Semi‐solid MT and APTw CEST‐MRI predict clinical outcome of patients with glioma early after radiotherapy

The purpose of this study was to compare the potential of asymmetry‐based (APTwasym), Lorentzian‐fit‐based (PeakAreaAPT and MTconst), and relaxation‐compensated (MTRRexAPT and MTRRexMT) CEST contrasts of the amide proton transfer (APT) and semi‐solid magnetization transfer (ssMT) for early response assessment and prediction of progression‐free survival (PFS) in patients with glioma.


INTRODUCTION
Diffuse gliomas have an incidence of 5.2 / 100000 per year and are the second most common class of primary brain tumors over all. 1 Therapy involves maximal surgical resection followed by radiotherapy or concurrent radio/chemotherapy and adjuvant chemotherapy, depending on World Health Organization (WHO) grade. 2 Under current clinical standard of care, life expectancy depends on histopathologic subtype, tumor size, completeness of resection and patient age. [3][4][5][6] Time to treatment failure amounts to 1.4 y for glioblastoma (GBM), 4.5 y for astrocytomas with oncogenic mutations of the isocitrate dehydrogenase isotype 1 or 2 (IDH-mutant) and 9.8 y for IDH-mutant oligodendrogliomas. 6 Treatment-related brain damage and tumor progression can equally result in blood-brain-barrier damage and increasing gliosis, making differentiation on MRI challenging. 7 Misdiagnosis of either can result in severe consequences for the patient, as they require different therapeutic strategies. Growing awareness to pseudoprogression (PP), occurring in up to 30 % of all cases following radiation, has led to the introduction of response assessment in neuro-oncology (RANO) criteria in 2010. 8,9 However, RANO requires longitudinal follow-up as well as integration of image-based and clinical parameters, making correct response assessment challenging. For this reason, innovative MRI modalities such as diffusion-and perfusion-weighted imaging, PET-MRI and MRI spectroscopy have been evaluated for their clinical value in patients with glioma before and after therapy over the past two decades. This has led to the incorporation of some of these techniques into clinical routine. 10,11 CEST MRI is an innovative and contrast agent-free MRI approach, for which several groups have demonstrated great potential for therapy response assessment and prognostication in patients with glioma undergoing radio-and chemotherapy. [12][13][14][15][16][17][18][19] However, CEST contrasts are heavily dependent on the metrics used for their extraction from the Z-spectrum and the applied magnetic field strengths. 20,21 There is a fast growing body of evidence in the literature about the prognostic capacities of asymmetry-based, relaxation-compensated and Lorentzian-fit-based CEST contrasts of the amide proton transfer (APT) and semi-solid magnetization transfer (ssMT), that is for predicting therapy response at different field strengths of 1.5, 3, and 7 T. [16][17][18][19][22][23][24] Furthermore, several studies have demonstrated the diagnostic capacities of asymmetry-based APT-weighted (APTw asym ) CEST imaging for differentiation of treatment-related changes and glioma progression, as early as 3 mo after the completion of radiotherapy. [12][13][14] However, results from cross-sectional prospective clinical studies, comparing the prognostic and diagnostic potential of the APTw asym , as well as relaxation-compensated (MTR Rex APT and MTR rex MT) and Lorentzian-fit-based (PeakAreaAPT and MT const ) CEST contrasts of the APT and ssMT early after the completion of radiotherapy in a larger cohort of patients with glioma have not been reported before. Therefore, the purpose of this study was to compare the potential of the APTw asym , MTR Rex APT, MTR Rex MT, PeakAreaAPT, and MT const for response assessment and prediction of progression-free survival (PFS) in patients with glioma in a prospective clinical trial at the first follow-up 4 -6 wk after the completion of radiotherapy.

Study design
The local institutional review board committee approved this prospective study. Written informed consent was obtained from each participant prior to study inclusion. From September 2018 until December 2021 72 study participants (61 with initial disease and 11 with relapsing/progressive disease) being treated for diffuse glioma at the department of radiation-oncology of the University Hospital Heidelberg were prospectively enrolled. MRI scans were acquired at the first follow-up 4 -6 wk after the completion of radiotherapy. Inclusion criteria were minimum age of 18 y, Karnofsky-Performance-Score of 50 or higher and legal capacity to consent to study inclusion. Eleven participants had been excluded from analysis due to heavy motion artifacts (two participants), incomplete data sets (seven participants), large perioperative infarction (one participant), and missing clinical follow-up data at data cut off on January 26, 2022 (one participant) (Figure 1).

