Detection of tissue pH with quantitative chemical exchange saturation transfer MRI

Abstract

Chemical exchange saturation transfer (CEST) MRI has emerged as a novel means for sensitive detection of dilute labile protons and chemical exchange rates. By sensitizing to pH-dependent chemical exchange, CEST MRI has shown promising results in monitoring tissue statuses such as pH changes in disorders like acute stroke, tumor, and acute kidney injury. This article briefly reviews the basic principles for CEST imaging and quantitative measures, from the simplistic asymmetry analysis to multipool Lorentzian decoupling and quasi-steady-state (QUASS) reconstruction. In particular, the advantages and limitations of commonly used quantitative approaches for CEST applications are discussed.

Graphical Abstract

Transition from pH-weighted MTR_asym image (A) and tissue pH map reconstructed from pH-specific MRAPT analysis (B).

Introduction

pH is a physiological index tightly regulated in healthy tissue (1,2). A notable pH change often signals altered tissue states, such as those following acute ischemia (3-8) and the development of cancer (9-15). For example, tissue acidosis is associated with anaerobic glycolysis following acute stroke (16-19). pH drop compromises essential ATP-dependent functions, often leading to cell death and ultimately tissue infarction (20-24). In the case of cancers, excessive amounts of lactate are produced, disrupting pH homeostasis, a hallmark of the Warburg effect (25-27). It has been recognized that pH change profoundly affects tumor growth and metastasis (28). As such, pH measurement may provide novel insights into tissue microenvironment and response to therapy, and potentially facilitate the development of new therapeutics.

pH measurement in vivo, however, is not straightforward because common strategies such as pH immunohistology, fluorescence microscopy, or phosphorus magnetic resonance approaches are either invasive (29-33), of limited penetration depth (34,35), or insensitive (36-38). To address such an unmet biomedical need, Balaban and co-workers introduced chemical exchange saturation transfer (CEST) MRI as a new sensitive means of imaging pH and dilute labile protons (39-42), laying the groundwork for CEST MRI research over the following two decades (43-48). The proton exchange between macromolecule backbone labile (e.g., amide, amine, and guanidyl groups) and solvent protons is highly pH-dependent, often presenting as a U-shaped curve for acid-base catalysis and conferring pH sensitivity to CEST MRI (49). The typical peptide proton exchange ranges from the low teens for amide protons to a few thousand for amine and hydroxyl groups, depending on the molecule size and exchangeable groups (50). When the exchange rate between two proton groups is much lower than their chemical shift difference, it is considered slow exchange (51). The slow exchange process is particularly amenable to CEST MRI because the saturation of labile proton can be isolated from the solvent proton signal, resulting in efficient saturation of labile protons and their transfer to the bulk water for induction of the CEST effect. CEST MRI has been increasingly adopted to measure certain key metabolites such as glutamate, glutamine, creatine, ensemble amide protons from mobile proteins and peptides, and intracellular pH (52-59). Although the endogenous CEST signal has been attributed mainly to the intracellular compartment, the development of exogenous CEST contrast agents enables imaging of the extracellular space, complementing the endogenous intracellular imaging. Therefore, it is foreseeable that pH imaging (intracellular and extracellular) is beneficial, particularly in tumor imaging (60,61).

CEST MRI experiments are complex, highly dependent on the scan conditions and post-processing analysis. Despite the substantial progress the field has achieved in the theoretical framework of the CEST MRI quantification, many different CEST indices have been adopted. For example, the simplistic magnetization transfer ratio asymmetry analysis has been commonly used despite its limitation in the presence of multiple exchangeable groups. On the other hand, the development of Lorentzian and model-based fittings can account for multiple partially overlapping CEST effects. Such analyses are time-consuming, however, and often limited by image sensitivity. Where quantitative analyses such as omega plots and ratiometric approaches are feasible in contrast agent-based applications, their in vivo implementation requires thorough optimization and Lorentzian fitting. We aim to provide a concise overview of commonly used CEST MRI quantification methods and discuss their advantages and limitations.

