensity is shifted and interpolated, per pixel. As such, the scan time is unavoidably prolonged, making it challenging to implement in the acute stroke setting where the imaging time has to be minimized so timely intervention can be performed. In the context of scan time, the proposed B₀ inhomogeneity approach is fast because it does not require the Z-spectral acquisition. To illustrate the proposed B₀ correction in vivo, our study first derived ΔMRAPTR without B₀ correction and then demonstrated the correlation between ΔMRAPTR and B₀ inhomogeneity. The generalized approach shall directly include B₀ inhomogeneity into the MRAPT analysis so that MTR_asym is regressed again R_1w, MMTR, and B₀ inhomogeneity (i.e., $\Delta \text{MRAPTR}={\text{R}}_{1\text{w}}{\bullet \text{MTR}}_{\text{asym}}-\text{F}\left({\text{R}}_{1\text{w}},\text{MMTR},{\text{B}}_0\right)$ for correction (Supporting Information Figure S1). Our study succeeded in achieving reasonable B₀ inhomogeneity correction in acute stroke imaging without Z-spectrum. Because ΔMRAPTR is zero for the intact tissue, the C₀ term from the intact tissue is negligible, and the first and second-order correction terms were determined from the contralateral intact tissue. One limitation of the proposed approach is the second-order correction is estimated from the exchange rate of the intact tissue. This may slightly overcompensate the B₀ inhomogeneity effect in the ischemic lesion. One potential solution is to solve the exchange rate from B₀ inhomogeneity-corrected ΔMRAPTR and attenuate the second-order correction based on the exchange rate per pixel, which can be iterated to improve the correction. Fortunately, for field inhomogeneity less than r_2s (e.g, about 50 Hz), the magnitude of the second-order correction shall be dominated by the first-order correction. Indeed, the pH CNR between the diffusion lesion and the contralateral normal tissue significantly improved with the proposed B₀ inhomogeneity correction. pH MRI, in the absence of B₀ correction, may miss the ischemic region with positive field shift (Fig. 3). Because the used gradient coil did not have high order shims, the pH lesion without B₀ inhomogeneity correction was underestimated, and the pH/diffusion lesion mismatch became nonsignificant. The proposed B₀ field inhomogeneity recovers the missed ischemic lesion from 140 ± 75 to 211 ± 47 mm³ (P<0.01), revealing the pH/diffusion lesion mismatch (P<0.05). Nevertheless, fast B₀ inhomogeneity correction without Z-spectrum is not trivial. Future experiments that improve shimming, enhance the SNR, and incorporate relaxation and MT images to the fast B₀ inhomogeneity correction are urgently needed to improve pH imaging.

Because the CEST MRI effect decreases following T₁ relaxation after the RF saturation (52), our in vivo study used the fast CEST MRI with segmented RF saturation scheme (45). The pulse sequence was designed so that the CEST MRI steady state is reached after the primary RF saturation while the secondary short RF saturation retains the steady state for efficient signal averaging and multi-slice acquisition. Whereas the initial implementation of fast CEST MRI used gradient echo (GE) EPI readout with an excitation angle of 75 degrees, our current study used spin echo readout to minimize the EPI distortion due to the absence of high order shimming. As such, the magnitude of fast CEST SE EPI may be somewhat reduced because of the timing of finite secondary labeling RF pulse. Specifically, the magnetization recovers from null after the spin echo readout, which requires longer secondary saturation duration in order to reach the same CEST MRI steady state as that using GE EPI readout. As such, pH drop calculated from fast CEST SE EPI using the calibration standard established from GE CEST MRI may be slightly underestimated. It is feasible to correct the small pH scaling factor by accounting for the finite secondary saturation duration with concurrent fast CEST acquisition using both SE and GE EPI readout. Nevertheless, the significant improvement in homogenizing ΔMRAPTR and pH in the intact tissue undoubtedly demonstrates the advantage of the proposed fast B₀ inhomogeneity correction. It is worth mentioning that one of the challenges for whole-brain pH MRI was how to acquire multi-slice CEST images and correct for field inhomogeneity in a timely manner. The combined use of the unevenly segmented RF saturation scheme and the proposed fast B₀ inhomogeneity correction makes it feasible to attempt whole-brain quantitative CEST MRI, which shall further facilitate its in vivo adoption.

Conclusion

Our study demonstrated a fast B₀ inhomogeneity correction approach for in vivo pH-specific imaging. We confirmed its usefulness in normal rat brains, showing significantly more uniform pH. The acute stroke MRI results showed significant enhancement in pH contrast to noise ratio between the ipsilateral diffusion lesion and the contralateral normal tissue. The proposed fast B₀ correction algorithm is simple to use yet it provides satisfactory correction of typical in vivo B₀ inhomogeneity.

Supplementary Material

fS1

Figure S1. Flow chart of data processing steps of the proposed fast B₀ inhomogeneity corrected MRAPT analysis.