Review and consensus recommendations on clinical APT-weighted imaging approaches at 3T: Application to brain tumors

1 INTRODUCTION

Amide proton transfer-weighted (APTw) imaging is a molecular MRI technique that generates image contrast based on endogenous mobile proteins and peptides in tissue.^{1, 2} As a type of CEST imaging,³ the principles and applications of APTw imaging have been reviewed in several articles.^4-21 Key abbreviations and nomenclatures used in the field of APTw imaging are listed in Table 1. Data from numerous institutions worldwide have demonstrated that APTw imaging adds important value to the standard clinical MRI sequences in brain cancer diagnoses, such as the detection and grading of tumors,^22-37 the assessment of treatment effect versus tumor recurrence,^38-45 prognosis related to tumor progression and survival,^46-48 and the identification of genetic markers.^49-56 It is worth mentioning that brain tumor patients require frequent MRI exams, and the exposure to gadolinium (Gd)-based contrast agents has been indicated as a risk for people with moderate to advanced kidney failure and for Gd deposition in the brain.^57-59 Although promising, there are several remaining challenging issues for clinical APTw imaging.^11-14 These include scanner RF amplifier constraints, specific absorption rate (SAR) limits, low SNR, long scan times, a complicated contrast mechanism with multiple contributions, B₁ inhomogeneity, and the possibility of artifacts because of motion, B₀ inhomogeneity, and lipids.

The most concerning issue from a radiologist's point of view is that APTw signal intensities depend on the APTw pulse sequence features and parameters used, which may lead to differences in image contrast and in interpretation between sites. Currently, there are large variations in parameters used for APTw imaging in the literature and, except for 1 Food and Drug Administration (FDA)-approved product,^{60, 61} most vendor sequences are available only as “work in progress” (WIP) software (Tables 2 and 3). Differences in data processing strategies have further complicated the reproducibility and comparison of results between centers. Based on its demonstrated ability to enhance diagnostic specificity in brain tumor assessment, there is a need for the clinical APTw imaging community to work together to develop this emerging technology into an optimized, reproducible, and standardized approach.

Toward this goal, the 7th International Workshop on CEST Imaging (2018)⁶² featured a special session to discuss standardization involving the 3 main MRI vendors (GE Healthcare, Milwaukee, Wisconsin, USA; Philips Healthcare, Best, Netherlands; Siemens Healthineers, Erlangen, Germany), illustrating its high priority for this community. The organizing committee of the 8th International Workshop on CEST Imaging (2020)⁶³ established an APTw Imaging Subgroup to evaluate APTw MRI standardization for brain tumor imaging. The purpose of this subgroup was to review the development of clinical APTw imaging techniques on 3 T MRI scanners and, in collaboration with some of the other leading groups in the field, to provide consensus recommendations for APTw imaging of brain tumors. Because different clinical applications may require a different set of parameters for optimal contrast, this work is limited to the currently most-used application of APTw MRI, namely, the diagnosis of brain tumors and only for the field strength commonly applied to brain tumors in the clinical setting, 3 T. Consistent implementation of these recommendations by MRI vendors will allow more medical centers, worldwide, to routinely use this promising technology in their multi-center clinical trials and in daily clinical practice. This will help to ultimately achieve biomarker status for APTw MRI contrast in the assessment of brain tumors at 3 T.

2 BACKGROUND AND THEORY

In tissue, most proteins and peptides are present in μM concentrations, and all contain multiple backbone amide protons. The amide proton exchange rate measured using MR spectroscopy was found to be about 30 Hz in the original APT paper and has been assumed widely in APT studies since.¹ However, MR spectroscopy detects mainly the slowly exchanging protons, and the amide proton exchange rate likely has a range of tens to hundreds of Hz in vivo, as reported in later imaging studies.^69-71 These various amide groups resonate at a similar frequency (around 8.3 ppm in the ¹H spectrum, or with an offset $$$ \Delta \upomega $$$ of +3.5 ppm downfield from the water resonance). This leads to a large composite resonance in NMR spectroscopy that reflects a total accessible amide proton concentration of approximately 50-100 mM (hence, justifying the assumption that f_sk_sw < R_1w in Equation [1]).^{1, 69-71} The magnitude of the APTR effect in vivo resulting from these protons is typically on the order of a few percent of the bulk water signal. Although this is a small effect on the water signal, APTw MRI offers a large detection sensitivity enhancement for metabolites present in millimolar concentrations.

3 HISTORY OF THE MECHANISM AND THE EVOLUTION OF ITS UNDERSTANDING

Originally, APTw imaging was designed for in vivo imaging of mobile protein content in tumors and pH changes in tissue during ischemia (because of the strong dependence of the amide proton exchange rate on pH in the physiological range).^{1, 2} It exploits the concept that the CEST mechanism can detect changes in amide proton signals. These 2 applications are based on early spectroscopy studies that showed an increased amide proton signal in proton spectra during ischemia (because of slower exchange)⁷⁸ and very large amide proton signals in perfused tumor cells.^{78, 79} These hypotheses were confirmed in vivo where ischemia experiments in rats at 4.7 T showed a small decrease in MTR_asym spectra over a chemical shift range of 2.5 to 5 ppm from the water, with the clearest decrease at 3.5 ppm, attributed to reduced pH.¹ The first in vivo studies in animal tumor models at the same field and RF settings showed a broad, increased MTR_asym effect ranging from 1.5 to 5 ppm.² In that paper, the signal change over this range was attributed to backbone amide and other exchangeable protons, which are typically seen in this range in high-resolution NMR protein studies and in the MR spectra of perfused cells and brain. Because of the results of MR spectroscopy, the authors focused on the 3.5 ppm offset.