F I G U R E 1
Flow chart of the study. Seventy-two participants were prospectively included in the study and underwent CEST MRI 4 -6 wk following completion of radiotherapy. Eleven participants had to be excluded from evaluation due to the given reasons. Association of CEST contrast values with therapy response assessment at the first follow-up was tested by Mann Whitney-U-test and ROC analysis in all 61 evaluable study participants (all participants, n = 61) and 42 participants with residual contrast enhancement on MRI (participants with CE, n = 42). Gliomas with midline locations were excluded for PFS assessment, since these tumors are known to be associated with an especially poor prognosis (participants with CE, n = 37). CE, contrast enhancement; KPI, Karnovsky Performance Score; PFS, progression-free survival; ROC, receiver operating characteristic. Assessment of therapy response at the first follow-up 4 -6 wk after the completion of radiotherapy and PFS were performed according to the revised RANO criteria based on clinical data and MRI with a median follow-up of 9.2 mo (range 1.6 -40.47 mo). 9,25 Pseudoprogression was distinguished from progressive disease based on follow-up. PFS was assessed as the time between the start of radiotherapy and disease progression. Final decisions were made in consensus by two radiologists with 5 and 10 y of experience in neuroimaging (N.K. and D.P.). To account for prescribed antiangiogenic drugs, potential adaptions of therapy regimens and clinical status changes, assessment results were matched with institutional interdisciplinary tumor-board decisions.

Histology
Histology after biopsy or surgical resection was available for 60 out of 61 study participants. Histology of one participant being treated for newly diagnosed glioma with midline location could not be secured due to poor clinical status. Clinical routine work-up included determination of IDH-, ATRX-, LOH1p19q-, H3K27M-, and MGMT-status and glioma classification was performed according to WHO 2016 criteria. For detailed information on tumor histopathologies of the study cohort, please see Table 1.

Image acquisition and postprocessing
All images were acquired on a 3 T whole-body MR scanner (MAGNETOM Prisma; Siemens Healthineers) with a 64-channel receive head/neck coil and an integrated transmit body coil. Processing of CEST data was performed using customized scripts in Matlab® (Mathworks) version 2019b.

Lorentzian-fit based and relaxation-compensated CEST imaging of the APT and ssMT
The image-readout parameters (matrix = 128 × 104 × 16, resolution = 1.7 × 1.7 × 3 mm 3 ) of the 3D spiral-centricreordered gradient-echo acquisition sequence (i.e., snapshot CEST 26,27 ) and the CEST presaturation, originally from Zaiss et al., 27 were set up as described in detail by Goerke et al. 21 Presaturation was performed with a mean amplitude of B 1 = flip angle/( ⋅t p ) = 0.6 and 0.9 μT respectively. For both CEST scans, 57 unequally distributed frequency offsets between ±250 ppm and two M 0 at −300 ppm for Clinical characteristics of all evaluated 61 study participants who received CEST MRI in the first follow-up (FU) 4 -6 wk after completion of radiotherapy normalization were acquired, resulting in a total measurement time of 7:36 min per CEST scan. The WASABI 28 (3:41 min) approach, which provides B 0 and B 1 maps, was used with the same image readout and a similar presaturation with an equally distributed sampling of 31 offsets between ±2 ppm.
For postprocessing, the CEST and WASABI data were coregistered using a rigid registration algorithm as described by. 29 Subsequently, all Z-spectra were B 0 corrected by shifting them along Δ according to the B 0 -map. Finally, the Z-spectra were denoised using a principle component analysis based algorithm. 30 Two different Lorentzian-fit based evaluation models were applied to isolate the signal contributions.
The first model (MTR Rex APT and MTR Rex MT) is additionally compensated for spillover of the direct water saturation and MT influences. These contrasts are referred to as "relaxation compensated" in the following. This model is based on a four-pool Lorentzian-fit between ±250 ppm around the water offset (0 ppm: direct water saturation, 3.5 ppm: APT, −3.5 ppm: rNOE, and −2.5 ppm: ssMT). 21 The MTR Rex contrasts of the APT and ssMT were cal- Additional B 1 -correction was performed with the two-point "contrast-correction" method as proposed by Windschuh et al. 31 The second data fit approach (PeakAreaAPT and MT const ) was adapted from Mehrabian et al., 16 and only uses the region between ±6 ppm to fit a four-pool Lorentzian (APT, AMINE, rNOE, DS) and a constant ssMT to the data of the CEST measurement at 0.6 μT: Afterward, the constant ssMT offset (MT const ) and the area under curve of the Lorentzian describing the APT-peak (PeakArea APT) were calculated. The PeakAreaAPT and MT const are collectively being referred to as Lorentzian-fit-based contrasts in the following. A representative fitted Z-spectrum and residuals of both models can be found in Figure S1. The displayed Z-spectrum is located in the CE-region of the participant with progressive disease shown in Figure 2.

