Endogenous Chemical Exchange Saturation Transfer MRI for the Diagnosis and Therapy Response Assessment of Brain Tumors: A Systematic Review

Abstract

Purpose

To generate a narrative synthesis of published data on the use of endogenous chemical exchange saturation transfer (CEST) MRI in brain tumors.

Materials and Methods

A systematic database search (PubMed, Ovid Embase, Cochrane Library) was used to collate eligible studies. Two researchers independently screened publications according to predefined exclusion and inclusion criteria, followed by comprehensive data extraction. All included studies were subjected to a bias risk assessment using the Quality Assessment of Diagnostic Accuracy Studies tool.

Results

The electronic database search identified 430 studies, of which 36 fulfilled the inclusion criteria. The final selection of included studies was categorized into five groups as follows: grading gliomas, 19 studies (area under the receiver operating characteristic curve [AUC], 0.500–1.000); predicting molecular subtypes of gliomas, five studies (AUC, 0.610–0.920); distinction of different brain tumor types, seven studies (AUC, 0.707–0.905); therapy response assessment, three studies (AUC not given); and differentiating recurrence from treatment-related changes, five studies (AUC, 0.880–0.980). A high bias risk was observed in a substantial proportion of studies.

Conclusion

Endogenous CEST MRI offers valuable, potentially unique information in brain tumors, but its diagnostic accuracy remains incompletely known. Further research is required to assess the method’s role in support of molecular genetic diagnosis, to investigate its use in the posttreatment phase, and to compare techniques with a view to standardization.

Keywords: Brain/Brain Stem, MR-Imaging, Neuro-Oncology

Supplemental material is available for this article.

© RSNA, 2020

Summary

Chemical exchange saturation transfer can be further developed as a biomarker for metabolically active brain tumors, evidenced by correlations to tissue findings, including proliferative indexes; further study is required to assess its diagnostic power with respect to specific clinical indications.

Key Points

• ■ Endogenous chemical exchange saturation transfer (CEST) methods can support glioma grading, molecular subtyping, and differential diagnosis.

• ■ CEST signal may aid the identification of metabolically active tumor following treatment.

• ■ Study data are heterogeneous with a substantial bias risk, highlighting the importance of future prospective research and technical standardization.

Introduction

Gliomas account for most malignant intrinsic brain tumors in adults and, despite being a relatively rare disease, represent a major cause of mortality (1). Diffuse gliomas are categorized into World Health Organization (WHO) grades II to IV, based on histologic evidence of proliferation and vascular invasion. However, histologic (WHO) grade and glioma cell lineage (oligodendroglioma vs astrocytoma) are limited predictors of disease progression, which is predominantly influenced by genetic factors (2). Recent studies have identified molecular markers, such as the isocitrate dehydrogenase (IDH) gene and methylguanyl methyltransferase (MGMT) enzyme, as key determinants of clinical outcomes (1). The optimal treatment and overall prognosis of glioma subtypes depend on the combination of molecular features and histologic grade (1); however, tumor malignant potential remains incompletely captured by clinical imaging techniques (3). In addition, MRI features can overlap between gliomas and different brain tumors (eg, lymphoma, metastases) to such an extent that only tissue diagnosis is conclusive (3). Postoperative radiation and chemotherapy with temozolomide may result in predominantly transient (pseudoprogression) or permanent (radiation necrosis) phenomena, which notoriously resemble contrast-enhancing tumor progression due to breakdown of the blood–brain barrier. Definitive distinction of these entities frequently requires serial imaging with a combination of structural and advanced techniques (4).