Quantitative Description of CEST MRI Signal

2-pool CEST solution

CEST MRI measurement depends not only on the concentration and exchange properties of labile protons, but also on experimental conditions and relaxation constants (62-64). The CEST MRI phenomenon can be mathematically described using Bloch-McConnell equations (65-67). However, the numerical nature of the Bloch-McConnell equations provides limited insights into the contrast mechanism. To address this, analytical CEST MRI solutions, albeit only approximations, have been developed. For a classical 2-pool exchange model under the condition of long RF saturation, the equilibrium CEST effect can be approximated by (68-70)

$$CESTR\approx \frac{f_r\cdot k_{sw}}{R_{1w}}\cdot \alpha \cdot (1-\sigma )$$ (1)

in which CESTR stands for CEST ratio, k_sw and f_r are the labile proton exchange rate and fraction ratio, and R_1w is the bulk water longitudinal relaxation rate, with α and σ being two experimental factors. Briefly, α is the labeling coefficient that accounts for the saturation efficiency of labile protons (71). σ is the RF spillover factor, which describes the concomitant direct RF saturation of the bulk water signal that competes with the CEST-specific MRI effect (72). The product of the experimental factors needs to be optimized to maximize the measurable CEST MRI effect. Overall, the 2-pool CEST MRI solution (Eq. 1) depicts a balance between bulk water signal decrease caused by its exchange with saturated labile protons and its signal recovery due to its T₁ relaxation.

Spinlock-based multipool CEST solution

Although the multipool CEST MRI phenomenon can be modeled by extending Bloch-McConnell equations, it quickly becomes tedious for more than three labile proton groups (73,74). The multipool CEST asymmetry effect can be described by superimposing multiple 2-pool models, provided that each labile proton group is sufficiently dilute so the interaction term can be accounted for (75). Fortunately, the introduction of the spinlock theorem to the CEST MRI field enormously simplified the modeling of the CEST effect (76-78). Briefly, the Z-spectral intensity is described by

$$\frac{S_{sat}}{S_0}=e^{-R_{1\rho }\cdot Ts}+\frac{R_{1w}}{R_{1\rho }}co{s}^2\theta \cdot (1-e^{-R_{1\rho }\cdot Ts})$$ (2)

where Ts is the RF saturation time, R_1ρ is the spinlock relaxation rate under the RF saturation, being $R_{1\rho }=R_{1w}co{s}^2\theta +R_{2w}si{n}^2\theta +{\sum }_{i=1}^N{R}_{ex}^i$, with $R_{ex}^i$ being the ith pool CEST effect, $\theta =ta{n}^{-1}({\omega }_1∕\Delta \omega )$, in which ${\omega }_1$ and $\Delta \omega$ are the RF saturation amplitude and offset, in radian, respectively. The spinlock formula has been extended to describe magnetization transfer (MT) ratio asymmetry (MTR_asym), a commonly used metric to quantify CEST effects (79), and the pulse RF saturation scheme (80-82).

Generalized CEST solution

It is worth mentioning that Eq. 2 is accurate under the condition of complete relaxation recovery, which may not be fulfilled when the repetition time is shortened to reduce the overall scan time. Jiang et al. extended the CEST solution to incorporate experimental parameters such as flip angle (FA), relaxation delay (Td) and Ts (83). The Z-spectral intensity for CEST MRI with gradient-echo (GE) readout can be shown as

$$\frac{S_{sat}(R_{1\rho }(B_1,\Delta \omega ),TR,FA,Ts)}{S_0(TR,FA)}=\frac{\frac{\left(1-e^{-Td∕T_{1w}}\right)\cdot e^{-R_{1\rho }(B_1,\Delta \omega )\cdot Ts}+\frac{R_{1w}co{s}^2\theta }{R_{1\rho }(B_1,\Delta \omega )}\cdot \left(1-e^{-R_{1\rho }(B_1,\Delta \omega )\cdot Ts}\right)}{1-cos(FA)\cdot e^{-Td∕T_{1w}}\cdot e^{-R_{1\rho }\cdot Ts}}}{(1-e^{-TR∕T_{1w}})∕(1-cos(FA)\cdot e^{-TR∕T_{1w}})}$$ (3)

This generalized formula facilitates scan optimization and quantification.

Recently, Sun proposed quasi-steady-state (QUASS) CEST reconstruction, which solves the equilibrium $R_{1\rho }$ (i.e., $R_{1\rho }^{QUASS}(B_1,\Delta \omega )$) from the experimental CEST measurement under the condition of finite saturation time and relaxation delay (84). The equilibrium CEST effect can be reconstructed as

$${\left(\frac{S_{sat}}{S_0}\right)}^{QUASS}=\frac{R_{1w}}{R_{1\rho }^{QUASS}(B_1,\Delta \omega )}co{s}^2\theta$$ (4)

One of the advantages of QUASS CEST MRI is that the CEST scans can be expedited without resorting to long scan duration, while the magnitude of the CEST effect can be recovered from QUASS reconstruction.