2.5
Asymmetry-based CEST imaging of the APTw asym APTw asym imaging was realized with the identical pulse sequence and image readout as the CEST measurement in the previous section. Presaturation was achieved according to Zhou J,et al. 32 in line with recent consensus and recommendation guidelines 33 employing four rectangular RF pulses with a B 1 of 2 μT, pulse length (t p ) of 0.2 s and 95% duty cycle (t sat = 0.83 s). Sixteen offsets were acquired at ±4 (1), ±3.75 (2), ±3.5 (2), ±3.25 (2), and ±3 (1) ppm. Additionally, an M 0 at −300 ppm was acquired for normalization purposes.
Similar coregistration and B 0 -correction approaches were used as described in the previous section. Afterward, the APTw asym = Z(−3.5 ppm) − Z(3.5 ppm) contrast was calculated from the corrected data.

Quantitative T 1 mapping
Given that the investigated CEST contrasts are influenced by T 1 to varying degrees, we also assessed the association of T 1 with the RANO assessment results and PFS. Quantitative mapping of the longitudinal relaxation time of water (T 1 ) was performed by the application of a saturation recovery sequence 21 with recovery times (t rec ) of 0.1, 0.25, 0.5, 1.0, 1.5, 2.0, 2.5, 3.5, 5.0, 7.5, and 10.0 s and

Tumor segmentations
3D segmentations of contrast-enhancing tumor (CE) volumes, whole tumor (WT) volumes, and normal appearing white matter (NAWM) were performed by a radiologist with 5 y of experience in neuroimaging (N.K.) on contrast-enhanced T 1w and T 2w fluid-attenuated inversion recovery (T 2w -FLAIR) images with a custom-made segmentation tool in Matlab® (Mathworks) version 2019b. Segmentations were performed in all evaluable study participants (all participants, n = 61) and in one sub-cohort encompassing only participants with residual contrast enhancement on MRI (participants with CE, n = 42) ( Figure 1). The WT volume in the study participants with residual contrast enhancement encompassed CE plus peritumoral T 2w -FLAIR hyperintense signal alterations (WT [participants with CE]). In the entire study sample, the WT volumes encompassed additional 19 cases that contained only peritumoral T 2w -FLAIR hyperintense signal alterations (WT [all participants]), given the absence of residual contrast enhancement in these patients (please see Table 1). Relevant surgically induced changes (i.e., hemorrhage) were excluded by visual correlation with T 1w and T * 2w images.

Statistical analyses
Receiver operating characteristic (ROC) analyses and Mann-Whitney-U-test were performed to test for quantitative differences of CEST contrast values between participants with stable and progressive disease, as well as pseudoprogressive and progressive disease. Kaplan-Meier curves and logrank-tests were used for analyses of PFS in association with mean CEST contrast values per ROI.
In six participants, tumors showed involvement of midline structures as thalamus, midbrain, or brainstem. Diffuse midline gliomas are common in pediatric patients, but rare in adults. 34 Since glioma midline location is known to be associated with worse clinical outcome, midline gliomas were additionally excluded for PFS analysis (participants with CE, n = 37) 35 (Figure 1).
All statistical analyses were performed using in-house software in Matlab® (Mathworks) version 2019b.

Demographics of study participants
Seventy-two study participants (mean age 59 ± 16 y; 43 male) underwent CEST MRI at the first follow-up 4 -6 wk after the completion of radiotherapy. Out of 61 evaluable participants (35 male, 26 female; mean age 59 ± 16 y) 19 had progressive disease (PD) and 8 had pseudoprogression (PP). Overall, 42 participants were diagnosed as having stable disease (SD), which included the eight participants with PP. Median PFS was 5.3 mo (range 1.5 -27.1). Detailed characteristics of study participants are provided in Table 1.

MT const showed a stronger association with early response assessment according to RANO compared to PeakAreaAPT and MTR Rex MT
Contrast maps of the APTw asym , MTR Rex APT, MTR Rex MT, PeakAreaAPT, and MT const for two study participants, one with PD, and one with PP, are illustrated in Figure 2.
When comparing participants with PD and SD, the ROC analyses yielded matching results. When comparing participants with PD and SD, areas under the curve (AUC) for the MTR Rex MT were 0.71 for CE (p = 0.02) and 0.68 for WT (participants with CE) (p = 0.05) in participants with residual contrast enhancement on MRI (participants with CE). In the same sub-cohort AUCs for the PeakAreaAPT were 0.71 (p = 0.02) for CE and 0.69 for WT (participants with CE) (p = 0.04). For the MT const AUCs were 0.79 (p < 0.01) for CE and 0.78 (p < 0.01) for WT (participants with CE) in participants with residual contrast enhancement on MRI and 0.72 (p = 0.01) for WT (all participants) in all evaluable participants (all participants). When testing for differences between participants with PD and PP the MT const yielded AUCs of 0.79 (p = 0.02) for CE and 0.75 (p = 0.05) for WT (participants with CE) in participants with residual contrast enhancement on MRI. All