Chemical exchange saturation transfer (CEST) is a promising imaging technique that has recently emerged as an alternative contrast mechanism for MRI (5). CEST signal can be generated through application of a radiofrequency “saturation” pulse targeted at the resonance frequency of solute (eg, protein or metabolite bound) protons, from which the saturation is transferred to bulk water via chemical exchange. The much larger water proton pool ensures a continuous flux of unsaturated protons close to the exchangeable sites, thereby leading to a measurable reduction in the water signal amplitude after a few seconds (6). CEST contrasts are classified into diamagnetic CEST, mostly consisting of endogenous agents, and paramagnetic CEST, which usually involves the use of exogenous agent administration (6). Diamagnetic CEST uses chemical compounds with a range between 0 and 7 ppm from water (eg, -NH,-NH₂,-OH groups), representing the first discovered and most studied CEST contrast (7). CEST techniques can be classified according to the type of molecular construct, such as amide proton transfer (APT), amine CEST, glucoCEST (glucose-based CEST contrast), and gagCEST (CEST contrast originating from glycosaminoglycans) (6). APT imaging targets endogenous mobile proteins and peptides featuring amide protons and is the most widely used CEST imaging method, whereby the APT-weighted signal can be quantified by magnetization transfer ratio (MTR) asymmetry (MTR_asym) analysis at +3.5 ppm, using the water peak as reference (5). In addition, nuclear Overhauser enhancement (NOE)–mediated signal arises from mobile protein and lipid spin cross-relaxation effects between 0 and −5 ppm (8). It has been proposed that NOE could also become an imaging biomarker to characterize brain tumors, similar to APT (9).

Numerous single-center studies have highlighted the potential of CEST MRI in stratifying brain tumors; however, the exact diagnostic contribution of the method remains uncertain. To date, a single systematic review and meta-analysis have evaluated the diagnostic performance of only APT in grading gliomas (10). To our knowledge, ours is the first systematic review to explore the diagnostic and prognostic value of endogenous CEST for a variety of brain tumor indications. Our analysis aims to evaluate (a) the diagnostic value for grading gliomas, (b) the accuracy for predicting glioma molecular subtypes, (c) the distinction of glioma from other brain tumor types, (d) the assessment of brain tumor therapy response, and (e) the power of differentiating tumor recurrence from treatment-related changes.

Materials and Methods

This study was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis of Diagnostic Test Accuracy Studies criteria (11). The research was registered in the PROSPERO online database of systematic reviews (CRD42019122320).

Search Strategy

In November 2018, a medical researcher performed a systematic search in PubMed, Ovid Embase, and the Cochrane Library. We used the following search key words: (“brain tumor,” “glioma,” “brain neoplasm,” “brain metastasis,” “glioblastoma”) and (“CEST,” “chemical exchange saturation transfer,” “amide proton transfer,” “magnetization transfer,” “chemical exchange,” “nuclear Overhauser effect”). Further details of the search strategy are shown in Appendix E1 (supplement).

Selection Criteria

The abstracts of all articles retrieved in the initial search were screened by two board-certified radiologists (S.O. and A.H.) with research experience in neuro-oncology. Selected full-text manuscripts were reviewed in detail to determine their relevance. A stepwise selection was performed by two independent reviewers (S.O. and A.H.) according to the following criteria: The exclusion criteria were as follows: (a) no CEST technique (eg, CEST, APT, NOE) performed; (b) no patients with brain tumor examined; (c) animal and/or laboratory study; (d) technical study or diagnostic and/or prognostic value in brain tumors not evaluated; (e) comparisons confined to different MRI acquisition technique; (f) review article, case report (defined as fewer than five cases), letter, commentary, or conference proceeding; and (g) non-English full text. The inclusion criteria were (a) CEST technique was performed in patients with brain tumor before, during, or after treatment and (b) study assessed diagnostic or prognostic value of CEST parameters in brain tumors or examined pseudoprogression or recurrent tumors. Disagreement was resolved in consensus with a senior reviewer (S.T.).

Data Extraction

The following data were extracted from the included studies: CEST parameter values, diagnostic or prognostic accuracy, and method characteristics. The latter included study design, country of origin, number of patients, participant age, tumor histologic features, and, where available, molecular data, MRI field strength, type of CEST contrast, CEST acquisition parameters, methods of correcting B0 field inhomogeneity, and region-of-interest placements. The same two reviewers independently performed the full-text screening, followed by data extraction, and any discrepancies were resolved in consensus with the third reviewer.

Study Quality Assessment

The study quality was examined by using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) instrument (12). We evaluated concerns regarding applicability in three domains and the risk of bias in four different domains. Each study was independently assessed for quality and potential bias by the same two researchers. Disagreements were resolved as described above.

Statistical Analysis

Descriptive data are presented in the form of a narrative synthesis because of the perceived heterogeneity of research questions, CEST technical parameters, and brain tumor cohorts studied.