Quantitative CEST MRI Measurements

MTR asymmetry

The commonly used MTR_asym takes the difference between the signals with RF saturation applied on the reference frequency (−Δω) and the labile frequency (Δω), symmetric around the bulk water resonance

$$MT{R}_{asym}=\frac{S_{sat}(-\Delta \omega )-S_{sat}(\Delta \omega )}{S_0}$$ (5)

The asymmetry calculation corrects RF spillover and, to a lesser degree, the semisolid MT effects (47). Zhou et al. determined the amide proton exchange rate from water exchange spectroscopy and calibrated it against pH determined by phosphorus (31P) spectroscopy. The calibration was further applied to map MTR_asym for absolute tissue pH (85). Due to the slow amide exchange rate, ischemia-induced MTR_asym contrast peaks with a relatively mild RF saturation amplitude (86). In addition to APT, there are multiple partially overlapping exchangeable groups such as amine and guanidyl CEST MRI effects, suggesting the possibility of combining them and enhancing the pH contrast. For example, Jin et al. took the signal difference between 3.6 ppm (amide) and 2 ppm (guanidyl) as a new pH-weighted measure (87). Because of their exchange rate difference, a small pH change causes their signals to vary in an opposite direction, resulting in a larger signal change (pH_enh). Jin et al. showed that whereas APT-weighted (APTw) at 3.6 ppm indicated a small signal drop (Fig. 1a) following tissue acidosis induced by CO₂ challenge, pH_enh detects a more substantive signal increase (Fig. 1b). To rule out that the observed pH_enh is not due to perfusion change, the experiments were repeated with an O₂ challenge, which induces hemodynamic changes but not pH (Fig. 1c). Fig. 1d shows the average time course of APTw and pH_enh obtained from the cortex, with the peak signal change in pH_enh much stronger than APTw. It is important to note that exchange rates of amide, amine, and guanidyl groups with respect to pH differ substantially, allowing variant pH imaging methodologies. For example, MTR_asym at 3 ppm (amine) obtained under a high RF saturation amplitude has been applied to map tumor pH (88), which may serve as a potential early biomarker for recurrent glioblastoma (89,90). pH-weighted imaging has also been correlated with tumor hypervascularity (91).

Fig. 1.

Comparison of pH-weighted APTw and pH_enh following hypercapnic and hyperoxia challenges. a) Signal change in APTw map following hypercapnia. b) Signal change in pH_enh map following hypercapnia. c) pH_enh MRI signal change following hyperoxia. d) The timecourse of APTw and pH_enh following hypercapnic and hyperoxia challenges (Jin et al. Neuroimage 2017;157:341-350.)

It is worth noting that an index similar to MTR_asym is the apparent relaxation exchange (AREX), which utilizes inverse normalization to linearize the CEST calculation (77). Inverse asymmetry calculation also allows for T₁ normalization, which is not easily applicable when using MTR_asym. Specifically, we have

$$AREX=R_{1w}\cdot (\frac{S_0}{S_{sat}(+\Delta \omega )}-\frac{S_0}{S_{sat}(\Delta \omega )})$$ (6)

In the presence of multiple CEST effects that are partially overlapping or bilaterally around the bulk water resonance, both MTR_asym and AREX reflect a mixture of contrasts. The asymmetry analysis is simple to use, yet prone to confounding contributions from multipool CEST and changes in relaxation and asymmetric semisolid macromolecular effects. As a result, pH-weighted MTR_asym has been demonstrated in acute stroke imaging without substantial changes in relaxation and semisolid MT effects.

Direct Saturation-Corrected (DISC) CEST

Instead of the asymmetry calculation, Sun et al. (92) and Jin et al. (93) proposed interpolating signals at two neighboring offsets around the labile proton offset as the baseline to isolate APT and nuclear Overhauser enhancement (NOE) effects. However, linear baseline interpolation is simplistic because the direct RF saturation is nonlinear concerning the RF offset. Zhou et al. proposed a direct saturation-correction (DISC) algorithm to correct the direct RF saturation based on relaxation images. Furthermore, because the MT effect has a broad spectrum, it can be described as a linear baseline (94). The DISC-CEST method allows for using asymmetric boundary offsets

$$APT{R}_{DISC}(\Delta {\omega }_i)=\frac{{\delta }_1\cdot \Delta Z(\Delta {\omega }_i+{\delta }_2)+{\delta }_2\cdot \Delta Z(\Delta {\omega }_i-{\delta }_1)}{{\delta }_1+{\delta }_2}-\Delta Z(\Delta {\omega }_i)$$ (7)

in which the upfield and downfield boundaries are $\Delta {\omega }_i-{\delta }_1$ and $\Delta {\omega }_i+{\delta }_2$, respectively. In addition, ΔZ is the Z-spectral intensity difference between the experimental measurement (multipool) and bulk tissue water signal due to direct RF saturation alone (one-pool). The DISC approach has been extended to 3T, which is advantageous over the simplistic linear estimation of the baseline (95). The limitation of the DISC analysis is that it provides a mixed measurement in the presence of multiple overlapping CEST effects. Therefore, weak RF saturation is often preferred to isolate the CEST effect (slow chemical exchange) of interest.

