Radiofrequency (RF) Labeling Strategies in Chemical Exchange Saturation Transfer (CEST) MRI

Abstract

Chemical exchange saturation transfer (CEST) MRI has generated great interest for molecular imaging applications since it can image low concentration solute molecules in vivo with enhanced sensitivity. CEST effects are detected indirectly through a reduction in the bulk water signal after repeated perturbation of the solute proton magnetization using one or more radiofrequency (RF) irradiation pulses. The parameters used for these RF pulses – frequency offset, duration, shape, strength, phase, and inter-pulse spacing – determine molecular specificity and detection sensitivity, thus their judicious selection is critical for successful CEST MRI scans. This review article describes the effects of applying RF pulses on spin systems and compares conventional saturation-based RF labeling with more recent excitation-based approaches that provide spectral editing capabilities for selectively detecting molecules of interest and obtaining maximal contrast.

Graphical Abstract

CEST MRI is a promising technique which measures low-concentration metabolites with enhanced sensitivity by indirectly detecting the signal reduction of bulk water after applying RF irradiation pulses. However, the generated Z-spectrum includes multiple signal components, which confounds quantitative analysis. The judicious selection of RF pulse parameters used can improve the detection specificity and sensitivity. This review begins with the RF labeling effects on spin systems, and detailed broadly used approaches which are based on saturation, excitation, or hybrid form labeling.

1. Introduction

The perturbation of a solute proton pool using a radiofrequency (RF) pulse and the subsequent exchange of perturbed magnetization between proton pools underpins conventional magnetization transfer contrast (MTC)¹, chemical exchange saturation transfer (CEST)^2–7, and relayed nuclear Overhauser Effect (rNOE)^3,8–10 MRI studies. This magnetization exchange can be in the form of a physical exchange of protons from the solute molecule to the solvent³, intra- or inter-molecular cross-relaxation between spins in close proximity that experience through-space dipolar coupling¹¹, or a combination of these effects.^3–10 Furthermore, depending on the exchangeable proton chemical and structural environments or molecular dynamics, exchange and cross-relaxation rates can vary by several orders of magnitude. This broad scope hints at the potential for using magnetization exchange methods (e.g., CEST, MTC, rNOE) to study a large number of biological systems in situ, including signals from mobile proteins and peptides^12–16, creatine and phosphocreatine^17–20, small sugars and sugar polymers^10,21–27, glutamate^28–30, myo-inositol³¹, urea^32,33, glycosaminoglycans³⁴, but it can be challenging to separate these multiple signal sources in vivo.

A key improvement in detection specificity and sensitivity occurred when Balaban and colleagues² showed how narrow bandwidth RF pulses can selectively label exchangeable proton pools in specific solute molecules and how the prolonged irradiation of these protons can lead to enhanced detection sensitivity for these solutes due to multiple label-transfer events. Therefore, the judicious selection of RF labeling pulse parameters, such as frequency offset, duration, shape, B₁ field strength, phase, and inter-pulse spacing, is key to optimizing the specificity and sensitivity of CEST, rNOE, and MTC MRI studies. The following sections discuss the effect of RF pulses on spin pools, define the saturation transfer MRI signal, and introduce key components of the water saturation spectrum, also known as the Z-spectrum.^32,35 These sections are followed by a discussion of strategies and techniques for improving the sensitivity and specificity of CEST MRI studies using saturation and excitation-based RF labeling approaches. We focus on protons but the concepts discussed can be generalized to heteronuclear CEST experiments.^36–41 Additionally, the RF labeling methods can be combined with novel acquisition approaches^42–46 to achieve specific and rapid CEST imaging.

2. RF labeling

Magnetization exchange MRI experiments consist of three stages in which (i) solute nuclear pools are magnetically labeled using RF pulses, (ii) the magnetically labeled components transfer to the solvent (typically bulk water), and (iii) the signal of the solvent, containing the labeled components, is measured. These stages can overlap but in general, the period before detection (labeling + transfer) is called the preparation period with duration t_prep. The combination of RF pulses along with other pulse sequence elements (magnetic gradient pulses and inter-pulse delays) in this stage determines the characteristics of the magnetization of the solute pool and, subsequently, the magnitude of the resulting MRI signal. Several types of labeling schemes have been proposed to achieve specific experimental requirements but most of these methods can be summarized into two categories, (i) saturation-based and, (ii) excitation-based RF labeling methods.⁴ In this section, we describe the effect of RF labeling on spin systems, define the saturation transfer signal, and introduce the main components of the Z-spectrum.

2.1. Effect of RF labeling on spins

When a sample containing independent identical spins is placed in a static magnetic field, the magnetic dipole vectors comprising this spin ensemble (or pool) have a slightly higher population in the lower energy state (|α>) compared to the higher energy state (|β>). This difference in population results in a net longitudinal spin polarization. The measured MRI signal is proportional to this polarization. Note that we are limiting this description to two pure states for simplicity but spins may also be in a superposition of |α> and |β>.⁴⁷

The magnitude of an oscillating magnetic field (B₁) generated by an RF pulse is orders of magnitude smaller than the static magnetic field (B₀). Still, the cumulative effect of a resonant B₁ field can change the spin state. For instance, an RF pulse with pulse duration (t_dur) much shorter than the longitudinal relaxation time constant (T₁) will result in spins transferring from the |α> state to the |β> state and vice versa (Figure 1A). This process is called excitation. The net result is a higher energy state in total. In the classical picture, not depictable with this energy level diagram that just shows population differences, this polarization or full magnetization can be rotated away from the magnetic field axis. When perpendicular to B₀ (after a 90-degree pulse) this will give full transverse magnetization and a signal, S, but one can also invert the full magnetization. Eventually the system will relax back to equilibrium with T₁. On the other hand, when t_dur approaches T₁, transitions between |α> and |β> occur concurrently with longitudinal relaxation and at even longer pulse durations (t_dur > T₁), the spin system settles into a steady-state where there is an approximately equal number of spins in the two states (Figure 1B). The negligible difference means there is no significant longitudinal magnetization, and the spin system is said to be saturated. If multiple spins in either excitation or saturation state (i.e., labeled) exchange physically with another spin ensemble or pool, the excitation/saturation state is also transferred to the other pool (e.g., in CEST experiments) and the signal in that other pool will be reduced upon detection, with the magnitude of the reduction depending on the labeling properties. Since we detect this effect as a signal reduction, the term (partial) saturation is used. Thus, the Z-spectrum detects the water saturation after labeling spin pools as a function of irradiation frequency.

Figure 1.

Effect of (A) excitation (t_dur << T₁) and (B) strong saturation (t_dur > T₁) on a pure two spin state system, with the population difference proportional to the magnitude of the polarization. Shorter RF pulses (t_dur ~ T₁) can induce partial saturation. t_dur is the RF pulse duration and T₁ is the longitudinal relaxation time constant of a pool.