Results

Search Results

A total of 430 studies were identified through the electronic database searches. After we removed duplicate studies and screened the studies’ titles and abstracts, 68 studies that provisionally satisfied the inclusion criteria remained. Of these, 36 studies were relevant in subsequent full-text screening. We categorized the final selection of 36 studies into five groups as follows: grading gliomas, 19 studies (9,13–30); predicting molecular subtypes of gliomas, five studies (13,14,31–33); distinction of different brain tumor types, seven studies (5,8,34–38); therapy response assessment, three studies (39–41); and differentiating recurrence from treatment-related changes, five studies (25,42–45). Two studies (13,14) contained data on glioma grading and predicting molecular subtypes, and one study (25) was assigned to both glioma grading and differentiating recurrence from treatment-related changes. A flowchart of the study selection process is presented in Figure 1. All studies included in the analysis are summarized in Tables E1–E6 (see supplement).

Figure 1:

Flowchart describes the study selection process. Two studies contained data on glioma grading and predicting molecular subtypes, and one study was assigned to both glioma grading and differentiating recurrence from treatment-related changes.

CEST Techniques

Thirty-three of the 36 studies used APT-weighted imaging. Six studies presented NOE-weighted images, and four studies assessed amine CEST. Three studies tested conventional magnetization transfer imaging, which depicts semisolid macromolecules in the more solid environment of the cell than APT (37), and one study used fitted magnetization transfer and NOE.

Glioma Grading

A total of 596 patients with glioma (one with WHO grade I, 232 with WHO grade II, 129 with WHO grade III, 193 with WHO grade V, and 41 with WHO grades III or IV) were included from 19 studies. Studies summarized WHO grades I and II into low-grade gliomas (LGGs), whereby WHO grade I corresponds to indolent entities other than diffuse glioma (eg, pilocytic astrocytoma [2]) and WHO grades III and IV into high-grade glioma (HGGs). Seventeen of 19 studies for glioma grading used light microscopic analysis according to the WHO 2007 Classification of CNS Tumors; only two of 19 (more recent) studies adopted the WHO 2016 Classification of CNS Tumors as the diagnostic reference standard. Of these, both studies performed immunohistochemistry testing for IDH1 status, and one study performed analysis for MGMT genetic status. In five of 19 studies, investigators reported the Ki-67 labeling index as a biomarker of tumor cellularity. Seventeen studies used APT-weighted imaging, two studies evaluated amine CEST, two studies presented NOE-weighted images, and one study used fitted magnetization transfer and NOE. Seventeen studies used 3-T MRI and two studies used 7-T MRI. The imaging parameters and grading results are shown in Tables E1 and E2 (see supplement).

Statistically significant differences of APT signals between HGGs and LGGs (with greater and lower signal, respectively) were identified in 16 of 17 studies using APT-weighted images (P < .0001–0.0497); the other study by Heo et al (9) found no difference. Furthermore, significant differences were demonstrated between WHO grades II, III, and IV in studies by Bai et al (23) and Togao et al (28). A significant difference between WHO grades II and III but no difference between WHO grades III and IV was reported in the studies by Zou et al (15) and Jiang et al (21). In contrast, in the studies published by Choi et al (22) and Sakata et al (27), no difference was shown between WHO grade II and III, but WHO III differed significantly from WHO grade IV. Receiver operating characteristic (ROC) curve analyses were carried out in 13 of 17 studies. These demonstrated low to high diagnostic performance with areas under the ROC curve (AUC) of 0.500–1.000.

Paech et al (13) and Heo et al (9) evaluated NOE-weighted MR images by using 7-T imaging. Paech et al (13) showed a lower diagnostic performance for NOE-weighted images than APT-weighted images and downfield-relayed NOE-suppressed APT. Conversely, Heo et al (9) reported that NOE-based signals of HGGs were significantly lower than those of LGGs (P < .05), with no statistically significant difference in APT-based signals.

Harris et al (14,26) performed two studies to evaluate diagnostic performance of pH-weighted amine CEST for gliomas. The initial research, in 2016 (26), yielded a statistically significant amine CEST signal difference for glioma of WHO grades II, III, and IV (P < .05 for WHO grade III vs grade IV and WHO grade II vs grade IV), but the subsequent study, in 2018 (14), identified a difference only for WHO grade II versus WHO grade IV (P < .05). CEST signals increased with increasing tumor grades in both studies.