Magnetization transfer and relaxation normalized APT (MRAPT) analysis

Although MTR_asym at 3.5 ppm has been associated with tissue acidification in acute stroke (96), it includes not only the APT effect but also T₁, NOE effect at −3.5 ppm, and slightly asymmetric semisolid MT contributions (97-103). Studies from Jin et al. and Zhou et al. showed little NOE signal change in hyperacute stroke (93,104). However, both APT and NOE signal changes have been observed in acute stroke (105,106). Recently, Wu et al. traced the apparent NOE signal change during the acute stroke to a small T₁ change. Specifically, there have been reports of a non-negligible T₁ increase in the acute ischemic lesion (107,108). The exchange spectrum (R_ex) was calculated by subtracting the water spinlock relaxation rate from the experimental measurement. Whereas Z-spectra showed comparable APT and NOE signal changes, the CEST-specific exchange spectra revealed that APT signal change dominates that of NOE.

The MRAPT analysis has been developed to improve pH specificity in the pH-weighted MTR_asym image (109). This is built on the observations that 1) there is little pH difference in the intact white matter and gray matter (110); 2) pH effect in MT contrast is negligible (111); and 3) APT signal change dominates NOE's in acute ischemia (93,107). The background homogenization provides a straightforward approach to model and correct non-pH signals in the routine MTR_asym image. Specifically, we have

$$\Delta MRAPTR=R_{1w}\cdot MT{R}_{asym}-F(R_{1w},MMTR)$$ (8)

where the mean MTR at ±3.5 ppm is denoted as MMTR. pH-specific ΔMRAPTR has been calibrated against pH determined from lactate MR spectroscopy (108,112), with the correlation between ΔMRAPTR and pH substantially improved over that of MTR_asym. Specifically, Fig. 2a shows that pH-specific ΔMRAPTR from the diffusion lesion highly correlates with tissue pH. Fig. 2b shows the pH-weighted MTR_asym image, while Fig. 2c shows the pH-specific ΔMRAPTR image. Fig. 2d converts the ΔMRAPTR image to absolute tissue pH. Note that pH captured not only the severe pH drop in the ischemic core, but also mild acidic tissue in the primary motor cortex. It is also worth pointing out that the development of pH-specific MRI makes fast B₀ inhomogeneity correction feasible by incorporating the field inhomogeneity into the MRAPT regression model (113). Also, the MRAPT analysis can be generalized to APT MRI with pulsed-RF saturation (114) and different RF saturation amplitudes (115). Although the MRAPT analysis is simple to implement and highly pH specific, it assumes little concomitant MT change, so regression established from the intact tissue applies to the diseased tissue. The approach is limited to well-defined applications such as acute stroke imaging, where MT change is minimal.

Fig. 2.

Tissue pH MRI from pH-specific MRAPT analysis. a) The pH-specific ΔMRAPTR and pH are highly correlated. b) pH-weighted MTR_asym map. c) pH-specific ΔMRAPTR map. d) Tissue pH image derived from ΔMRAPTR image and its pH calibration (Wang et al. Neuroimage 2019;191:610-7.)

Quantitative CEST (qCEST) analysis

There has been tremendous interest in solving the labile proton ratio and exchange rate beyond the CEST-weighted images (116,117). McMahon et al. quantified the exchange rate using either saturation power or time dependence (QUESP and QUEST) by fitting the CEST signal as a function of RF saturation amplitudes and durations (118). Randtke et al. proposed Hanes-Woolf linear QUESP and inverse QUEST to improve the quantification (119,120). Sun extended the QUEST algorithm with ratiometric analysis (QUESTRA) to enhance the quantification accuracy (121). This approach addresses the issue that the rate at which the CEST effect approaches its equilibrium state is not governed by R_1w but by R_1ρ. Hence, using the reference signal provides a simple normalization strategy to account for the spinlock relaxation without the need to measure it. These algorithms were extended to describe non-equilibrium initial and weak labeling conditions (122). Although a quantitative CEST solution determines labile proton exchange rate and concentration of the underlying CEST system, it has been mainly limited to applications using exogenous CEST contrast agents. The multipool contribution needs to be isolated for in vivo applications, so the same labile group is measured under different conditions (e.g., B₁). However, this task is not trivial because the typical endogenous CEST effect is a few percent; hence, of very limited sensitivity.