2.2. The Z-spectrum

The most common acquisition scheme for exchange transfer MRI is to acquire a spectrum in which the water signal intensity after saturation (S_sat) is normalized to the signal without saturation (S₀) and displayed as a function of irradiation frequency. As the water signal saturation is a consequence of a reduction in the water Z-magnetization after the labeling period, this is called a Z-spectrum⁴⁸. However, in vivo tissue signals contain multiple contributions, including direct water saturation (DS)⁴⁹, CEST effects from multiple types of exchangeable protons^50,51, MTC of semi-solid macromolecules^52,53, and rNOEs of mobile macromolecules^3,8,34,54,55. Due to the inherent broadness of saturation signals, a Z-spectrum often presents as a sum of multiple overlapping peaks with an appearance heavily dependent on the static magnetic field strength (B₀), relaxation parameters (both T₁ and T₂), and B₁ strength (Figure 2).^56,57 Higher B₀ results in greater spectral dispersion (larger frequency spread in Hz and slower exchange regime) and thus more resolved peaks. One practical approach for improving Z-spectral resolution is to use lower amplitude saturation pulses (e.g., 0.5 vs 2.0 μT in Figure 2A). Simulation parameters for Figure 2 are shown in Table 1.

Figure 2.

Simulated Z-spectra of continuous wave (CW)-based saturation with components from: direct water saturation (DS, 0 ppm), symmetric magnetization transfer contrast (MTC, 0 ppm), hydroxyl (+0.9 ppm), phosphocreatine (PCr, +2.6 ppm), creatine (Cr, +2 ppm), amine (+3 ppm), amide (+3.5 ppm), and rNOE (−3.5 ppm) effects under different B₁ (0.5, 1.0, 1.5, and 2.0 μT) at (A) 11.7 T, (B) 7 T, and (C) 3 T. The simulation parameters are shown in Table 1, t_dur = 4 s. Signal specificity reduces with lower B₀ and increased B₁. Also, fast exchanging protons (hydroxyl: -OH, amine: -NH₂) are broad and often partially merged with water and thus not resolved.

Table 1.

Exchange parameters for select solute pools used in Bloch equation-based simulations.^*

		Offset from water (ppm)	Exchange rate ksw (s−1)	Concentration** (mM)	T2 (ms)
Water		0.0	1	111000	62
Hydroxyl (Myoinositol)		+0.9	2,000	45	55
Guanidinium	(creatine)	+2.0	1,100	20	170
	(phosphocreatine)	+2.6	260	10	100
Amine (glutamate)		+3.0	5,500	20	200
Amide (larger proteins)		+3.5	30	72	100
rNOE (aliphatic)		−3.5	20	100	5
MTC***		0.0	40	5,500	0.009

^*T_1w = 1.6s, T_1s = 1s.

^**of the protons in the solute.

^***The MTC pool spans a broad frequency range but here its center offset is set to the same frequency as water (0 ppm).

The simulations in Figures 2, 4, 7 were performed using custom-written scripts in Python and the code can be downloaded from a public repository, https://github.com/ChongxueBie/RF-labeling-techniques-in-CEST-MRI. The simulations in Figures 5, 6 were performed on MATLAB 2022b using source code downloaded from https://github.com/kherz/pulseq-cest-library.⁵⁸

Figure 4.

Comparison of (A) saturation efficiency of continuous wave (CW)-based saturation at varied B₁ (t_dur = 4 s) and (B) excitation efficiency of inversion-based excitation for various solute protons with different exchange rates (k_sw). Two-pool model (solute / water), T_1w/T_2w = 1.6 s/62 ms, T_1s/T_2s = 1 s/100 ms, [H_s]/[H_w] = 6.5×10⁻⁴. Δω was set to 20 ppm to avoid spurious saturation of the water pool.

Figure 7.

Simulations of CEST contrast with continuous wave (CW) pulse for the following solute protons at 7 T: (A) amide (+3.5 ppm), (B) creatine (Cr, +2 ppm), (C) phosphocreatine (PCr, +2.6 ppm), and (D) amine (+3 ppm). The simulation parameters of the multi-pool model are same as Figure 2 and shown in Table 1. Contrast for each solute is the amplitude of the signal of interest was calculated by taking the Z-spectral difference at the solute frequency with and without the solute pool.

Figure 5.

(A) Saturation efficiency (α) and (B) proton transfer ratio (PTR) values at 7 T as a function of duty cycle and B₁ for solute protons with different exchange rates. Duty cycle is defined as the ratio of the period of RF application for a pulse train and the total preparation time (t_prep = 4s) including RF pulses (duration of 80 ms) and delays. Simulations used rectangular pulses with phase accumulation. Other simulation parameters were: two-pool model (solute/water), T_1w/T_2w = 1.6 s/62 ms, T_1s/T_2s = 1 s/ 100 ms, [H_s]/[H_w] = 6.5×10⁻⁴. Δω was set to 20 ppm to avoid spurious saturation of the water pool. The saturation efficiency is calculated by $\alpha =1-S_{zs}^{ss}/S_{0s}$ and PTR was calculated by measuring the signal reduction in the water pool.

Figure 6.

(A) Simulated Z-spectra for amide proton pool (Δω = 3.5 ppm, k_sw = 30 Hz) obtained by pulsed-CEST with different duration rectangular pulses (with phase accumulation). Delay time of 5 ms, preparation time of 4 s, B_1rms = 0.5 μT, B₀ = 7 T. Other simulation parameters were: two-pool model (solute/water), T_1w/T_2w = 1.6 s/62 ms, T_1s/T_2s = 1 s/ 100 ms, [H_s]/[H_w] = 6.5×10⁻⁴. (B) B₁-amplitude spectra which show the effect of the rectangular pulses (panel A) in frequency domain. These B₁-amplitude spectra were generated by Fourier transforming the time domain pulse profile. (C, D) Simulated Z-spectra obtained by pulsed-CEST with different RF pulse shapes showing the effects of not taking into account phase accumulation off-resonance for rectangular pulses (black). This effect can be removed either by using shaped pulses (C) or by adding phase-accumulation for the rectangular pulses (D). Simulations used pulses with duration of 80 ms. Black arrows indicate the off-resonance pulse effects. Other simulation parameters are same as (A).

2.3. Description of the CEST signal

The difference in spin populations depicted in Figure 1 is greatly exaggerated and in reality, the population distribution at physiological temperatures is given by a Boltzmann distribution (difference between the |α> and |β> populations of 1 to 10 per million, depending on B₀). The behavior of this number of spins is mathematically intractable but it is possible to approximate the behavior of many spin systems using a magnetization vector M which indicates the magnitude and direction of the net polarization. A simple case of magnetization exchanging between a solute pool (H_s) and water pool (H_w) at rate k can be described by a two-pool model,

$$H_s\mathrm{ }\begin{array}{c}{k}_{sw}\\ \rightleftarrows \\ k_{ws}\end{array}\mathrm{ }{H}_w$$

The CEST effect is measured as an accumulated attenuation in the water signal intensity after prolonged RF labeling applied on pool H_s, using amplitude B₁ and a frequency corresponding to H_s. Under a low amplitude RF field and continuous labeling, the magnetization of the two pools can approach a steady-state (dM/dt = 0), and the reduction in the water signal intensity (proton transfer ratio, PTR) can be derived⁵⁹:

$$\text{PTR}=\frac{S_0-S_{sat}}{S_0}=\frac{k_{sw}\alpha f_s}{(R_{1w}+k_{ws}\alpha )}[1-e^{-\left(R_{1w}+k_{ws}\alpha \right)t_{dur}}]$$ (1)

Where S₀ and S_sat are the water signal with and without saturation. R₁ and R₂ are the spin-lattice and spin-spin relaxation rates, respectively. f_s is the fractional concentration of the solute protons, $f_s=\frac{\left[H_s\right]}{\left[H_w\right]}=\frac{k_{ws}}{k_{sw}}$, square brackets indicate concentration. t_dur = t – t₀. Saturation efficiency $\alpha =1-S_{zs}^{ss}/S_{0s}$,^57,60 where $S_{zs}^{ss}$ is the steady-state magnetization of solute pool, and $S_{0s}$ is the equilibrium magnetization of solute pool, α = 1 corresponds to complete saturation.