Some study authors proposed a combination of CEST and multimodal techniques to increase the diagnostic accuracy. Zou et al (15) reported that the combined use of intravoxel incoherent motion resulted in the increase of AUC from 0.957 to 0.986. Sakata et al (17) observed that the combined use of fluorodeoxyglucose (FDG) PET improved the AUC from 0.76 to 0.85, and Choi et al (22) found that the addition of relative cerebral blood volume derived from dynamic susceptibility contrast material–enhanced MRI increased the AUC from 0.877 to 0.923. The correlation of APT signals and MR spectroscopic parameters (choline, choline-to-N-acetylaspartate ratio, N-acetylaspartate, choline-to-creatine ratio, N-acetylaspartate-to-creatine ratio) were investigated in three studies with moderate correlations (r = 0.4–0.6).

Predicting Molecular Subtypes of Gliomas

A total of 165 patients with glioma (60 with IDH wild type, 44 with IDH mutant, 23 with MGMT methylated, 17 with MGMT unmethylated, 38 with positive MGMT immunostaining, four with negative MGMT immunostaining) were included from five studies. Three of the five studies performed immunohistochemistry testing for IDH1 status, two of five did so for MGMT promotor methylation status, and one of five did so for MGMT protein expression. The MGMT methylation status was assessed with a methylation-specific polymerase chain reaction, and MGMT protein expression in tumor cells was reviewed with light microscopy. Four studies used APT-weighted imaging, one study performed amine CEST, one study used NOE-weighted imaging, and one study tested conventional magnetization transfer imaging. Four studies were undertaken by using 3-T and one study using 7-T magnetic field strength. Details of MRI parameters and molecular subtyping results are shown in Tables E1 and E3 (see supplement).

Jiang et al (33) and Paech et al (13) investigated the value of CEST to predict IDH mutation status. Jiang et al reported an AUC of 0.89 with use of a maximum region-of-interest value (“hot spot”) analysis of APT imaging in WHO grade II gliomas (n = 27), with greater APT signal identified in IDH wild-type gliomas. Paech et al proposed that downfield-relayed NOE-suppressed APT had a high diagnostic performance (AUC, 0.92–0.98) for IDH typing in a mixture of gliomas with various WHO grades (II–IV, n = 31) with increased APT signal in IDH wild-type gliomas. Harris et al (14) evaluated IDH status using pH-sensitive and oxygen-sensitive amine CEST, reporting marginally greater signal in IDH mutant (P = .0434).

Studies by Su et al (31), Jiang et al (2018) (32), and Paech et al (13) evaluated APT for the prediction of MGMT methylation status. Su et al reported a moderate diagnostic accuracy (AUC, 0.849) for a visual scale (qualitative) assessment of APT characteristics. Tumors with greater signal intensity on the solid part or peripheral abnormality tended to be MGMT-positive gliomas. Jiang et al observed a moderate performance (AUC, 0.856) using histogram analysis of MTR_asym at 3.5 ppm in a comparison of the MGMT-unmethylated glioblastomas (GBMs) versus the MGMT-methylated GBMs. APT signals were significantly higher in the unmethylated GBMs than in the methylated GBMs (mean APT, P = .022; 90th percentile APT, P = .006). Paech et al presented APT and NOE results, which achieved low diagnostic accuracy (AUC, 0.61–0.69) although slightly greater compared with perfusion (relative cerebral blood volume AUC 0.59) and diffusion-weighted MRI (apparent diffusion coefficient AUC 0.59). APT and NOE between the unmethylated gliomas and the methylated gliomas had no statistically differences (P = .13–.39).

Distinction of Different Brain Tumor Types

A total of 215 patients (124 gliomas [four WHO grade I, 20 WHO grade II, 17 WHO grade III, 77 WHO grade IV, six unclear], 59 metastases, 11 primary central nervous system lymphomas (PCNSLs), eight meningiomas, two pituitary adenomas, three hemangioblastomas, one angiosarcoma, six cavernous malformations, and one angiosarcoma) were included from seven studies. Six brain metastases and nontumor lesions were confirmed by clinical diagnosis, and the remaining tumors were confirmed by histopathologic assessment. The MRI parameters and CEST characteristics are shown in Tables E1 and E4 (see supplement).