In early work,² the increased APTw signal in tumors (relative to normal brain tissue) was attributed to (1) the increased mobile amide proton concentration in the tumor associated with the increased concentration of mobile cytosolic proteins and peptides because of higher cellularity, and (2) the slightly increased amide proton exchange rate because of marginally higher intracellular pH (<0.1 unit, as reported from phosphorus MR spectroscopy studies in patients)^80-82 in tumor cells. This has been confirmed in subsequent studies^83-86 and is consistent with increasing protein concentrations in tumors, as revealed by proteomics^{87, 88} and by in vivo MR spectroscopy.⁸⁹ It has been reported that solid tumors have an acidic extracellular pH and a neutral-to-alkaline intracellular pH.^{90, 91} Notably, reduced acidic extracellular pH in tumors would substantially reduce the amide exchange rate and, therefore, the contribution of extracellular proteins and peptides. The intracellular origin is supported by a recent study showing that there was no correlation between the amide proton exchange rate and extracellular pH.⁹²

Over the years, the understanding of the APTw contrast mechanism has evolved. Several other possible effects that may contribute to the APTw hyperintensity in tumors include: (3) reduced semi-solid MTC_asym (3.5 ppm) in tumors because their Z-spectra are less asymmetric thanin white matter and gray matter. This effect becomes more visible when using a longer saturation time, with B₁ increasing from 0.5 to 3 μT.^{76, 93} (4) Decreased aliphatic rNOEs of mobile macromolecules at and around the opposite frequency offset −3.5 ppm.^{6, 73-75} Based on Equation (3), the MTR_asym-based APTw image signal has multiple contributions (primarily including aliphatic rNOE and MTC_asym). In the brain, $$$ \mathrm{MT}{\mathrm{R}}_{\mathrm{asym}}^{\prime } $$$(3.5 ppm) (≈ −rNOE − MTC_asym) is negative and reduces the MTR_asym (3.5 ppm). Therefore, this contribution is actually a synergistic factor that enhances the APTw image contrast of the tumor because of the reductions in aliphatic rNOE and MTC_asym in tumors relative to brain tissue.^94-96 (5) Downfield rNOEs (e.g., in aromatic residues). This possible confounder at and around +3.5 ppm downfield of water has been previously explored in studies of protein solutions, tissue homogenates, and brains in vivo, and will partially counteract the aliphatic rNOE contribution.^{97, 98} (6) Spillover and MTC dilution effects.⁶⁷ When performing the MTR_asym analysis, asymmetric effects relative to water are visualized, and symmetric background signals (water T₂ [T_2w]-based spillover and the symmetric portion of MTC) are removed under the zeroth-order approximation. However, the APT effect is always diluted by competing spillover and MTC effects, resulting in the fact that T_2w and MTC alterations may influence the magnitude of APTw contrast. (7) Contamination by T_1w changes⁶⁷ (but there is a nonlinear relationship between APTw and T_1w). As shown in Equation (1), the APT effect scales with T_1w. However, the effects of increased T_1w (decreased R_1w) and water content (denominator of f_s) in brain tumors are canceled out partially.⁸⁴ Further, apart from the T_1w recovery effect, there is an opposing T_1w effect on APTw through dilution effects, as longer T_1w leads to lower Z-spectra, therefore, increasing the dilution effects, which lowers the APT effect.⁹⁹ These 2 effects cancel each other out to some extent, depending on experimental RF settings. Simulation studies have shown that APTw increases with T_1w at lower B₁, but is roughly insensitive to T_1w or even decreases with T_1w at higher B₁.^{99, 100} In addition, the dependence of APTw MRI on T_1w can be reduced using non-steady-state saturation. Fortunately, when RF strength is approximately equal to 2 μT (as recommended for brain tumor imaging at 3 T), APTw intensity has been found to be reasonably robust against the change in T_1w.^{99, 100} Therefore, a correction for T_1w changes is not necessary for APTw imaging of brain tumors at 2 μT on clinical 3 T scanners. Of course, this robustness still has limits.^{101, 102} Because the significant T_1w effect of Gd contrast agents may bias APTw imaging, it is important to remember that the APTw acquisition should always be performed before the injection of contrast agents.¹⁰³ (8) CEST signals from exchangeable protons resonating nearby. At 3 T, most exchangeable protons of macromolecules and metabolites (such as amines and hydroxyls) are in the fast-exchange regime (k_sw » Δω_s), and their resonances start to coalesce with water. Reduced extracellular pH in tumors may reduce the exchange rates of extracellular amines and other fast-exchanging protons and make them become detectable.¹⁰⁴ It is worth noting that the linewidth of some nearby intermediate-exchange resonances (including those from guanidinium protons at 2 ppm, which have an exchange rate of about 800-1000 Hz^{105, 106} and, therefore, an effective linewidth of about 2.0-2.5 ppm at 3 T) may be sufficiently large to be partially irradiated and detected at 3.5 ppm.^{107, 108} (9) APTw effect from mobile proteins in liquefactive necrosis. Proteinaceous fluid compartments in tumors, such as liquefactive necrosis, would result in large hyperintensity on APTw images^{12, 24} because of the abundance of proteins and protein fractions at higher mobility. In addition, plausible protein denaturation processes would generate such a signal increase.^{109, 110} Notably, reduced dilution effects in liquefactive necrosis would lead to an apparently high APTw contrast between this compartment and normal brain tissue. To simplify radiology readings, a new APTw metric based on the background MTR value has been proposed to suppress APTw signals of large liquefactive necrosis.¹¹¹ However, a further validation is needed for this APTw metric. (10) APTw effect from mobile proteins in blood vessels and hemorrhage. Blood has high concentrations of hemoglobin in erythrocytes and albumin in plasma, and a higher pH (∼0.2 pH units, relative to brain tissue),¹¹² which would contribute to the increased APTw in well-perfused tumors because of induced angiogenesis.^{37, 113} In addition, like liquefactive necrosis, intratumoral hemorrhage would demonstrate large APTw hyperintensity, particularly at hyperacute and acute stages.^{24, 114, 115}