It has been noted that RF amplitude provides a unique dimension for quantitative CEST analysis (123). Dixon et al. proposed an omega plot to solve the exchange rate without a priori knowledge of the labile proton ratio in paramagnetic CEST (PARACEST) MRI (124). In short, the plot of the CEST effect versus inverse RF power level is linear, with the x-intercept providing a direct measurement of the exchange rate. It has been shown that with the correction of the RF spillover effect, the omega plot could be extended to describe diamagnetic CEST (DIACEST) agents (125,126). Also, quantitative CEST MRI has been preliminarily applied to study muscle and brain tissue (127,128).

QUASS CEST MRI

The finite relaxation delay and saturation time can be accounted for using the recent QUASS algorithm, which reconstructs the equilibrium CEST effect from experimental measurement. Fig. 3 shows the routine and QUASS MTR_asym images from a dual pH creatine phantom using Td/Ts ranging from 1.5 to 7.5 s. Fig. 3a shows that the routine MTR_asym intensity increases with Td and Ts as it approaches the equilibrium CEST state. Figs. 3b and 3c show the corresponding CEST Z- and asymmetry spectra, respectively, revealing large variations as a function of Ts and Td, as expected. In comparison, the QUASS Z- (Fig. 3d) and asymmetry- (Fig. 3e) spectra overlap with each other, revealing consistent pH-specific CEST MRI contrast (Fig. 3f). The data shows that the QUASS algorithm can unify CEST scans obtained under different conditions. Also, it has been shown that the ratiometric analysis based on the QUASS solution provides improved quantification of the labile proton fraction ratio and exchange rate over those with apparent measurement (129). In addition, using the inverse Z-spectrum normalization combined with QUASS reconstruction enables proper T₁ correction (130). Furthermore, the QUASS processing has been recently demonstrated in fast multislice CEST imaging at 3T (131) and brain tumor applications (132,133). Although the QUASS algorithm enables equilibrium CEST quantification, it requires parametric T₁ mapping (134), which could lengthen the total scan time.

Fig. 3.

Comparison of the conventional and QUASS CEST MRI from a creatine pH phantom. a) Routine MTR_asym at 1.9 ppm obtained under different Td and Ts. b) Z-spectra and c) asymmetry spectra. d) The corresponding QUASS Z-spectra and d) QUASS asymmetry spectra. e) The QUASS MTR_asym at 1.9 ppm under different Td and Ts. (Sun. Magn Reson Med 2021;86(2):765-776.)

Lorentzian fitting

The Lorentzian fitting has been developed to describe CEST measurement in the presence of multiple labile proton groups. Briefly, the normalized Z-signal change can be described by

$$1-\frac{S_{sat}}{S_0}\approx \frac{\sum _{i=1}^N{f}_r^i{k}_{sw}^i\cdot {\alpha }^i}{R_{1\rho }}\approx {\sum }_{i=1}^N\frac{f_r^i{k}_{sw}^i}{R_{1w}}\cdot {\alpha }^i$$ (9)

in which the superscript i represents the i^th labile proton group. The labeling coefficient can be approximated as a Lorentzian function. The choice of the number of labile pools depends on the field strength, RF saturation scheme, Z-spectral sampling density, and SNR. For example, multipool Lorentzian models including water (0 ppm), semisolid macromolecular MT (−2 ppm), amide (3.5 ppm), amine (2.75 ppm), guanidinium (2 ppm), and NOE effects at 1.6 ppm and 3.5 ppm upfield from water have been deployed to describe brain CEST effects (104,135-137). Although the accuracy improves with the number of CEST pools, the nonlinear multiparameter fitting is subject to substantial errors when the spectral resolution and SNR are insufficient. To enhance the robustness of fitting, Zhou et al. proposed the image downsampling expedited adaptive least-squares (IDEAL) algorithm to regularize the fitting against noise (138). Nevertheless, for applications at low fields such as 3T, the Z-spectrum has a relatively coarse spectral resolution. Therefore, models of 3 or 4 exchangeable pools are often chosen to minimize the risk of overfitting (139,140). To address the pronounced MT effect, Heo et al. proposed fitting and extrapolating semisolid magnetization transfer reference (EMR) to remove it from Z-spectra (141,142). The EMR approach has been applied to isolate APT from NOE effects in acute stroke patients and showed improved lesion depiction (105).