Yu et al (34) proposed that APT may help differentiate solitary brain metastases from GBM. In their study of 45 patients with solitary brain metastases versus 43 patients with GBM, APT values in perilesional tissue were significantly lower in the solitary brain metastases group, in which the minimum APT-weighted values produced the highest AUC (0.905) compared with mean APT-weighted values (AUC, 0.868) for lesion discrimination.

Jiang et al (37) reported a high accuracy (AUC, 0.963) for a subtraction parameter (APT weighted_max−min) to differentiate 11 PCNSLs from 21 HGGs, whereby the PCNSLs had significantly lower APT weighted_max−min (mean, 0.76% ± 0.42 [standard deviation]) than the HGGs (2.55% ± 1.20). Jeong et al (36) compared APT signals in hemorrhagic brain lesions of 16 tumors and seven lesions with nonneoplastic causes, observing that MTR_asym in acute to subacute hemorrhage was greater than in surrounding brain, regardless of the underlying pathologic condition.

Park et al (38) analyzed 45 gadolinium-enhanced tumors, consisting of 19 “low-grade” tumors (four pilocytic astrocytomas, two hemangioblastomas, three low-grade astrocytomas, seven low-grade oligodendrogliomas, three pleomorphic xanthoastrocytomas) and 26 “high-grade” tumors (five anaplastic astrocytomas, three anaplastic oligodendrogliomas, two anaplastic oligoastrocytomas, 11 GBMs, five brain metastases), reporting that APT 90th percentile had an AUC of 0.85–0.86 in discriminating low-grade tumors and high-grade tumors. Compared with normalized 90th percentile cerebral blood volume alone, adding APT 90th percentile significantly improved the AUC for the identification of contrast-enhanced low-grade tumor from 0.80–0.82 to 0.97.

Of three studies (5,8,35) featuring gliomas and meningiomas, Jones et al (5) were the first group to demonstrate that the APT effect is quantifiable (eight gliomas and two meningiomas). Shen et al (8) used NOE maps, observing a significantly lower signal within tumor than contralateral normal-appearing white matter for six gliomas (P < .001) versus no significant difference for five meningiomas (P = .116). Khlebnikov et al (35) used the effect of water T1 relaxation on APT to compare three different metrics of APT contrast: MTR, relaxation-compensated magnetization transfer ratio, and traditional asymmetry (MTR_asym) in five gliomas and one meningioma. This study identified a difference that appeared between LGG and HGG in non–gadolinium-enhanced solid tumor regions using MRT and no difference in relaxation-compensated MTR.

Differentiating Tumor Recurrence from Treatment-related Changes

A total of 161 patients with glioma (15 WHO grade II, 15 WHO grade III, 131 WHO grade IV; 108 with tumor progression, 53 with treatment-related effects) and 16 patients with brain metastasis (five with tumor progression, 11 with radiation necrosis) were included from five studies. Final diagnoses were confirmed by second-look surgery or clinical-radiologic follow-up by using the Response Assessment in Neuro-Oncology criteria. All studies used APT-weighted imaging, and one study also assessed magnetization transfer and NOE signals. All studies were completed with 3-T MRI. The patient characteristics and study results are listed in Tables E1 and E5 (see supplement).

One study (43) found a significant difference between tumor progression and radiation necrosis for brain metastases. An ROC analysis was not performed, but NOE MTR and amide MTR differed between tumor progression and radiation necrosis (P < .0001). The remaining four studies (25,42,44,45) enrolled patients with glioma (15 with WHO grade II, 15 with WHO grade III, 131 with WHO grade IV). In all four studies, APT signals were significantly higher in tumor progression than in therapy-induced lesion changes; diagnostic accuracies were high (AUC, 0.88–0.98). In a 2018 study, Park et al (42) compared APT and PET imaging and reported greater diagnostic accuracy for APT than carbon 11 (¹¹C) methionine PET. Previously, Park et al (44) had combined gadolinium enhancement features and normalized cerebral blood volume with APT, resulting in increased diagnostic accuracy (AUC, 0.97) over APT alone (AUC, 0.89) for the distinction of glioma recurrence from therapy effects.