According to Eq.1, CEST contrast (PTR) is dependent on the magnitude of the solute spin pool size relative to the detected spin pool size ([H_s]/[H_w]), the longitudinal relaxation rate or conversely T₁ (1/R₁), the exchange rate (k_sw), and the irradiation parameters that determine α and the labeling time t_dur. Often, it is not possible to change relaxation or solute parameters but there is much more flexibility to vary RF parameters. The following sections detail how RF irradiation parameters can be optimized for different experimental needs.

3. Saturation-based labeling

This section details saturation-based labeling approaches, including continuous-wave (CW) irradiation, pulsed-saturation schemes, and innovative saturation-based editing methods, summarized in Figure 3. Notice that spin-lock approaches can also be applied in a pulsed manner, but their discussion is beyond the scope of this paper. The suitability of each approach is dependent on the MRI scanner hardware available, specific absorption rate (SAR) limitations, and the system under investigation.

Figure 3.

Illustration of saturation-based labeling approaches. Solute protons are labeled within the preparation period (t_prep) using: (A) continuous saturation by low-power RF irradiation (CW-CEST) with a duration time of t_dur (= t_prep), (B) saturation by pulse train (pulsed-CEST), and (C) saturation-based editing approaches, including (C.1) adjusting RF pulse duration and amplitude, (C.2) alternating additional RF irradiation parameters, such as (C.2a) saturation with frequency alternating RF irradiation (SAFARI), (C.2b) Z-spectroscopy with alternating-phase irradiation (ZAPI) as well as uniform magnetization transfer ratio (uMT), (C.2c) length and offset varied saturation (LOVARS), (C.2d) multi-echo length and offset varied saturation (MeLOVARS), and (C.3) modulating inter-pulse delays, for example (C.3a) variable delay multi-pulse (VDMP). Then the magnetic label is transferred continuously from solute protons to bulk water. Δω is the radial frequency offset of the RF pulse from the water resonance, φ is the RF pulse phase, θ the flip angle, N is the total number of saturation modules applied. “Acq” is the acquisition module.

3.1. Continuous wave (CW)-based saturation

The most straightforward manner of approximating prolonged low-amplitude RF irradiation is through a CW pulse (Figure 3A). Almost all preclinical systems and many human MRI scanners can generate low amplitude RF pulses of several seconds duration using single RF amplifiers or a combination of amplifiers.^61–66 For a CW pulse, the saturation efficiency α is given by⁶⁰,

$$\alpha ={\omega }_1^2/({\omega }_1^2+pq)$$ (2)

where $p=r_{2s}-\left(k_{sw}{k}_{ws}/r_{2w}\right)$ and $q=r_{1s}-\left(k_{sw}{k}_{ws}/r_{1w}\right)$. r₁ and r₂ are the effective longitudinal and transverse relaxation rates, respectively, of the pools involved during RF irradiation (r_1s = R_1s + k_sw; r_1w = R_1w + k_ws; r_2s = R_2s + k_sw; r_2w = R_2w + k_ws). The saturation efficiency for a four-second-long RF pulse with constant amplitude (duty cycle of 100%) across a range of exchange rates is shown in Figure 4A. Under these conditions, it is possible to achieve almost complete saturation of the solute pool (α ≈ 1) for exchange rates up to 200 s⁻¹ using low B₁ strength (0.5 μT) thus increasing contrast as given by PTR (Eq. 1) while retaining proton pool specificity as shown in Figure 2. Amide proton transfer (APT) is one case where these low pulse powers would be appropriate since the exchange rates of these protons have been measured to be as low as 28 s⁻¹.¹² Higher exchange rates can still be efficiently saturated at higher B₁ but these higher strengths concomitantly broaden the saturation spectral bandwidth leading to reduced signal specificity (Figure 2).

CW-based saturation approaches often produce the largest total CEST effect size since magnetic labeling occurs over the entire preparation period. These seconds-long pulses are generally easy to implement on preclinical scanners where smaller RF transmission coils negate concerns with RF amplifier duty cycles. Human translation of CW-CEST methods is more challenging due to hardware concerns and safety (specific absorption rate, SAR) limits the RF power delivered to subjects⁶⁷, especially since most clinical scanners now have body coil transmit and SAR increases proportionally to the square of the coil radius. Still, recent studies have shown that alternating between transmit channels using multiple RF amplifiers allows for pseudo-continuous pulses on clinical MRI scanners⁶⁸ thus enabling long saturation pulses which utilize the whole preparation period for labeling.

3.2. Pulsed saturation

SAR and RF hardware duty cycle limits often compel the use of a train of shorter RF saturation pulses separated by delays to achieve a steady state at some level of saturation (Figure 3B).^69–72 These pulsed saturation approaches lead to lower saturation efficiencies, especially when irradiating fast-exchanging protons. Figure 5A shows that high saturation efficiencies can be obtained for slowly-exchanging pools using a pulsed pre-saturation period but for faster exchange pools (e.g., k_sw > 260 s⁻¹), especially at low B₁ amplitudes, saturation is less efficient because exchange occurs before the solute proton can be efficiently saturated. Still, achieving high saturation efficiency should not be the overriding concern in all cases. As shown in Figure 5B and Eq. 1, high saturation efficiency but low exchange rates lead to low PTR or CEST contrast. Faster exchange rates lead to lower saturation efficiencies but many exchange-transfer events with partially saturated protons can have a large cumulative effect on the water signal. Separately, post-processing algorithms can simplify CEST parameter measurement. For example, the QUAsi-Steady State (QUASS) CEST solution⁷³ minimizes the influence of t_dur and t_delay on CEST effects, thus only B₁ needs to be optimized for quantifying CEST parameters when using QUASS thus allowing for a broad choice t_dur and t_delay values.

An important consideration for pulsed-CEST experiments is that shorter RF pulses result in the perturbation of a wider spectral bandwidth (Figures 6A, B) and the creation of off-resonance pulse artifacts (Figures 6C-F).⁷⁴ Wider pulse bandwidths are unfavorable for CEST experiments because they generate competing signals from unwanted sources (e.g., other proton pools in the same or other molecules, direct water saturation) and broaden components in the Z-spectrum (see Figure 2 showing the Z-spectral dependence on pulse bandwidth via higher pulse amplitudes and Figure 6A showing the Z-spectral dependence on pulse bandwidth via shorter pulse durations). The B₁-amplitude spectrum for rectangular pulses of different pulse durations in the frequency domain is shown in Figure 6B, showing that for pulses as short as 50 ms, most of the saturation is limited to a narrow frequency bandwidth but that for 10 ms pulses, the sidebands can create ripples in the Z-spectrum which are particularly noticeable around the solute peak in Figure 6A.