Several factors need to be considered when deploying the multipool Lorentzian fitting. First, the Lorentzian fitting is valid when the saturation time is sufficiently long: The transient terms could introduce an asymmetric baseline because the spinlock relaxation rate increases when the RF saturation approaches the bulk water resonance (143). Second, in the presence of notable changes in relaxation rates, which is not uncommon in diseased tissues, the bulk water spinlock relaxation term will induce a symmetric baseline shift. Such a baseline change needs to be considered when assigning the origin of the CEST contrast. Third, although T₁ normalization improves the measurement of the labile proton ratio and exchange rate (144-146), how to properly account for the T₁ effect in CEST imaging in vivo (147,148) is controversial. One particular limitation of Lorentzian fitting is that it requires a reasonably densely sampled Z-spectrum; therefore, it could be time-consuming.

Model-based fitting

The Bloch-McConnell equations have been extended to describe the multipool CEST effect. For example, Sun et al. applied 5-pool models to determine exchange properties of iopamidol (74). However, multipool fitting requires many free parameters and is at the risk of overfitting. Chappell et al. introduced the Bayesian model to regularize the fitting (149). Specifically, Ray et al. demonstrated that Bayesian model-based analysis (BayCEST) with relaxation compensation provides robust pH-sensitive measurement (150). BayCEST was further shown in acute stroke patients to be advantageous over MTR_asym in defining the ischemic lesion (151). Msayib et al. compared different indices in acute stroke patients, including MTR_asym, AREX, and APT* with local linear baseline assumptions, post-acquisition Lorentzian difference analysis, and APTR* from 3-pool and 4-pool Bayesian models (152). Fig. 4 shows that the 3-pool Bayesian model exhibits the lowest spatial variability, yet captures pH drop in the ischemic lesion. Foo et al. also concluded that the model-based fitting is advantageous when SNR is inadequate (153). It is interesting to point out that using the BayCEST approach, Ray et al. concluded that 2/3 of APT contrast originates from changes in protein concentration, while the remaining 1/3 comes from changes in tumor pH (154). Whereas the model-based fitting approach is robust, it requires a reasonably densely sampled Z-spectrum, which could be computation-intensive and time-consuming.

Fig. 4.

Comparison of different CEST indices in acute stroke patient imaging. The model-based Bayesian CEST images capture pH drop in the ischemic lesion with the least spatial variability (Msayib et al., NeuroImage: Clinical 2019;23:101833.)

Ratiometric analysis

The original ratiometric pH imaging requires the CEST agent to have at least two exchangeable groups of different chemical shifts and exchange properties, so their ratio depends only on the exchange rate and, hence, pH (42). Aime et al. reported iopamidol, a CT contrast agent, as a new CEST MRI contrast (155-157). The iopamidol ratiometric analysis depends only on pH, not on T₁ and CEST agent concentration. Due to its FDA-approved status, iopamidol is highly translational for in vivo CEST imaging (158,159). Longo et al. showed that pH is a sensitive biomarker of early acute ischemic injury before creatinine level change (160). Pavuluri showed that in addition to pH, iopamidol MRI could quantify renal perfusion status (161).

Beyond iopamidol, several iodinated contrast agents have been explored for pH imaging, including Iopromide (162) and Imidazole (163). Furthermore, the ratiometric CEST MRI has been extended to CEST agents of a single liable proton group. Specifically, RF power-based ratiometric takes the ratio of CEST measurements performed under different B₁ saturation amplitudes instead of two different chemical shifts, and was demonstrated using iobitridol (164-166). Wu et al. further generalized the ratiometric pH imaging by mixing CEST effects at different chemical shifts and RF saturation levels (167). Fig. 5 shows the generalized ratiometric renal pH imaging in a rodent kidney. The iopamidol CEST effects at 4.3 ppm and 5.5 ppm were isolated using Lorentzian fitting with RF amplitudes of 1 and 2 μT, respectively. Absolute renal pH can be mapped using the modified ratiometric pH imaging. The generalized pH imaging is versatile for CEST applications at non-high magnetic fields, where the CEST effects might partially overlap, decreasing the dynamic range of routine ratiometric analysis. Similarly, Zu et al. used the RF saturation pulse flip angle as an alternative to measuring the chemical exchange rate (168). The ratiometric pH imaging approach has also been explored for endogenous pH mapping. McVicar et al. demonstrated the endogenous ratiometric MRI of amine and amide concentration-independent detection (AACID) for mapping ischemic tissue pH (169). However, endogenous ratiometric pH imaging assumes that the concentration of different labile groups is relatively stable, which needs to be validated independently under pathological conditions.