Therapy Response Assessment and Prognosis Prediction

Three studies examined therapy response assessment and prognosis prediction by using CEST MRI. Of note, each study differs in its research purposes and investigated different types of brain tumors. The patient characteristics and study results are presented in Tables E1 and E6 (supplement). Regnery et al (39) examined NOE and APT signals at 7-T MRI in 20 patients with GBM to predict early tumor progression after first-line treatment. Pretreatment tumor signal in NOE-Lorentzian difference differed significantly according to responsiveness to first-line treatment (AUC, 0.98).

Desmond et al (40) evaluated the predictive value of various CEST metrics in 25 brain metastases treated with stereotactic radiosurgery at baseline compared with 1 week after treatment and related these to changes in tumor volume at 1 month. A significant association was observed between metastasis volume changes and the relative change in NOE peak amplitude in contralateral normal-appearing white matter.

Harris et al (41) performed pH-weighted imaging in 20 patients with GBM and evaluated differences between acidic tumors and nonacidic tumors in progression-free survival. The median progression-free survival intervals for acidic tumors and nonacidic tumors were 125 days and 450 days, respectively.

Study Quality

The results of the study quality assessment, performed by using the QUADAS-2 tool, are demonstrated in Figure 2. Several studies had a high risk of bias regarding the selection of patients (17 of 36) and/or concerning the conduct or interpretation of the index test (six of 36) due to retrospective design and/or region-of-interest placement by a single researcher. In a high proportion of studies (approximately 80%), it was unclear whether radiologists were blinded to histologic results when placing regions of interest, and in approximately 50% it was unknown whether the interval between imaging and tissue diagnosis was appropriate (ie, when imaging signals were compared to subsequently diagnosed histologic glioma grades).

Figure 2:

Results of the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) quality assessment of the included studies. The risk of bias in four different domains and concerns regarding applicability in three domains are shown.

Discussion

Glioma Grading

Through this systematic review, we identified 36 research studies on the value of endogenous CEST techniques to depict brain tumor metabolism. Approximately half of this research was aimed at predicting glioma histologic (WHO) grades. Broadly, these grading studies indicate a link between greater cellularity in HGGs, higher concentration of proteins and peptides, and APT signal intensity (15,18). Most grading research found higher APT image signals in HGGs than in LGGs, with variable diagnostic accuracy for individual WHO grade distinction. According to the ROC curve analyses, which produced moderate to high AUC values in many studies (13 of 19), the evidence for the use of CEST in glioma grading is judged to be moderate, whereas the diagnostic accuracy differs among glioma grading studies. For example, Zou et al (15) and Jiang et al (21) reported AUC values of 0.957 and 1.000, respectively, whereas Zhang et al (16) and Sakata et al (27) achieved AUC values of 0.723 and 0.760, respectively, for differentiating between HGGs and LGGs using APT. Aside from technical differences and sampling limitations, the heterogeneity in these data sets is likely to be influenced by the lack of glioma grouping according to molecular genetics. A fundamental change has occurred in the reference standard of the WHO classification of central nervous system tumors from the previous 2007 version (histologic grading only) to the 2016 classification (integrated diagnosis considering histologic grading and molecular markers), whereby most CEST studies carried out for glioma grading (17 of 19) took into account histologic findings only. Specifically, LGGs indistinguishable by histologic criteria may differ in malignant potential (eg, according to IDH status), which may affect the CEST signal through difference in the number of solutes (related to the proteasome content) and the pH, depending on the presence or lack of an IDH mutation (2,33). Whereas numeric thresholds from individual studies lacking molecular data should be interpreted with caution, in its entirety the research on glioma grading underscores the potential of CEST to quantify malignant metabolism. This is further supported by the statistical associations between APT metrics and Ki-67 in two prospective studies (16,21).

CEST signals contain complex information from various technical factors, whose contributions strongly depend on the experimental setup, such as power, length, and shape of the radiofrequency saturation pulses (24,26), all of which may affect results. A recent meta-analysis by Suh et al (10) focused on the use of APT for glioma grading and attributed variations in radiofrequency saturation power as a probable factor on the heterogeneity of study results.