A separate issue caused by wide bandwidth pulses is that away from their resonant frequency, the nutation axis of RF pulses is tilted out of their intended plane thus causing a change in the magnitude of the rotation frequency. For prolonged RF pulses intended for saturation, a tilt in the effective field causes a wobble in the spin polarization (see the off-resonance effect of rectangular-shaped pulses in Figure 6C) which can resemble CEST signals in the Z-spectrum and thus can be misattributed to a CEST peak. For saturation pulse trains, off-resonance effects can be alleviated using shaped RF pulses (Figure 6C), adding crusher gradients during the inter-pulse delay periods, or accounting for the accumulation of the phase offset over the course of the RF pulse train (Figure 6D). Herz et al.⁷⁵ noted that oscillations can occur if the phase of the magnetization vector after the previous pulse and the phase of the current saturation pulse are different. Further optimization of RF pulse shapes using optimal control theory can alleviate artifacts and superfluous signals while maintaining CEST contrast.^74,76,77

3.3. Saturation-based editing techniques

For many CEST agents, particularly diamagnetic compounds (diaCEST agents) which resonate over a narrow frequency band in the proton MR spectrum and usually within 6 ppm from water, RF irradiation at the agent resonance frequency coincidently perturbs other signal-generating components (see Section 2.3). Consequently, extraction and quantification of specific proton signals in vivo is nontrivial. For saturation-based labeling techniques, signal specificity can be improved by utilizing signal editing methods that modulate the CEST signal using a combination of acquisitions with various saturation pulse amplitudes, durations, phases, and saturation frequencies, as well as inter-pulse delays. CEST signals from solute proton pools with different intrinsic properties (e.g., exchange rates) will vary depending on changes to these pulse sequence elements, and hence signal editing can be used to separate different contributions.

3.3.1. Adjusting RF pulse duration and amplitude

Figure 3C.1 shows the saturation efficiency (α) of solute protons with different exchange rates, which after transfer determine the CEST signal intensity in the Z-spectrum. The effect of varying saturation pulse amplitude and duration for the selection of CEST signal⁷⁸ of solute molecules is shown in Figure 7. The CEST contrast in Figure 7 was based on Bloch simulations using all the pools listed in Table 1. Contrast for each solute is the amplitude of the signal of interest calculated by taking the Z-spectral difference at the solute frequency with and without the solute pool. Slow-exchanging protons (e.g., backbone amide protons in mobile proteins and peptides, Table 1, Figure 7A) can be almost fully saturated at low B₁ amplitudes (~0.5 μT) and sufficiently long saturation time. At longer preparation times, which allow more saturation exchange transfer events to occur, the CEST contrast will reach a plateau as longitudinal relaxation of the water pool results in a saturation steady-state of the water pool. Faster-exchanging protons, such as guanidinium protons in creatine (Table 1, Figure 7C), require higher pulse amplitudes (>2 μT) to achieve more saturation before exchange occurs (Eq. 2), but many saturation transfer events can occur in a short amount of time and maximal CEST contrast can be achieved at short t_dur times. Thus, for creatine guanidinium protons and large protein amide protons, the overlapping signals can be separated by using different saturation pulse parameters. Also note that higher pulse amplitudes and durations can diminish the solute signal amplitude due to spillover dilution effect from other Z-spectral components (e.g., DS⁴⁹ and MTC^5,52,53). The very fast-exchanging glutamate protons have low saturation efficiency thus low CEST contrast (Table 1, Figure 7D), but their contribution can be modulated using a combination of different pulse parameters (saturation-based editing discussed below) thus possibly allowing signal extraction. For example, Jin et al. developed the “average saturation efficiency filter” (ASEF) method which utilizes these saturation parameters to achieve exchange rate filtering,^79,80 separating out signals of intermediate-exchanging protons from fast-exchanging protons and MTC. Zu et al. also compared CEST signals acquired using CW- and pulsed-saturation schemes to filter out unwanted components.⁸¹

3.3.2. Alternating RF irradiation to remove short T₂ components

Immobile tissue components have protons with very short transverse relaxation times (μs range, Table 1) and thus wide resonances, which appear as background in the Z-spectrum and overlap with other signal components. Lee et al. suggested that alternating the saturation pulse parameters could be used to isolate these strongly-coupled spin systems,^82,83 based on the fact that these very broad peaks are less sensitive to saturation offset frequency compared to narrow resonances from mobile molecules. Hence, applying a saturation pulse at the mobile CEST agent frequency can be compared to a reference frequency that does not perturb the CEST agent but still saturates immobile components to approximately the same degree. The difference in the signal can be used to separate mobile and immobile components.

A more detailed approach based on this principle is the SAFARI (saturation with frequency alternating RF irradiation) method developed by Scheidegger et al (Figure 3C.2a).^84,85 Assuming that slow-exchanging protons (e.g., amide) are almost completely saturated at low pulse amplitudes and their signal amplitude will remain relatively constant with increasing B₁ (Figure 4A), however, water saturation due to protons in immobile macromolecules and DS will continue to increase at higher pulse powers. SAFARI aims to extract CEST signals by comparing images acquired at the solute resonance frequency S_sat(+Δω) (containing CEST, DS, and MTC signals), S_sat(−Δω) (containing only DS and MTC signals), and a dual frequency irradiation S_sat(SAFARI) interleaved +/−Δω, containing CEST, DS, and MTC components and the pulse power is only half that of S_sat(±Δω). A quantification map MTR_SAFARI, similar to magnetization transfer ratio asymmetry (MTR_asym)^32,86,87, is generated,

$$\begin{gathered} MT{R}_{SAFARI}=[2S_{sat}\left(SAFARI\right)-S_{sat}\left(+\text{Δω}\right)-S_{sat}\left(-\text{Δω}\right)]/S_0 \\ =2\left(CEST+\frac{1}{2}DS\left(+\text{Δω}\right)+\frac{1}{2}MTC\left(+\text{Δω}\right)+\frac{1}{2}DS\left(-\text{Δω}\right)+\frac{1}{2}MTC\left(-\text{Δω}\right)\right) \\ -\left(CEST+DS\left(+\text{Δω}\right)+MTC\left(+\text{Δω}\right)\right)-\left(DS\left(-\text{Δω}\right)+MTC\left(-\text{Δω}\right)\right) \\ =CEST \end{gathered}$$ (3)

However, SAFARI assumes that only a single solute pool is selectively and almost fully saturated with minimal perturbation of other components. This is almost impossible in vivo where multiple CEST resonances overlap. Also, S_sat(−Δω) images will contain rNOE signals that will resemble the saturation pulse dependence of CEST signals and thus will be intermixed when using the MTR_SAFARI quantification (Eq. 3). Another issue is that as background signal components (DS, MTC) increase, CEST and rNOE signals become diminished due to the attenuation of the water signal (e.g., spillover). This behavior is not accounted for and thus complicates image interpretation. Further, inhomogeneously broadened immobile components (e.g., membranes) may not be suppressed^88–90. Overall, SAFARI still mixes signal components and are therefore less useful.