Fig. 5.

A representative normal kidney pH map from the generalized ratiometric pH imaging at 4.7 T. The iopamidol CEST effects were measured with two RF amplitudes, and the peaks were isolated for 4.3 ppm (1 μT) and 5.5 ppm (2 μT). Their ratio map is pH-dependent, which permits absolute pH mapping (Wu et al. Magnetic Resonance in Medicine, 2018;79:1553-58.)

The ratiometric MRI approach has also been applied in vivo to map pH in cancerous tissues. Chen demonstrated dual RF power ratiometric pH imaging using Ioversol in mouse breast cancer (170). Recently, Tang et al. translated Ioversol pH MRI to examine hepatic tumor patients at 3T. Fig. 6a shows a patient with hepatic hemangioma, with its corresponding pH map (Fig. 6b). Note that pH was consistent with the surrounding liver tissue, confirming benign hemangioma. Fig. 6c shows a patient with hepatic carcinoma, with its pH in Fig. 6d. Its pH is significantly lower than the normal tissue. Indeed, pH in hepatic carcinoma was acidic (6.66 ± 0.19), while the extracellular pH was 7.34 ± 0.09 in hemangioma, demonstrating pH MRI as a differential diagnostic imaging methodology (171). Increasing data have documented that tumor extracellular pH is lower in faster-growing tumors, serving as a surrogate marker for differentiating benign and malignant tumors and early response to treatment.

Fig. 6.

Liver pH imaging with CEST MRI of Ioversol. a) A structural image of a hepatic hemangioma patient. b) pH map shows consistent pHe with the surrounding normal tissue. c) A structural image of a hepatic carcinoma patient. d) Tumor pH map is noticeably lower than normal tissue (Tang et al. Frontiers in oncology 2020;10:578985.)

It is helpful to mention that CEST pH MRI has certain advantages over the relaxation-based pH imaging approach. For example, although the relaxivity of macromolecular Gd complexes can be extracellular pH-responsive (172), it requires a double-contrast injection approach, sequential or concurrent, to determine the contrast concentration, which is cumbersome (173-175). Another notable extracellular pH imaging approach is biosensor imaging of redundant deviation in shifts (BIRDS), which utilizes pH-responsive chemical shifts to measure pH (176-178). However, its major limitation is that the contrast agent has very short T₁ and T₂ relaxation times, requiring customized imaging sequences. In comparison, pH CEST agents, particularly FDA-approved iodinated-based ones, allow concentration-independent pH imaging and are highly translational. Nevertheless, exogenous CEST agents-based pH imaging requires the administration of contrast agents at relatively high concentrations, mostly appliable in tumor and renal imaging applications.

Practical Considerations of CEST MRI Experiments

It has been well recognized that experimental parameters such as the RF saturation scheme (RF saturation pulse shape, amplitude, and duration), repetition time, and flip angle could profoundly impact the CEST MRI measurement. Therefore, the CEST MRI needs to be carefully optimized to improve its sensitivity and specificity.

Pulse sequence

The most straightforward implementation of the CEST MRI pulse sequence includes a relaxation delay, during which magnetizations recover toward their thermal equilibrium states, followed by a continuous wave (CW) RF irradiation to induce the CEST effect, during which the bulk water signal approaches its CEST equilibrium state following the spinlock relaxation rate. The sequence often concludes with a fast image readout such as echo planar imaging (EPI) or fast spin echo (FSE) readout (179). Although the CW saturation scheme is simple to implement and model mathematically, the pulse RF saturation scheme has been chosen due to the RF duty cycle, specific absorption rate, and hardware limitation (180-182). In addition, the flip angle has been explored as a novel means for isolating the amide CEST effect from the baseline (183,184). Fortunately, CW or pseudo-CW (90% or higher RF duty cycle) RF saturation has been recently demonstrated in mainstream MRI systems, including General Electric, Philips, and Siemens scanners. Although single slice readout was adopted in early clinical CEST imaging, recent CEST development includes substantially improved spatial coverage, including fast multislice (185-187) and 3D readout (141,188-190). Demetriou et al. recently provided a comprehensive review of CEST MRI pulse sequences (191).