NOE signals, which are hypothesized to originate from magnetization transfer between water protons and proteins or lipids mediated through intramolecular NOE effects (9), have been identified as valuable to support glioma characterization. However, the extent to which NOE plays a role remains uncertain; Paech et al (13) observed no significant differences for glioma WHO grades, whereas Heo et al (9) reported WHO grade differences for a study of only 10 patients (molecular data unknown). In the study by Paech et al, downfield-relayed NOE-suppressed APT had higher diagnostic performance than conventional APT at 7-T MRI, indicating that NOE contributes to CEST image signal, probably as a confounding effect. Of note, NOE effects are thought to be substantial at 7-T but smaller at 3-T clinical field strength (46).

The comparison of APT CEST with techniques such as diffusion-weighted imaging, FDG PET, and MR spectroscopy for glioma characterization could be of interest for a multimodal diagnostic approach. APT was reported to provide greater diagnostic accuracy for grading than other techniques, and in several studies (13,15,17,22) the combination of CEST with other sequences (intravoxel incoherent motion, FDG PET, and dynamic susceptibility contrast-enhanced MRI) increased diagnostic performance. Therefore, the use of APT together with other modalities has been proposed to aid in grading gliomas. For the combination with APT, it has been reported that intravoxel incoherent motion resulted in an increase in AUC from 0.957 to 0.986 (15), that FDG PET improved the AUC from 0.76 to 0.85 (17), and that dynamic susceptibility contrast-enhanced MRI produced an AUC increase from 0.877 to 0.923 (22). However, the diagnostic accuracy of the combined use of APT and MR spectroscopy has not been comprehensively investigated.

Predicting Molecular Subtypes of Gliomas

Research into the ability of CEST to predict glioma molecular subtypes remains confined to a small number of studies on IDH and MGMT typing (32,33). IDH-mutant gliomas predominantly consist of WHO II–III gliomas and rarely (<10%) of secondary GBM, with an overall better clinical prognosis (1). Distinct from this are IDH wild-type gliomas, many of which correspond to the genetic equivalent of primary GBM with a similarly dismal prognosis, regardless of WHO grade (1). Key disturbances of cellular metabolism, including alterations of amino acid concentrations and reduction of protein expression, are caused by mutations in IDH gene-encoded enzymes (33). In addition, IDH mutations result in accumulation of the oncometabolite 2-hydroxygluterate, which inhibits oxidative phosphorylation and promotes aerobic glycolysis (14). However, lactic acidosis due to anaerobic glycolysis in the context of nutrient depletion and growing tumor hypoxia is a key property of IDH wild-type gliomas, which could confound a pH-based distinction (47). The reported diagnostic accuracy for IDH typing by Jiang et al (21) at 3 T (AUC, 0.89) and Paech et al (13) at 7 T (AUC, 0.98, including downfield-relayed NOE suppression) is very high. These results are promising, with the caveat that no information on blinding to immunohistochemistry is stated for either. Larger studies, including multicenter research on CEST imaging for glioma characterization, would be desirable, for example to investigate LGGs, which carry other mutational risk factors for malignant progression (48).

MGMT is a DNA repair enzyme, the activity of which determines glioma susceptibility to alkylating chemotherapy (temozolomide), whereby the methylated MGMT promoter status increases chemosensitivity. Both immunohistochemical MGMT protein expression and MGMT promoter methylation status are prognostic markers of survival in patients with gliomas (31,32). With regard to AUC, the results of Su et al (31) correlating APT signals with MGMT protein expression are similar to those of Jiang et al (32) assessing MGMT promoter methylation status, but differences in the glioma cohorts and analysis methods limit direct comparability. It has been proposed that MGMT promoter methylation in gliomas produces a decrease of protein expression, which may affect other protein activity downstream of MGMT (31). Therefore, CEST could be considered as a biomarker for predicting MGMT methylation status, but whether it has accuracy sufficient to affect clinical decisions is yet unclear (13).

Paech et al (13) compared CEST with diffusion-weighted imaging and dynamic susceptibility contrast-enhanced MRI for predicting IDH and MGMT status and found the diagnostic performance of CEST was marginally better than that of the other modalities.

The number of studies aimed at predicting glioma molecular subtypes is still limited, and the evidence for CEST in this context, although promising, is uncertain. Further research is desirable to confirm the method’s role in predicting specific genetic signatures and/or tumor biologic behavior.