Given all these potential synergistic or competitive effects, the contributions of which are affected by experimental RF settings and analysis approaches, the question arises whether the chosen term “APT-weighted” imaging is still justified, because it implies that the APT component is the major source of this signal. With the above list as a reference and amide protons as one important contributor, this naming is valid under the careful choice of acquisition and analysis.^94-96 For clinical APTw imaging of brain tumors, we would seek a pragmatic compromise that not only effectively detects APT, but also can be standardized so that all other effects are consistent between vendors and studies. Notably, recent radiographic-histopathologic correlation studies^{26, 29, 43} have clearly demonstrated that MTR_asym(3.5 ppm) is a valid metric that adds clinical value to the imaging of brain tumors at 3 T. Further validation is still needed on different pathology types for the various recommended APTw approaches described below.

In light of the above list of possible confounding contributions, there is still a great need for continued work on CEST methods with increased signal specificity and/or parameter quantification, as such methods may ultimately have improved diagnostic use. Some of such pulse sequences and data analyses that attempt to separate APT effects from the other contributions in APTw imaging are discussed in Table 3 and “Data Processing” below. Additional methods that incorporate different (and sometimes radically different) acquisition protocols are reviewed elsewhere.^{116, 117} The current paper is focused solely on providing a reasonable standard for assessing brain tumors in daily clinical practice using a currently established MTR_asym-based APTw approach.

4 PULSE SEQUENCES

4.1 RF saturation approaches and parameters

Currently, no consensus-based APTw MRI pulse sequences or parameters are available for clinical MRI systems between different vendors, even for the most studied application of brain tumors. As mentioned above, the APTw signal depends on the APTw pulse sequence features and parameters used. Notably, the APTw signal is affected by dilution effects, the contributions of which vary with the saturation amplitude and time. Because of this dependence, to achieve reproducible APTw imaging contrast for brain tumors, a consensus choice of certain saturation parameters is needed. Based on the abundant literature of the last decade, and taking into account saturation time limitations for amplifiers on some equipment, we recommend:

$$$$ {\displaystyle \begin{array}{cc} APTw& ={MTR}_{\mathrm{asym}}\left(3.5 ppm,{B}_{1\mathrm{rms}}=2\ \mu T,\right.\\ {}& \kern1em \left.{T}_{\mathrm{sat}}>0.8\ s,{DC}_{\mathrm{sat}}\ge 50\%\right),\end{array}} $$$$ ()

where T_sat (during which saturation is applied and transfer occurs) may consist of different combinations of RF pulses and inter-pulse delays, DC_sat is the RF saturation duty cycle (DC), and B_1rms is the root-mean-square B₁ value of a saturation pulse train with duration t_p and inter-pulse delay t_d, defined as:

$$$$ {B}_{1\mathrm{rms}}=\sqrt{\frac{1}{t_{\mathrm{p}}+{t}_{\mathrm{d}}}{\int}_0^{t_{\mathrm{p}}}{\left[{B}_1(t)\right]}^2 dt}. $$$$ ()

Notably, Equation (4) provides some flexibility that is needed at this stage. However, to enhance reproducibility and the comparison of results across vendors and sites, our preferred and more specific recommendation as a long-term goal is:

$$$$ {\displaystyle \begin{array}{cc} APTw& ={MTR}_{\mathrm{asym}}\left(3.5 ppm,{B}_{1\mathrm{rms}}=2\ \mu T,\right.\\ {}& \kern1em \left.{T}_{\mathrm{sat}}=2\ s,{DC}_{\mathrm{sat}}\ge 90\%\right).\end{array}} $$$$ ()

The choice of these RF saturation parameters is based on the following rationales.