Several groups have also used dual-frequency irradiation to study different T₂ components. Narvainen et al. introduced a method called Z-spectroscopy with Alternating-Phase Irradiation (ZAPI), which periodically modulates the phase and/or amplitude of RF pulse (Figure 3C.2b).⁹¹ Components with different T₂ exhibit different timescales of coherence loss with respect to modulation, so ZAPI allows for selectively detecting different T₂ components. For example, mobile molecules with longer T₂ can be efficiently saturated by setting a sufficiently long modulation period (τ, the period of the modulated RF pulse).⁹² Shorter τ will only efficiently saturate shorter T₂ components (e.g., immobile molecules).⁹¹ However, ZAPI was only applied to detect macromolecular protons with short T₂, and the feasibility of selectivity measuring exchangeable protons has not yet been demonstrated. Still, this method needs further development as the low B₁ power used is unsuitable for fast-exchanging protons and the selection of long-T₂ components (CEST) can be challenging. More recently, Goerke et al. developed a dual-frequency irradiation CEST (dual-CEST) method to provide unique specificity to mobile proteins with removing interference from MTC asymmetry.⁹³

A major impediment to the extraction of CEST signals is that the broad MTC component is not symmetric about the DS signal in vivo.^88,89 Consequently, conventional CEST quantification techniques such as MTR_asym have a non-zero background thus complicating signal interpretation. To balance out MTC asymmetry and generate a uniform MTC background (uMT) for the simple extraction of CEST signals, Lee et al. proposed a dual-frequency irradiation scheme^82,94,95 that uniformly saturates broad spectral components when both frequencies simultaneously irradiate the peak (Figure 3C.2b and Figures 8A, B).^94,95 Sweeping the saturation offset about the center of the two frequencies provides a Z-spectrum containing a uniform MT peak and narrow CEST peaks which are not saturated by both frequencies at the same time. Asymmetry analysis of the resultant spectrum can be performed to extract CEST effects (Figures 8A, B). Lee et al. showed that the negative contrast often measured using conventional CEST methods was removed with uMT in the human brain at 7 T (Figure 8C). The removal of negative MTR_asym contrast was most noticeable between 800 to 1200 Hz from water which largely corresponds to the APT frequency range.⁹⁵ Additionally, uMT aided the detection of glycosaminoglycans (gagCEST) in human knee cartilage and demarcated glycosaminoglycan measurements from cartilage and synovial fluid regions.⁹⁴

Figure 8.

Illustration of the uniform MTC (uMT) method. (A) Simulated Z-spectra obtained with the uMT method (red) and conventional CEST (black). “CEST” and “MT” respectively indicate the estimation of the CEST and MT contrast in the Z-spectrum. (B) By assigning the water resonance as 0 ppm in the MTR_asym spectrum of uMT, the negative peak appearing up-field from water includes only the CEST effect because background MTC effects become uniform and are nulled using asymmetry analysis. f_m, the frequency offset of RF irradiation, black line: CEST + asymmetric MTC, red line: only CEST, blue dash line: only asymmetric MTC, green dash line: the position of shifted water resonance. (C) The MTR_asym maps obtained by conventional CEST and uMT at different frequency offsets for a healthy subject at 7T. MTR_asym values of uMT were negated and the frequency offset was replaced with −Δω. Reproduced in part with permission from Lee et al., Scientific Reports. 2013;3:1707 and from Lee et al., NeuroImage. 2014;95:22-28.

Eq. 1 shows that the amplitude of Z-spectral components will vary due to saturation pulse duration (t_dur). Also, CEST effects are frequency offset (Δω) specific. Similar to variations in pulse amplitudes (see SAFARI section), DS and MTC effects will increase with saturation time but the magnitude of this effect is relatively symmetric about the water resonance. Conversely, the CEST signal will vary depending on both the saturation time and frequency offset from the water peak. Song et al.⁹⁶ exploited the length and offset varied saturation (LOVARS) approach to model this dependence after repeated image acquisitions. For example, a four-image scheme was employed (Figure 3C.2c):

$$\left[S_1,S_2,S_3,S_4\right]=\left[S_{sat}\left(-\text{Δ}\omega ,t_{dur,2}\right),S_{sat}\left(-\text{Δ}\omega ,t_{dur,1}\right),S_{sat}\left(+\text{Δ}\omega ,t_{dur,2}\right),S_{sat}\left(+\text{Δ}\omega ,t_{dur,1}\right)\right]$$ (4)

where t_dur,1 is longer than t_dur,2. S_sat(±Δω) is the saturated image acquired at the on-resonance position or the opposite position of the exchangeable proton. By assuming symmetric MTC and DS effects, this four-image scheme is repeated to create a cosine modulation of water signal, including different phases which correspond to asymmetric and symmetric components (Figure 9A). This allows for separating the CEST effects from DS and MTC components by using time-domain analysis methods, such as fast Fourier transform, or principal component analysis (PCA). In a study on mice bearing brain tumors, Song et al. showed that LOVARS acquisitions with saturation pulse durations of 3 s and 0.8 s could be used to generate APT images with minimal MTC and DS interference, which with higher contrast-to-noise (CNR) for the tumor region when compared to MTR_asym (Figure 9B).⁹⁶ In addition, the effects of B₀ field inhomogeneity could be removed due to differences in how it affects the LOVARS signal. This approach was extended to a multi-echo readout that allows the acquisition of multiple saturated images within a single TR (Figure 3C.2d and Figures 9C, D).⁹⁷

Figure 9.

Illustration of the Length and Offset VARied Saturation (LOVARS) and Multi-Echo Length and Offset VARied Saturation (MeLOVARS) methods. (A) The acquisition scheme of LOVARS. (B) LOVARS results for a mouse bearing a 9L gliosarcoma at 9.4T. Reproduced with permission from Song, et al. Magn. Reson. Med. 2014;72,471-476. (C) The change of MTR_asym as a function of module number in MeLOVARS acquisition. (D) MeLOVARS MTR_asym maps for 5 modules for a mouse bearing a glioblastoma at 11.7T. Reproduced with permission from Song, et al. Magn. Reson. Med. 2015;73, 488-496.

3.3.3. Modulating inter-pulse delays to filter out unwanted components

Inter-pulse delays in the pulsed-CEST sequence are often necessary to avoid exceeding hardware RF duty cycle and SAR limitations. Xu et al. showed how this delay can also be used to act as a filter for removing selected components based on their exchange and/or transverse relaxation rates.⁹⁸ The Variable Delay Multi-Pulse (VDMP) approach employs the pulsed-CEST scheme with varied delay (t_exch, i.e., t_delay) between RF pulses, as shown in Figure 3C.3a. Specifically, the pulse sequence consists of a block of selective 180° pulses with duration t_dur and inter-pulse delay t_exch, followed by a signal readout module. Depending on the RF pulse length, these labeling pulses can be classified as excitation (coherent transfer of magnetization) or saturation. At long t_exch periods, both fast and slow-exchanging protons have sufficient time to exchange to the water pool but at shorter t_exch periods, the transferred magnetization is mostly a result of fast-exchanging protons. The signal dependence with different t_exch can be modeled (Figure 10A), and experiments conducted with specific t_exch values to maximize signal differences or null a particular component.

Figure 10.

Illustration of signal editing using the Variable Delay Multi-Pulse (VDMP) method. (A) Simulations of (top) normalized VDMP-CEST signal S(t_exch)/S(t_exch = 0) at various delay time (t_exch) and exchange rates (k_sw), and (bottom) the projections of VDMP-CEST signal as a function of t_exch. (B) (top) Z-spectra of white matter (WM) and gray matter (GM) acquired using t_exch = 0 ms and 100 ms at 7T, and (bottom) the VDMP difference (ΔVDMP) spectra for WM and GM. (C) The ΔVDMP maps for (top) APT and (bottom) rNOE for human brain. Reproduced with permission from X. Xu, et al. Magn. Reson. Med. 2016,75,88-96.