Correction of field inhomogeneity in CEST MRI measurement

Field inhomogeneity is not uncommon in MRI experiments, particularly in areas of the tissue air interface. Because the field inhomogeneity directly impacts the saturation of labile protons and the asymmetry analysis, CEST measurement is very prone to field inhomogeneity. The B₀ inhomogeneity on MTR_asym can be modeled as a baseline shift due to direct RF saturation and a Lorentzian-shaped modulation of the labeling coefficient (192). For in vivo applications, the water saturation shift reference (WASSR) approach has been widely used to correct the Z-spectrum from field inhomogeneity (193-195). More recently, deep learning has been applied to provide fast correction of field inhomogeneity without Z-spectrum (196). It is worth pointing out that the routine field inhomogeneity correction algorithms assume intravoxel homogeneity. However, due to the relatively coarse spatial resolution of CEST MRI, it is unavoidable to have inhomogeneity within a single voxel. To address this, an intravoxel inhomogeneity correction (CIVIC) post-processing was developed (197). It describes the experimental CEST spectrum (Z_app) as a convolution of the ideal CEST spectrum (Z_orig) and a point spread function (PSF) that characterizes the intravoxel inhomogeneity distribution (i.e., F(ΔB₀)), as

$$Z_{app}=Z_{orig}⊛F(\Delta B_0)$$ (10)

Using a high-resolution field map, Z_orig can be reconstructed, which complements the routine voxel-based field inhomogeneity correction. In addition, although B₁ inhomogeneity has been considered relatively mild at typical field strengths, fast B₁ mapping and correction have been established (198).

Optimization of CEST sensitivity efficiency

Because the typical endogenous CEST effect is only a few percent, optimization of CEST scans is necessary to improve its detection. Initial work was dedicated to optimizing the RF saturation amplitude, with the goal of inducing sufficient labile proton saturation without an excessive spillover effect on the bulk water signal (64). In addition to the magnitude of the CEST MRI effect, its signal-to-noise ratio per unit time (SNR efficiency) serves as an alternative parameter to optimize (199) because the CEST MRI effect increases monotonically with the RF saturation time, yet the incremental gain over an excessively long saturation time becomes marginal. Thus, the repetition time needs to be considered to balance the magnitude and scan time. For an illustrative 2-pool system, the SNR efficiency has been derived as

$$SN{R}_{put}\propto \frac{CESTR}{\sqrt{TR}\cdot \sqrt{2+CEST{R}^2}}\cdot SN{R}_{S_0}$$ (11)

where $SN{R}_{S_0}$ is the control image SNR, which depends on the flip angle, repetition time, and echo time. The derivation of the full analytical solution makes it feasible to simulate and optimize multiple parameters simultaneously before experimental validation (83). However, the optimization strategy based on the magnitude and SNR efficiency of CEST MRI measurement could contradict each other. Whereas the magnitude of the CEST MRI effect is higher when using a small flip angle, the CEST MRI SNR efficiency peaks when using flip angles between 60 to 75 degrees. The ultimate optimization depends on the purpose of the experiments. Fortunately, the QUASS algorithm allows for sensitivity-based experimental optimization with little loss in the quantification accuracy.

The scan time can be reduced by sampling fewer frequency offsets. For example, sampling multiple offsets only around ±3.5 ppm expedites MTR_asym measurement, circumventing time-consuming Z-spectral acquisition (200). For the Z-spectral imaging, it has been demonstrated that an uneven sampling scheme may provide similar results as the fully sampled Z-spectrum, particularly at 3T (201,202).

Conclusion

Steady progress has been achieved in CEST imaging, with increased specificity and sensitivity to pH. The technical advancement enables emerging applications in disorders such as acute stroke and cancer. Further development of robust CEST analysis and post-processing algorithms will expedite its adoption in the clinical setting, facilitating the transition from CEST-weighted imaging to quantitative maps of tissue pH and metabolites.

Abbreviations

APT

Amide proton transfer

CEST

Chemical exchange saturation transfer

CESTR

CEST ratio

CNR

Contrast to noise ratio

ΔMRAPTR

Delta MRAPT ratio

MT

Magnetization transfer

MMTR

Mean magnetization transfer ratio

MTR

Magnetization transfer ratio

MTR_asym

MTR asymmetry

MRAPT

MT and relaxation normalization APT

NOE

Nuclear overhauser enhancement

pHw

pH-weighted

QUAS

Quasi-steady-state

rCNR

Relative CNR

rCESTR

ratiometric CESTR

ΔMRAPTR

Delta MRAPT ratio