It has been shown that the APTw contrast in brain tumors (T_1w ≈ 1.5-1.6 s in the solid portion of gliomas at 3 T)^{118, 119} improves substantially (Figure 2A) when lengthening T_sat from 0.5 s to 2 s.¹²⁰ For clinical APTw imaging, a CW block pulse of several seconds is possible for RF saturation with transmit-receive head coils on a standard 3 T clinical system,²² similar to typical animal APTw experiments. However, most standard 3 T clinical systems use body coils for transmit, which results in stricter limitations on RF amplifier DC (typically 50%), saturation pulse length, and, to some extent, SAR. In the past decade, this has been addressed by a few different methods (Tables 2 and 3), such as pulse-train, time-interleaved parallel RF transmission (pTX), or pulsed steady state APTw MRI. The pulse-train pre-saturation (a train of pulses separated by brief delays) has been used in some early pre-clinical APTw studies^{1, 2} and in most 3 T clinical investigations.^121-128 The use of pulse trains can achieve a DC_sat >90% and a T_sat >0.8 s on most 3 T MRI scanners from different vendors, greatly alleviating the limitations in amplifier duty-cycles and saturation pulse lengths. Notably, in a recent study on a brain tumor patient using different pulse-train RF saturation modules, Herz et al¹²⁹ confirmed the decreased tumor APTw signal intensities with decreasing DC_sat (Figure 2B). To truly maximize the saturation efficiency, Keupp et al^{130, 131} introduced a time-multiplexed RF saturation method in pTX systems (Figure 3). By time-interleaving 2 RF sources of the body coil (the quadrature channels), each with a 50% idle-time, one can achieve an increased length of the pseudo-CW saturation pulse train, completely meeting the needs of CEST/APTw imaging. Finally, the pulsed steady-state CEST sequence, previously implemented on 1.5T¹³² and 7 T^{128, 129} human MRI systems,^{133, 134} but similar to the 3 T MTC steady-state sequence,¹³⁵ is currently being optimized for APTw MRI on 3 T clinical systems.^136-141 With this, the CEST effect is built up over multiple saturation pulses and each is followed by an imaging acquisition segment. However, the DC for such a pulsed steady-state sequence is inherently low.

The amide proton pool, which has a k_sw range of tens to hundreds of Hz in vivo, can be efficiently labeled using an RF saturation strength (B₁ or B_1rms) between 1 and 3 μT (1 μT = 42.567 Hz). Importantly, the use of 2 μT in MTR_asym-based APTw imaging of brain tumors provides some advantages. When a B₁ of 2 μT is used, the APTw signal is almost 0 for normal brain tissue (Figure 4A), because of the presence of a negative $$$ \mathrm{MT}{\mathrm{R}}_{\mathrm{asym}}^{\prime } $$$(3.5 ppm) compensating the APTR. In addition, the APTw signal should always be negligible in the cerebrospinal fluid (CSF) because of the symmetry of the CSF Z-spectrum. Therefore, using this B₁ and sufficient T_sat (0.8-2 s), the APTw images are zeroed in most normal brain areas, the ventricles and, in patients, the resection cavity (Figure 4B). This convenient background allows easy detection of hyperintense APTw signals in high-grade tumors or hypointense APTw signals in ischemic tissue, which is convenient for clinical assessment.¹⁴²

Four RF saturation types (types [a]-[d]) recommended for brain tumor APTw imaging on 3 T clinical MRI scanners are given in Table 4, in which types (a) and (b) are the 2 preferred options. For all pulse-train saturation types, we recommend B_1rms = 2 μT, T_sat = 0.8-2 s, and DC_sat ≥ 50%. This recommendation retains some flexibility for the parameters of shaped pulses (such as shape, length, and phase). Type (a) and several examples using type (b) are further compared in Figure 5A and B1-B3. Type (a) is the ideal approach and is allowed only with transmit-receive head coils, some state-of-the-art RF amplifiers, or single-slice APTw imaging protocols. Fortunately, the pulse-train methods in type (b) can be realized with the RF amplifier hardware configuration from all different manufacturers. Time-interleaved pTX is one good option, but not a requirement, because some state-of-the-art RF amplifiers can support a DC_sat >90% for body coils at the recommended B₁ strength for APTw MRI. The data in Figure 6 confirm that the pulse-train methods in type (b) can provide fairly similar Z-spectra, MTR_asym spectra, and APTw images in healthy volunteers from three 3 T MRI scanners of different vendors. When preferred types (a) and (b) are not feasible, we recommend using types (c) or (d). A good example for type (c) is shown in Figure 5C, which has been widely used in brain tumor studies previously.^{25, 30} A comparison of types (b)-(d) can also be found in Figure 2B. To allow maximal reproducibility, several of the pre-saturation schemes were recently shared in the open-source pulseq-CEST format¹⁴³ (see more details in the caption of Figure 5).