Several groups have employed the VDMP method to separate different components with varied exchange rates.^99–101 For instance, Xu et al. demonstrated that the relative intensity of the combined MTC component in the white and grey matter was the same at t_exch = 0 ms and t_exch = 100 ms.⁹⁹ Therefore, subtracting separate images at these two t_exch values will result in the nulling of MTC (Figure 10B). The APT and rNOE components with different transfer rates are not nulled at these times and thus residual APT and rNOE maps can be constructed (Figure 10C). VDMP relies on different signal buildup curves to separate CEST components. However, for very fast-exchanging components, exchangeable protons exchange to water before labeling is complete, thus no signal buildup is observed and only signal decay from zero t_exch. Shorter labeling pulses can more rapidly label these fast-exchanging protons but will be less frequency selective (see Section 3.2).

4. Excitation-based labeling

RF saturation is the most common method for labeling solute components in CEST experiments and varying saturation parameters has allowed for the editing of Z-spectral components. However, as mentioned in Section 2.1, RF excitation is an alternative approach for effectively labeling spins in short well-defined periods. As will be detailed below, this category of labeling techniques offers unique advantages over saturation-based techniques for certain CEST applications.

The application of an RF pulse with a duration t_dur, much shorter than the longitudinal and transverse relaxation times of a proton, will result in the coherent rotation of magnetization of a solute pool about the direction of the pulse. Similar to saturation transfer, this perturbed magnetization of solute can subsequently transfer to water via exchange, cross-relaxation, or a combination of both. A single instance of label transfer from a low-concentration solute to water is insufficient to change the water signal markedly so excitation-based labeling processes are repeated multiple times to provide signal enhancement. Each label-transfer element is defined as a label-transfer module or LTM (Figure 11).^4,102 LTMs can consist of several different forms of excitation labeling including inversion (Figure 11A), rotation (Figure 11B), dephasing (Figure 11C), frequency encoding (Figure 11D), inter-pulse delay modulating (Figure 11E), excitation with stimulated echo selective refocusing of exchanged solute protons (Figure 11F), or include both excitation and saturation labeling (Figure 11G). Each labeling method offers a unique set of possibilities for signal enhancement, molecular specificity, and spectral editing.

Figure 11.

Excitation-based labeling methods using label-transfer modules (LTMs). Within LTMs, the solute protons are rapidly labeled by several different forms of excitation labeling, including (A) frequency-selective inversion, (B) multiple rotation angles, (C) dephasing, as well as (D) multiple excitations with encoding frequency of protons over evolution time (t_evol), (E) excitations with varied inter-pulse delays, (F) solute-selective excitations with a pair of frequency selective excitation pulse separated by gradient G_s that are used for stimulated-echo refocused acquisition. Alternatively, proton pools can be labeled by both excitation and saturation pulses (G, H). (G) After the labeling period, the labeled protons are transferred to water. LTMs are repeated multiple times to enhance the effect on bulk water. After preparation, the signal of bulk water is acquired using an imaging pulse sequence. (H) The positive CEST (pCEST) is a hybrid labeling method by adding excitation (inversion) before the long saturation period. “Acq” is the acquisition module. Reproduced in part with permission from van Zijl and Sehgal, eMagRes.2016:1307-1332.

4.1. Inversion, Rotation, and Dephasing-based labeling

Inversion-based labeling uses a 180° RF pulse within each LTM to invert the magnetization of solute protons, followed by a delay (t_exch, i.e., t_delay) for this negative Z-magnetization to transfer to the water pool (Figure 11A). Since this inversion results in the labeled protons having a magnetization opposite to the equilibrium water magnetization, the effect of inverted solute protons on the water pool is twice as large as for saturated protons where the signal is nulled (Figure 4B). Labeling efficiency depends on the duration of the pulse relative to the exchange rate. Faster-exchanging protons require shorter inversion pulses for efficient labeling but this can be challenging for human scanners which are constrained by RF hardware and SAR limitations. Furthermore, shorter RF pulses result in wider pulse bandwidths which limit molecular specificity (see Section 3.2). Wider bandwidth pulses are more robust to B₀ field inhomogeneities but conversely, excitation-based approaches (which require exact flip angles) are more susceptible to B₁ field variations when compared to saturation-based labeling.

When inversion pulses are much longer than T₂ (e.g., for immobile MTC components with microsecond T₂) or the bound lifetimes, the transverse magnetization component during rotation is dephased before the full longitudinal component can be inverted by the RF pulse. This concept was utilized in the chemical exchange rotation transfer (CERT) approach to selectively detect slow transfer processes such as APT and rNOEs in vivo.^103–106 In CERT experiments (Figure 11B), exchangeable solute protons are inverted and the effect of inversion transfer is compared to a reference scan where the magnetization of the solute protons is rotated 360° under constant B_avg,power, which is defined as the square root of the mean square irradiation field over the entire pulse train period ($B_{avg,power}=\sqrt{\frac{1}{t_{pd}}{\int }_0^{t_{pd}}{B}_1^{2}dt}$):

$$MT{R}_{double}={\frac{S_{sat}\left(360°\right)-S_{sat}\left(180°\right)}{S_0}|}_{B_{avgpower}}$$ (5)

where MTR_double is the metric for quantifying the exchange effect and t_pd (= t_dur + t_exch) is the length of LTM. Under different flip angles with constant B_avg,power, broad components with faster exchange rates or short T₂ values (e.g., MTC) are dephased instead of being inverted and their contribution to S_sat(360°) and S_sat(180°) is similar (Figure 12A). In addition, water is continuously irradiated by the pulse train containing hundreds of pulses prior to each acquisition and thus water eventually becomes saturated to a similar extent under the same B_avg,power. Therefore, DS and MTC effects are approximately equal for both S_sat(360°) and S_sat(180°) acquisitions and can be subtracted. CERT was shown to be specific for APT and rNOE signals in a brain tumor model in rats after the MTR_double effectively removed DS and MTC signals (Figures 12B, C).¹⁰⁵ The CERT method can be further extended to spectral editing by utilizing multiple excitation pulse parameters (e.g., flip angle, inter-pulse delays)^107–110 and subsequent modeling of the water signal modulation as a function of the pulse parameters. Still, CERT is best suited to slow transfer processes where long-narrow bandwidth pulses can effectively invert and rotate solute protons. As mentioned above, faster-exchanging protons require shorter RF pulses for efficient labeling which limits molecular specificity due to the coincident labeling of multiple components. The concept of rotation-based excitation was also utilized to develop the “On resonance PARamagnetic Chemical Exchange Effects (OPARACHEE)” method to measure PARACEST agents, but the approach is somewhat opposite in that the water signal is continuously rotated by 360°, but water signal reduces because of fast exchange with the contrast agent during rotation.^111–115

Figure 12.

Illustration of the chemical exchange rotation transfer (CERT) method. (A) Simulated CERT data as a function of flip angle θ. The labeled signal (S₋), including CEST, DS, and MTC effects, varies with θ, while the reference signal (S₊) represents only DS, MTC effects and is independent of θ. Reproduced with permission from Zu et al., Magn. Reson. Med. 2013; 69: 637-647. (B) Z-spectra and magnetization transfer double angle ratio (MTR_double) spectra of rat brain-bearing tumor at 9.4T. (C) The corresponding MTR_asym and MTR_double maps. Reproduced with permission from Zu et al., Magn. Reson. Med. 2014; 72: 471-476.

A separate approach is to intentionally dephase solute protons using a single 90° RF excitation pulse followed by a pulsed gradient. The dephased solute signals will reduce the water signal intensity after transferring to the water pool (Figure 11C). Fast-exchanging protons may not need the dephasing gradient at all because these protons transfer to water quickly where they dephase slowly with $T_{2w}^{*}$ (see CERT description).