Diagnosing Different Types of Brain Tumors

The study reporting the highest diagnostic accuracy (37) for differentiating PCNSL from GBM (AUC, 0.963) used a parameter not studied in other research, derived from a calculation (APT_max-min) as opposed to one measurement. However, the result is noteworthy, possibly reflecting greater APT signal heterogeneity in GBM, which is known to contain areas of rapid proliferation mixed with (metabolically inactive) necrosis. Also of interest is the finding of greater APT signal in GBM perilesional tissue compared with metastases (34) because it raises the possibility that CEST could improve the delineation of MRI-occult GBM infiltration.

Park et al (38) reported that adding APT to dynamic susceptibility contrast-enhanced MRI increased the diagnostic accuracy in characterizing brain tumors. This finding suggests that a multiparametric approach could be valuable for differentiating malignant gliomas, PCNSL, and brain metastatic disease.

The CEST data on the distinction of different types of brain tumors are limited by small patient numbers (5,8,35), different purposes (34,36–38), and quantitative metrics presented, so that the evidence supporting CEST for this clinical indication remains uncertain.

Differentiating Recurrence from Treatment-related Changes

Conventional MRI sequences are unreliable for differentiating treatment-related changes from tumor recurrence (44) and, even with use of advanced techniques, the distinction can be challenging. Thus, there remains an unmet clinical need for a serial imaging method to provide information on tumor viability. The high reported accuracy in several studies (AUC, 0.88–0.98) suggests that APT may dramatically improve the diagnostic value of MRI for this clinical question. In fact, the performance of APT for differentiating recurrence from treatment-related changes appears to be higher than for differentiating LGGs and HGGs. Recurrent tumors include more protein species, whereas there are fewer proteins in regions of treatment-related changes due to reduced cell density and cytoplasm disruption (49). These metabolic conditions could explain differences in APT signals between recurrence and treatment-related changes. Both APT and methionine PET aim to depict endogenous protein metabolism. Park et al (42) observed a higher diagnostic accuracy for APT than for ¹¹C methionine PET, which could be influenced by differences in protein metabolism. APT signal depends on mobile protein concentration, whereas methionine PET signal originates from actively synthesized proteins. In addition, methionine accumulation may contribute to disruption of the blood–brain barrier in HGGs (42). As in many studies on the distinction of brain tumor recurrence from therapy effects, the reference standard in this study included both cases where the final diagnosis was secured via second-look operation and imaging-only follow-up (using the Response Assessment in Neuro-Oncology criteria).

The evidence for the use of CEST in differentiating recurrence from treatment-related changes is judged to be weak, with study numbers as the main limitation. Those studies consistently report positive results, and more evidence is required for evaluating the efficacy of CEST in differentiating recurrence from treatment-related changes.

Therapy Response Assessment and Prognosis Prediction

In the posttherapy phase, APT may be able to depict baseline and dynamic changes in lesion acidity as a biomarker signature of viable GBM, as suggested by Harris et al (41). This evidence originates from a single-center study and requires validation, particularly as certain metabolic features of therapy changes and disease recurrence are known to overlap (50).

In a study following stereotactic radiosurgery, Desmond et al (40) identified dynamic changes in normal-appearing white matter, which correlated with volume changes in recently treated brain metastases. As such, CEST signal measurement in normal-appearing tissue may be of interest in monitoring disease progression and disease response. Given these few studies evaluating the relationships between CEST and therapy response or prognosis, the evidence in support of this indication is uncertain.

In summary, CEST techniques can provide information on brain tumor pathologic metabolism and tissue viability in humans at clinical magnetic field strength. But many complexities are unresolved. In particular, the current evidence is shaped by a majority of studies, which solely examined image signals in relation to glioma histologic grade. This limits the clinical impact of these data in the context of WHO 2016 integrated brain tumor diagnosis. The heterogeneity of brain tumor cohorts, acquisition, and interpretative approaches is problematic, including a high risk of bias for a substantial proportion of the published data. From the QUADAS-2 analysis, there was no relationship identifiable between the severity of bias risk and diagnostic accuracy.

Conclusion

Endogenous CEST imaging offers valuable, potentially unique information on brain tumors, but its diagnostic accuracy is incompletely known. Further research is required to assess the method’s role in support of molecular genetic diagnosis, to investigate its use in the posttreatment phase, and to compare methods with a view to technical standardization.