TABLE 4. Recommendations for APTw imaging of brain tumors at 3 T

Pulse sequences
RF saturation approaches and parameters: CW RF saturation, Tsat = 2 s, B1 = 2 μT (ideal/preferred); Pulse-train, Tsat = 2 s, B1rms = 2 μT, DCsat = 90-100% (preferred); Pulse-train, Tsat = 800-1000 ms, B1rms = 2 μT, DCsat = 90-100%; Pulse-train, Tsat = 2 s, B1rms = 2 μT, DCsat = 50%; Lipid suppression: An effective lipid suppression method (such as SPIR); Readout (including recovery time): Fast 3D acquisition (in-plane resolution 1.8-2.2 mm; through-plane resolution 3-6 mm); Trec ≈ 2T1w (tumor T1w ≈ 1.5-1.6 s at 3 T)
Acquisition protocols
B0 shimming, preferably 2nd-order shimming, should be done; At least 6-offset/7-point APTw imaging protocols (S0, ±3, ±3.5, and ±4 ppm or S0, ±3.1, ±3.5, and ±3.9 ppm) should be used. More acquisitions at ±3.5 ppm are often needed to increase the APTw SNR. A saturated image at ±300 ppm or further from water should be acquired and used as S0, with a dummy scan or shot required; Proper water frequency mapping must be acquired
Data processing methods
Use MTRasym (3.5 ppm) as a metric; Use a rainbow color scale (±5%) leading to a green background with yellow/orange/red hyperintensities and blue hypointensities; Both rainbow color and gray-scale images are stored

4.3 Readout

Clinical application of the APTw MRI approach will be more feasible if fast volumetric imaging acquisition (<5 min) can be achieved. Both multi-slice and 3D approaches have been used in CEST/APTw imaging. In multi-slice acquisitions, there generally are CEST signal losses because of T_1w relaxation differences based on the order in which the slices are acquired, so corrections may be needed.¹⁵⁰ However, in a 3D acquisition with centric encoding, each line of k-space contributes equally to all reconstructed slices, and differences in saturation caused by differences in T_1w relaxation between slices can be minimized.¹²² Therefore, 3D acquisition is preferred for volumetric APTw imaging. For instance, Zhu et al¹²² developed a 3D APTw MRI technology that allows fast acquisition on 3 T clinical instruments, using 4 elements of 200 ms with 10 ms space in between (DC_sat = 95%) for gradient- and spin-echo (GRASE) imaging with adiabatic lipid suppression pulses. This sequence has been applied successfully to clinical studies at many sites.^{30, 151}

Volumetric APTw MRI currently requires a scan time of 3 to 10 min, because of the use of multiple RF saturation frequencies (to correct for the B₀ inhomogeneity) and multiple acquisitions (to increase the SNR). Various acceleration approaches and fast imaging acquisition techniques have been used to accelerate CEST/APTw imaging (Figure 7), including SENSE,^{152, 153} GRAPPA,¹⁵⁴ controlled aliasing in parallel imaging results in higher acceleration (CAIPIRINHA),¹⁵⁵ GRASE readout,¹²² turbo-spin-echo (TSE) readout^{22, 23, 156} (including sampling perfection with application-optimized contrasts by using different flip angle evolutions [SPACE]),¹⁴⁹ and a gradient-echo (GRE)-based snapshot approach.^157-159 The combination of 3D TSE readout with time-interleaved pTX saturation provides a fast and more sensitive 3D APTw imaging sequence (Figure 7A), which was the method used in the first commercial APTw imaging sequence on Philips 3 T MRI scanners.^{60, 61} In addition, several novel undersampling acquisition and reconstruction approaches (including keyhole, spectroscopy with linear algebraic modeling, compressed sensing, and deep learning)^160-165 have been used successfully to accelerate CEST/APTw acquisitions. The reconstruction-oriented, reduced k-space acquisition requires more advanced data processing, but this can all be automated on the scanners (including compressed sensing-based image reconstruction and B₀ inhomogeneity correction)^{60, 61} to streamline the clinical workflow.

Based on previous studies, a fast 3D acquisition technique (in-plane resolution 1.8-2.2 mm; through-plane resolution 3-6 mm), integrated with a feasible, optimized RF saturation scheme and an effective lipid suppression method, and is recommended for brain tumor APTw imaging on 3 T clinical MRI scanners (Table 4). T_rec or TR is generally limited by the SAR, and we suggest using T_rec ≈ 2T_1w (tumor T_1w ≈ 1.5-1.6 s at 3 T).^{118, 119} GRASE,¹²² TSE,⁶¹ SPACE,¹⁴⁹ and GRE^157-159 have widely been used in previous studies of brain tumor APTw imaging and are among the candidate readout sequences. However, we recommend retaining some flexibility for the 3D readout module, currently, because the choice of this pulse sequence component does not affect the APTw contrast per se.