4.2. Frequency encoding: modulating signal evaluation to edit multi-type protons

The inversion, dephasing, and rotation-based labeling methods detailed above are based on preferably selectively exciting a single solute pool prior to excitation transfer to water. In order to achieve such molecular specificity, these narrow bandwidth pulses need to be relatively long in duration which limits these approaches to slow transfer processes because efficient labeling requires t_dur <1/k_sw. Long labeling pulses remove a key advantage of excitation transfer which is the ability to label solute components rapidly. These excitation labeling techniques also mirror saturation transfer methods for acquiring exchange spectra (i.e., sweeping the labeling pulse offset frequency in a series of acquisitions). An alternative approach to specifically encode the chemical shift of solute components with short labeling pulses is frequency-labeled exchange transfer (FLEX).^{102,116–120} FLEX encodes the chemical shift of solute molecules using a binomial pair of excitation pulses separated by a period for chemical shift evolution (Figure 11D). The first θ_x pulse rotates the magnetization of multiple pools into the transverse plane where the magnetization of each component accumulates a phase depending on the frequency difference (Δω_s,o1) between the solute and the frequency offset of the pulse (O1). After the evolution period (t_evol), the magnetization of each pool along the minus Y-axis is flipped back to the Z-axis by the second θ_-x pulse. This binomial labeling period within each LTM is followed by a delay t_exch to allow the labeled solute protons to exchange to water. Crucially, the labeling offset frequency in FLEX experiments is often kept fixed and the t_evol period is varied. As a result, the water signal (I_w,_ex) modulates as a function of t_evol since each solute component (S) will accumulate a different phase change with each t_evol:

$$I_{w,ex}={\sum }_{s}PT{R}_s\cdot \text{exp}[-(k_{sw}+1/T_{2s}^{\text{*}})\cdot t_{evol}]\cdot \text{cos}(\text{Δ}{\omega }_{s,o1}\cdot t_{evol}+{\varphi }_s)$$ (6)

where PTR_s is the proton transfer ratio of each solute proton (Eq.1), $T_{2s}^{\mathrm{*}}$ is the effective transverse relaxation time of the solute proton, ${\varphi }_s$ is the phase of modulation. The resulting water signal, containing a superposition of modulations from each solute, can be Fourier transformed to produce a spectrum containing solute components or analyzed using time-domain analysis approaches (Figure 13A).^116,119 As seen from Eq. 6, k_sw can be extracted for each component from a single FLEX experiment. Another FLEX feature is that since water will also modulate depending on Δω_o1, B₀ inhomogeneity can be corrected by using this internal water frequency. By varying the timing parameters (t_dur, t_evol, t_exch), FLEX can be used as an exchange rate filter that limits which protons get labeled and transferred to water during an LTM (Figures 13C, D).

Figure 13.

Illustration of the frequency-labeled exchange transfer (FLEX) method. (A) In vivo FLEX signal of mouse kidney after injection of paraCEST agent (EuDOTA-(gly)₄⁻) decays with evolution time (t_evol), the corresponding FLEX spectrum after Fourier transfer, the time domain analysis, and the spectral components of the FLEX signals. (B) In vivo paraCEST map of mouse kidneys at 9.4T. Reproduced with permission from Lin et al., Magn. Reson. Med. 2014;71:286-293. (C) The simulated proton transfer rate (PTR) as a function of delay time (t_exch) within single label transfer module (LTM) at 7 T. T_1w/T_2w = 1.6 s/62 ms, T_1s/T_2s = 1 s/ 100 ms, [H_s]/[H_w] = 6.5×10⁻⁴, t_dur = 0.2 ms, t_evol = 0 ms. (D) On-resonance labeling efficiency for different pulse durations (t_dur) and exchange rates (k_sw). (E) The comparison between off- and on-resonance FLEX for thymidine solutions with different pH at 11.7T. Reproduced with permission from Yadav et al., Magn.Reson. Med. 2012;68:1048-1055.

In principle, FLEX labeling pulses can be as short as the RF hardware and SAR limitations allow. Shorter pulses allow fast transfer processes such as paraCEST agent exchange and cross-relaxation between immobile components to be efficiently labeled.¹¹⁹ For instance, Lin et al. showed that water exchange with EuDOTA-(gly)₄⁻, immobile bound water (MTC), and free water components could be selectively detected in vivo using FLEX labeling pulses of 29 μs (Figures 13A, B). The fitted FLEX exchange rates in the mouse kidneys were 15.5 × 10³ s⁻¹ for the EuDOTA-(gly)₄⁻ paraCEST agent and the dephasing of the MTC pool ($k_{sw}+1/T_{2s}^{\mathrm{*}}$) was approximately 50 × 10³ s⁻¹.

Even with FLEX experiments, pulses must be designed to not spuriously excite too much water. Significant excitation of the water pool can result in additional echoes and large water modulation which hampers the extraction of smaller solute signals.¹¹⁷ The original FLEX implementations set the labeling pulse frequency (o1) so the exchanging protons of interest experienced a 90° flip angle but the effective field at the water resonance resulted in a 360° flip angle which effectively nulled water.¹⁰² For solute pools close to water, satisfying this condition again requires longer pulse durations with the associated labeling efficiency/specificity issues discussed above. Yadav et al. proposed placing the FLEX labeling pulses directly on the water resonance (on-resonance FLEX) which results in a jump-return of the water magnetization but other solute pools away from water experience chemical shift evolution and thus are frequency encoded.¹¹⁷ By adjusting the pH of thymidine solutions, diamagnetic imino proton exchange rates of up 10,000 s⁻¹ could be imaged with high sensitivity (Figure 13E).¹¹⁷

Excitation pulses have much larger amplitudes compared to saturation pulses. A long train of such pulses may encounter SAR and hardware limitations when using a large number of LTMs. In such cases, longer repetition times can be avoided by reducing the number of LTMs. Yadav et al. used 100 LTMs for FLEX imaging of the human brain at 3 T ¹¹⁸ but a lower number of LTMs may also be sufficient depending on the application. Furthermore, B₁ fields can be reduced using lower flip angles for the FLEX binomial pulse pairs. For example, compared to 90° pulse pairs, a 30° flip angle may be more practical because the pulse requires only 1/3 of RF pulse area but will still generate transverse magnetization of 50% while alleviating RF concerns.

4.3. On-resonance VDMP: applying RF pulses on water resonance to efficiently label fast-exchanging protons

The VDMP method discussed in Section 3.3.3 has limited ability to measure fast-exchanging protons with frequency selective pulses. To address this, Xu et al.¹²¹ proposed an on-resonance VDMP method (Figure 11E), which applied a train of high-B₁ binomial pulses with alternating phases directly on the water resonance, resulting in a jump-return of the water signal (with minimal perturbation) but effective labeling of fast-exchanging protons (above 1 kHz) with the high B₁ strength inverting the fast-exchanging protons (also resulting in its classification as excitation-based labeling). For such an excitation pair when using 90° pulses, VDMP is basically a FLEX experiment, but with zero t_evol and variable t_exch. When using 180° excitation pulses, water saturation can be doubled. Although other frequency-selective inversion pulses can be used, binomial pulses are very suitable to minimize direct water excitation and are insensitive to B₁ inhomogeneity. Still, this approach can result in water dephasing during the binomial pulses when the magnetization is in the transverse plane. Depending on the binomial pulse durations and the number of repetitions (LTMs), this dephasing can generate $T_2^{*}$ water signal decay. Another issue with the on-resonance VDMP approach is the loss of frequency specificity since the off-resonance lobes of the binomial pulses extend over a wide frequency range and this can label multiple solute components.