5 ACQUISITION PROTOCOLS

The minimum data required to calculate MTR_asym is related to a 2-offset (±3.5 ppm) APTw imaging protocol. However, MTR_asym analysis is complicated by B₀ frequency inhomogeneity and scanner instability, which causes spatio-temporal resonance frequency variations. The 2-offset protocol is highly susceptible to B₀ variations between voxels, which can be problematic near air-tissue interfaces, even with up to 2nd-order shimming.¹²² Two kinds of APT imaging acquisition protocols have been reported in the literature that address this issue (Tables 2 and 3): acquiring a full Z-spectrum consisting of downfield and upfield frequency offsets from the water resonance, or acquiring limited frequency offsets at and around ±3.5 ppm, plus a ΔB₀ map for frequency offset referencing. A full Z-spectrum is often acquired for research purposes and allows for correction of B₀ variations because it includes samples near water and amide resonances. The APTw signal is known to appear at the offset of 3.5 ppm; therefore, only limited offsets at and around ±3.5 ppm need to be acquired.¹² To have the possibility to correct for B₀ differences on a voxel-by-voxel basis, it is necessary to acquire multiple offsets at and around ±3.5 ppm. The ΔB₀ map is often obtained using an extra, more time-efficient scan, such as the widely used water saturation shift referencing (WASSR)¹⁶⁶ or GRE phase-based methods.^{156, 167} According to the literature,¹²² the B₀ inhomogeneity in the brain is typically less than 20 Hz and up to 100 Hz near air-tissue interfaces (ears, sinuses) at 3 T. With a ΔB₀ map available, it has been shown that a 6-offset APTw imaging protocol (±3, ±3.5, and ±4 ppm) with 2-to-4 acquisitions at ±3.5 ppm (Figure 8A), can provide B₀ inhomogeneity-corrected APTw images of sufficient SNR.²³ This minimal Z-spectrum data acquisition allows 3D full brain imaging within clinically acceptable acquisition times (e.g., <5 min). More acquisitions at ±3.5 ppm can be used to increase the SNR. Of course, a relatively larger offset interval (e.g., ±2.75, ±3.5, ±4.25 ppm) or extra offsets (e.g., ±2.5, ±3, ±3.5, ±4, ±4.5 ppm) may be used when large B₀ inhomogeneities exist.^{23, 168} The tradeoffs for these are larger data interpolation errors (see next section) and more scan time, respectively.

Recently, a number of novel methods have been proposed to correct B₀/B₁ field inhomogeneity or frequency drift in clinical CEST/APTw imaging at both 3 T and 7 T.^169-178 Keupp et al^{169, 170} developed a so-called CEST-Dixon method at 3 T that can map the intrinsic B₀ inhomogeneity using echo shifts during the APTw acquisition (Figure 8B), concluding that this self-corrected method is more robust than separate B₀ mapping approaches. Schuenke et al¹⁷¹ proposed the water shift and B₁ (WASABI) method that can yield simultaneous B₀ and B₁ mapping within about 2 min at 3 T and 7 T. The B₁ variations in the brain at 3 T are typically within ±10% and can be up to ±30% in some regions, such as in the infratentorial region and at the superior part of the brain.¹⁵⁹ Theoretically, B₁ inhomogeneity affects APTw signal when different brain regions experience varying amounts of saturation. For example, the saturation efficiency α in Equation (1) is a function of B₁, and spillover/MTC dilution effects change with B₁.⁶⁷ At 3 T, when a body coil is used, these effects taken together may not be sufficient to substantially affect APTw contrast within most brain slices (Figure 6). However, this may be an issue only in the infratentorial and superior brain regions, and further assessments are needed. Technically, if available, parallel transmit and B₁ shimming, as well as B₁ inhomogeneity corrections during processing (from B₁ mapping), should be used to reduce the possible issue. Poblador Rodriguez et al¹⁷⁷ recently compared several static and dynamic ΔB₀ mapping methods for correcting CEST MRI in the presence of temporal B₀ field variations at 7 T. The results indicated that, in the presence of frequency drift, the 3 dynamic methods (which integrate ΔB₀ mapping into the CEST measurement) had significantly improved ΔB₀-correction performance over established static methods. Self-corrected CEST-GRE-2TE (using phase data directly generated by double-echo GRE readout), comparable to CEST-Dixon, was the most accurate and straightforward sequence to implement.

The APTw acquisition protocol currently recommended for brain tumors on 3 T clinical MRI scanners is listed in Table 4. We recommend the acquisition of at least 6 offsets and an unsaturated image (S₀, ±3, ±3.5, ±4 ppm, or S₀, ±3.1, ±3.5, ±3.9 ppm), in combination with B₀ shimming (2nd order preferred) and a B₀ shift reference. Further offsets can be applied, if a larger B₀ inhomogeneity is expected. Because the APTw effect in vivo is often small (2-4% of the water intensity in tumor), multiple acquisitions are often needed to increase the APTw SNR. We recommend that S_sat(±3.5 ppm) images should have an SNR of at least ∼50, corresponding to 2-to-3 acquisitions for the TSE readout, with an in-plane spatial resolution of about 1.8 to 2.2 mm and a through-plane resolution of 3 to 6 mm. Notably, S₀ without RF saturation should be acquired using the same TR as saturated images. It is acceptable that a saturated image at a very large offset is used as S₀, which avoids potential drift effects with RF power changes. In this case, ±300 ppm or further from water should be chosen. At least 1 dummy scan or shot is required for the single-shot or multi-shot acquisition, respectively. Finally, an intrinsically/dynamically referenced ΔB₀-correction method (such as CEST-Dixon or CEST-GRE-2TE) ^{128, 129, 177} is a good option to correct for B₀ variations, but a separate ΔB₀ mapping approach is generally acceptable at 3 T. For the latter, it is, of course, important to run the ΔB₀ mapping and APTw scans under the same shimming and ideally in immediate succession, because of potential B₀ drift effects (Figure 8A). The approaches and parameters suggested in Table 4 allow for the greatest degree of flexibility in the application of APTw imaging to multi-center clinical trials for the assessment of brain tumors and novel therapies.