4.4. RACETE: selective refocusing exchanged solute protons to achieve positive contrast

As mentioned in detail already, CEST signals are often mixed in with other background signals that complicate signal interpretation. Gutjahr et al. developed a Refocused Acquisition of Chemical Exchange Transferred Excitations (RACETE) method¹²², which allows for detecting positive contrast of exchanging protons with background suppression. As shown in Figure 11F, similar to the scheme of FLEX (Figure 11D)^{102,116–120}, RACETE employs LTMs, which consist of frequency selective 90°φ₁-90°φ₂ excitation pulse pairs applied on the solute resonance to excite and store solute magnetization in the longitudinal direction, and a gradient pulse G_s between the pulse pair which induces phase, following by t_exch to allow for chemical exchanging between solute and water pools. After repeated LTMs, signal enhancement can be achieved and the ultimate signal is detected selectively via water by using a 90° pulse applied on water resonance to refocus the exchanged magnetization and a G_s to dephase unwanted coherences. This stimulated echo (STE) detection results in positive RACETE contrast based on chemical exchange, while suppressing background signals, but the STE acquisition results in a 50% loss in signal. RACETE was successfully used to detect iopamidol and salicylic acid phantom simultaneously¹²², and to map the pH of a salicylic acid phantom by further exploring t_evol.¹²³. Similar to FLEX and VDMP, t_exch can be varied to filter components based on exchange rate. To date, RACETE has only been performed in vitro. In vivo studies will be challenging because RACETE relies on sweeping the offset frequency of the labeling pulses and RF off-resonance effects will vary for each point in the spectrum. Further complicating analysis in vivo will be the spurious excitation of other signal components or water, especially in the presence of B₀ inhomogeneity.

5. Hybrid labeling using both excitation and saturation

The methods discussed above utilize either RF saturation or excitation pulses to label solute protons. However, some groups have shown that a combination of excitation and saturation pulses can selectively detect certain groups of solute protons. For example, Friedman et al. used LTMs consisting of both excitation and saturation pulses to selectively enhance fast-exchanging components while simultaneously suppressing slow-transferring components (Figure 11G).¹²⁴ Here, an initial selective excitation is applied around water to simultaneously excite solute protons whilst avoiding perturbation of bulk water. A subsequent saturation pulse is applied and its frequency is swept in a series of acquisitions mirroring conventional CEST experiments. Label transfer from the initial excitation pulse period remains constant at all saturation frequencies. As shown in Figure 14A, for fast-exchanging protons, saturation transfer from that frequency will nullify the effect of excitation transfer and the water signal will be dominated by saturation transfer effects. Conversely, excitation transfer will dominate the water signal at saturation frequencies containing slowly transferring components and therefore contrast will be weighted towards faster-exchanging protons. This transfer rate edited CEST (TRE-CEST) method effectively suppressed slow rNOE processes in a BSA phantom whilst the APT signal was mostly unaffected by this exchange rate filtering (Figure 14B). Excitation pulse and saturation pulse parameters in the TRE-CEST experiment can be tuned to target specific exchange rates but applying such selective excitation pulses in vivo is challenging due to field inhomogeneity and RF hardware constraints.

Figure 14.

Hybrid labeling of excitation and saturation. (A) Simulated transfer rate edited (TRE)-CEST signal and bovine serum albumin (BSA) phantom data (black diamonds and circles) as a function of the number of label transfer modules (LTMs), and the Lorentzian difference (LD) spectra for this phantom. (B) The averaged LD spectrum of conventional (black line) and TRE-CEST (red line) of the BSA phantom, and the corresponding APT and rNOE images at 7T. Reproduced with permission from Friedman et al., J Magn Reson. 2015;256:4351. (C) Schematic of the “positive contrast generated by using CEST” (pCEST) approach. (D) The pCEST results of phantom containing agar and EuDOTA(gly)⁻₄ at 9.4T. Reproduced with permission from Vinogradov et al., J. Magn. Reson. 2012;215: 64-73.

Vinogradov et al. also developed a CEST technique combining excitation (inversion) and saturation to suppress background signals and measure CEST effects (Figure 11H and Figure 14C).¹²⁵ This so-called positive CEST (pCEST) method is based on the idea of off-resonance spin-lock, where an inversion pulse is placed before the saturation period. This pulse inverts the initial magnetization which then returns to its steady-state during the saturation period at a rate (i.e., apparent relaxation constant, R_1app) depending on whether RF saturation is applied at CEST agent frequency or not (applied at a reference frequency). The increased apparent relaxation rate in the presence of solute pool saturation is determined by measuring the water signal intensity at the inversion null time. Since the water signal will return to a steady-state more quickly in the presence of solute pool saturation, the signal difference is assigned as the pCEST signal. Although pCEST effects are relatively small when compared to MTR_asym contrast, MTC effects are nulled since these components are dephased instead of inverted (Figure 14D).

6. Conclusions

CEST MRI has emerged as a promising exchange transfer approach for the enhanced detection of low-concentration metabolites in vivo. CEST pulse sequences typically consist of three stages, including magnetic labeling, label transfer, and signal detection. The resulting Z-spectrum includes multiple signal components, including CEST, rNOE, DS, and MTC effects, which confounds quantitative analysis. Here, we detailed several approaches that were broadly based on saturation or excitation transfer. The parameters of these labeling methods can be tuned for improved specificity or sensitivity of signal detection in vitro or in vivo for pre-clinical and clinical studies. We believe that the possibilities of the principles outlined above are just starting to be explored and expect many new applications and sequences to be designed in the future.

Abbreviations

APT

amide proton transfer

BSA

bovine serum albumin

CEST

chemical exchange saturation transfer

CERT

chemical exchange rotation transfer

CNR

contrast-to-noise

CW

continuous-wave

DS

direct water saturation

FLEX

frequency-labeled exchange transfer

LOVARS

Length and Offset VARied Saturation

LD

Lorentzian difference

LTM

label transfer module

MeLOVARS

Multi-Echo Length and Offset VARied Saturation

MT

magnetization transfer

MTC

magnetization transfer contrast

MTR_asym

magnetization transfer ratio asymmetric

MTR_double

magnetization transfer double angle ratio

NOEs

nuclear Overhauser effects

PCA

principal component analysis

pCEST

positive CEST

PTR

proton transfer rate

OPARACHEE

PARamagnetic Chemical Exchange Effects

RF

radiofrequency

RACETE

Refocused Acquisition of Chemical Exchange Transferred Excitations

rNOEs

relayed nuclear Overhauser effects

SAFARI

saturation with frequency alternating RF irradiation

SAR

specific absorption rate

SET

stimulated-echo

TRE-CEST

transfer rate edited CEST

uMT

uniform MTC

VDMP

Variable Delay Multi-Pulse

ZAPI

Z-spectroscopy with Alternating-Phase Irradiation

ZAPISM

Z-spectroscopy with alternating-phase irradiation and sine modulation

Data availability statement

All materials are available on request; simulation code can be downloaded from a public repository, https://github.com/ChongxueBie/RF-labeling-techniques-in-CEST-MRI.