6 DATA PROCESSING

Different data processing approaches have been proposed to quantify the APT effect in vivo. Examples are the widely used MTR_asym analysis,¹⁷⁹ the 3-offset method,⁷⁵ and the extrapolated semi-solid magnetization transfer reference (EMR) method,^{95, 96} which were designed to, as much as possible, remove the DS and MTC background signals based on a reference signal. Other approaches include various model-based, Z-spectral fitting approaches, such as multi-pool Lorentzian fitting^180-182 or multi-pool Bloch-McConnell equation fitting.^{183, 184} These often allow a more specific CEST quantification and provide the ability to quantify multiple CEST parameters. The most suitable APT analysis method may depend on the setting of RF saturation amplitudes. At relatively low saturation amplitudes (<1 μT), APT and rNOE peaks are often distinguishable and relatively easy to fit. However, at 2 μT recommended for brain tumor imaging at 3 T, the distinct APT signal at 3.5 ppm is often invisible (because of a large saturation bandwidth on the order of ppm)⁹⁴ and the model-based fitting approaches, designed to separate out more pure APT effects, may not work well. Recently, a quasi-steady-state (QUASS) CEST analysis method was developed to account for the effects of finite saturation time and relaxation delay, which may facilitate more robust CEST quantification of individual resonances.^{185, 186} Here, we focus on the consensus on APTw imaging using MTR_asym analysis, whereas other quantification approaches will require a separate evaluation and consensus.

Notably, in addition to its simplicity and speed, MTR_asym-based APTw imaging at 2 μT has some other useful characteristics that are not directly related to amide proton exchange properties, such as the close-to-0 APTw signal in healthy tissue and the coarse T_1w independence, as discussed above. According to Table 2, when a B_1rms of 2 μT and a T_sat of 2 s are used on 3 T MRI scanners, very similar APTw intensities or contrasts can be observed from different vendors/sites (grade-4/3/2 glioma APTw contrast = 4.0%/2.2%/1.0% on GE; grade-4/3/2 glioma APTw signal = 4.1%/3.2%/2.1% on Philips; high-grade glioma APTw signal =3.5% on Philips),^{54,26, 164} ranging approximately from 1% to 4% in solid tumors of different grades. The MTR_asym(3.5 ppm) or APTw metric has, therefore, become the basis of the first commercial APTw imaging sequence on 3 T clinical MRI scanners.⁶⁰

Based on previous studies, we recommend MTR_asym (3.5 ppm, B_1rms = 2 μT, T_sat >0.8 s, DC_sat ≥50%) for brain tumor APTw imaging on 3 T clinical MRI scanners (Table 4). Using the recommended 6-offset acquisition protocol and estimate of B₀ offset, the equivalent value for the signal at ±3.5 ppm should be calculated using interpolation (linear or Lagrange interpolation being suitable with the small number of sample points) and MTR_asym(3.5 ppm) calculated from these values.^{26, 148, 151} Most APTw images have been displayed historically using a rainbow color scale. When the recommended B₁ and saturation time values are used, this often leads to a green background in normal brain regions with positive, yellow/orange/red hyperintensities in high-grade tumors (Figure 9) and negative, blue hypointensities in ischemic tissue.¹⁴² We, therefore, recommend that the results should be displayed on MRI scanner consoles in a window of ±5%, with a specific rainbow colorbar (no. 013), defined by Interactive Data Language (IDL; Harris Geospatial Solutions, Broomfield, CO),^{187, 188} to visually cover all possible APTw signal changes seen in different clinical applications, and that both rainbow color and gray-scale images are stored. In addition, to enable comparison of APTw images to previous published data, we suggest that the results should be displayed in publications using this colorbar and window. Nevertheless, we understand that radiologists may choose windows and levels themselves in reading rooms using different post-processing solutions and may use gray-scale or any color scales that they prefer for their specific clinical study with APTw imaging. Although radiologists usually prefer the rainbow scale, it is important to point out that it is increasingly recognized that this scale is not very suitable for color-blind people and that the sharp color transitions may be misleading.¹⁸⁹ We encourage experimentation with the use of perceptually uniform sequential colorbars for APTw imaging in the future. Because of a lack of published data with such scales, we can currently not make a consensus recommendation about this.

7 DATA INTERPRETATION

As is typical for all MRI approaches, there can be false-positive and false-negative findings because of low SNR. This certainly applies to CEST MRI, including APTw imaging, which detects changes in the range of a few percent of the water MRI signal. Consequently, the APTw MRI approach, based on difference images, is susceptible to motion and B₀ shifts that may cause artifacts in the APTw images. Although low SNR, together with motion, can average out small hypo- or hyperintensities, strong motion artifacts can also lead to false hyper- or hypo-intensities, as previously shown for dynamic CEST studies.¹⁹⁰ It has been reported on clinical scanners, with a saturation time of 2 s and a B₁ amplitude of 2 μT, that the repeatability of the APTw signal was excellent in supratentorial locations, but it was poor in infratentorial locations because of severe B₀ inhomogeneity and susceptibility that affect the APTw signal.^{163, 164} In addition, as discussed in “Acquisition Protocols,” the B₁ inhomogeneity can be up to ±30% in the infratentorial and superior regions of the brain, which may affect the APTw signal. Therefore, interpretation of APTw imaging must include possible SNR, B₀/B₁ inhomogeneity, and motion influences, or rule these out at the acquisition or post-processing stage, which may be possible with future developments.

Although APTw imaging is useful for brain tumor evaluation, there are pitfalls for APTw image interpretation. In addition to artifacts because of motion and B₀ alterations, areas of large liquefactive necrosis, hemorrhage, or large vessels typically demonstrate high APTw signals and should not be mistaken as viable tumor.¹¹ Figure 10 gives several representative images of liquefactive necrosis and hemorrhage at different stages. Careful interpretation is needed with post-operative-stage tumors with the surgical cavity filled with proteinaceous fluid, unless a fluid suppression method is used (Figure 10C and D).^{56, 111} To distinguish between viable tumor and proteinaceous fluid, APTw images should generally be interpreted together with anatomic MRI (such as T₂w, fluid-attenuated inversion recovery (FLAIR), and pre- and post-contrast T₁w), SWI, diffusion, and perfusion (including dynamic susceptibility contrast-enhanced and dynamic contrast-enhanced) MRI sequences that are acquired during routine clinical tumor protocols. This comparison helps, on one hand, to recognize and assign non-tumorous signals and potential artifacts on APTw images and, on the other hand, to identify tumor viability characteristics, which are not captured by the structural and perfusion-weighted MRI (Figure 10D). The information provided by APTw MRI should be regarded as complementary to existing approaches, further extending the repertoire of diagnostic tools in radiology. Finally, and importantly, the selection of the color scheme in the APTw images can affect the interpretability of the information contained, as discussed in “Data Processing.” Advanced post-processing solutions suited for APTw imaging can offer different color scales and windows, including perfusion-like color ranges, which may be more familiar to the radiologist and, therefore, more easily readable (Figure 10D).

As discussed in “Pulse Sequences,” a correction for T_1w changes is not necessary for APTw imaging of brain tumors at 2 μT on clinical 3 T scanners. T_1w values decrease after Gd injection, especially in Gd-enhancing regions that are of importance for tumor assessment. The Gd may remain there for a longer time, whereas it may clear faster from other tumor regions, leading to T₁ heterogeneity in the tissue. Shorter T_1w causes a reduced signal buildup in saturation transfer that leads to the region-dependence of quantification because of T_1w heterogeneity.¹⁰³ This could be corrected for if T_1w could be known; however, because of the change in T_1w over time with Gd concentration in the region, T_1w mapping after Gd is actually difficult. We, therefore, do not recommend APTw MRI after Gd, but if it is acquired, we suggest adding a note for this to help the interpretation.^{101, 102} Other metrics with relaxation compensation ability have been suggested,^{101, 102} and a test to compensate for Gd-induced T₁ changes on injection was performed in an animal tumor model.¹⁹¹ However, how these findings translate from animals to humans and from the sole amide CEST signal used to MTR_asym is not yet clear. It is, therefore, not part of this guidelines paper.

8 CONCLUSIONS AND REMARKS

This paper reviews and recommends the currently optimized APTw approaches at 3 T with an attempt to standardize this imaging technology in the clinical setting, and focus on APTw imaging applications to brain tumors (Tables 4 and 5). When the preferred pulse-train method (B_1rms = 2 μT, T_sat = 2 s) is applied to healthy volunteers, comparable Z-spectra and MTR_asym spectra can be obtained from the 3 T MRI scanners of the 3 main manufacturers. In addition, when using the recommended saturation sequences, currently published data from patients with brain tumors show that very similar APTw intensities or contrasts are observed for data from different vendors. We expect that these recommendations will become the first guidelines for APTw imaging of brain tumors on 3 T MRI systems from different vendors. When implemented, more medical centers will be able to use the same or comparable techniques in investigating the added clinical value of APTw MRI in larger independent patient cohorts and ultimately lead to biomarker status of this contrast for brain tumors. Reducing variability across vendors, sites, patients, and time will improve value and practicality of APTw imaging as a quantitative imaging biomarker that is consistent with the mission of the quantitative imaging biomarkers alliance (QIBA).¹⁹²

In addition to the brain tumor applications described above, APTw MRI has been applied to various neurological disorders and other diseases in the clinical setting. Notably, the extension of APTw MRI from the brain to other body regions is complicated by increased B₀ inhomogeneity, motion, and increased lipid contamination, which often lead to inferior imaging quality for many patients. Undoubtedly, APTw imaging methods must be optimized and standardized separately for each of these applications, and consensus recommendations are not possible at this time. In addition to 3 T, APTw imaging applications to patients with brain tumors have been studied at 7 T^{102, 193, 194} and 1.5 T,^{195, 196} with the latter being important clinically. Theoretically, the extension to lower magnetic field is straightforward; however, the increase in background interference (smaller frequency range in Hz) would affect the APTw MRI signal, and there is currently insufficient knowledge for a consensus. APTw imaging for other applications is still in its infancy. A continued effort to explore new APTw MRI pulse sequences, data acquisition protocols, and data processing methods is needed, particularly for neurological disorders other than brain tumors and various body diseases. We expect that significant advances will be achieved and updated new recommendations will be proposed within the next 5 years or so.

CONFLICT OF INTEREST

J.Z. and P.C.M.v.Z. are paid lecturers for Philips and are the inventors of technology (including APTw MRI) licensed to Philips. P.C.M.v.Z. also has research support from Philips Healthcare. These arrangements have been approved by Johns Hopkins University in accordance with its conflict-of-interest policies.