F I G U R E 4
Receiver operating characteristic (ROC) curves testing for capacities of the APTw asym , MTR Rex APT, MTR Rex MT, PeakAreaAPT, and MT const to differentiate between study participants with progressive disease (PD) and stable disease (SD) according to the response assessment in neuro-oncology (RANO) criteria. Participants with pseudoprogression (PP) are included in SD, given the fact that PP is formally counted as SD according to the response assessment other contrasts did not enable distinction between participants with PD and PP. Furthermore, the APTw asym and MTR Rex APT were not associated with RANO assessment at the first follow-up. ROC curves testing for differences in mean contrast values between participants with PD and SD for all CEST contrasts, as well as differences between participants with PD and PP for the MT const are shown in Figures 4 and 5.

APTw asym , PeakAreaAPT, and MT const signal intensities are associated with PFS
In Kaplan-Meier analyses, the APTw asym , PeakAreaAPT, and MT const showed significant association with PFS. Participants with APTw asym and MT const contrast values above the median of the (sub-)cohort showed shorter time to progression compared to participants with contrast values below the median. For the APTw asym , PFS was 127 versus 267 days (HR = 2.47, p = 0.01) for WT (all participants) in all evaluable participants (all participants) and 71 versus 174 days (HR = 2.63, p = 0.02) for CE after stratification for participants with residual contrast enhancement on MRI (participants with CE). For the MT const PFS was 68 versus 172 days (HR = 3.04, p = 0.01) for CE and 62 versus

F I G U R E 5
Receiver operating characteristic curves testing for the capacity of the MT const to differentiate between participants with progressive disease (PD, n = 19) and pseudo-progression (PP, n = 8) according to response assessment in neuro oncology criteria in participants with residual contrast enhancement on MRI (n = 42). AUCs were significant for contrast-enhancing tumor volume (CE, p = 0.02, red) and whole tumor volume (WT [participants with CE], p = 0.05, blue). 186 days (HR = 4.75, p < 0.01) for WT (participants with CE) in participants with residual contrast enhancement on MRI. On the contrary, for the PeakAreaAPT participants with contrast values above the median showed longer PFS with 173 versus 68 days (HR = 0.39, p = 0.03) for CE and a similar trend (148 vs. 70 days; HR = 0.48, p = 0.08) for WT (participants with CE) in participants with residual contrast enhancement on MRI. The MTR Rex APT and MTR Rex MT did not show any association with PFS ( Figure 6).

T 1 is weakly associated with RANO assessment but not with PFS
When comparing participants with PD and SD, the AUC for T 1 for the WT (all participants) volumes was 0.66 (p = 0.047) ( Figure S2). However, there was no association of T 1 with the RANO assessment results after the stratification for presence of residual contrast enhancement on MRI ( Figure S2). Furthermore, there were no significant differences of T 1 between participants with PD and SD ( Figure S3) and no significant association of T 1 with PFS ( Figure S4).

DISCUSSION
Several groups have demonstrated the potential of CEST MRI for therapy response assessment in diffuse gliomas. Kaplan-Meier analyses testing for the association of progression-free survival (PFS) with the APTw asym contrast values of whole tumor volume (WT) in all study participants (all participants, number at risk n = 61), as well as the association of PFS with the APTw asym , MTR Rex APT, MTR Rex MT, PeakAreaAPT, and MT const contrast values for contrast-enhancing tumor volume (CE) and WT in participants with residual contrast enhancement on MRI (participants with CE, number at risk n = 37). The APTw asym was the only contrast that was associated with PFS in all study participants (top row first column, HR = 2.47, p = 0.01) with shorter PFS of participants showing higher contrast values for WT (dark red). After stratification for participants with residual contrast enhancement on MRI, the APTw asym (bottom row right column, HR = 2.63, p = 0.02), PeakAreaAPT (bottom row middle column, HR = 0.39, p = 0.03), and MT const (bottom row left column, HR = 3.04, p = 0.01) were associated with PFS, with APTw asym and MT const showing higher contrast values for CE (light red) and WT (dark red) in participants with shorter survival. The PeakAreakAPT showed higher contrast values in participants with longer survival. For PFS analyses, participants with midline glioma location (n = 5) where additionally excluded from the subgroup of participants with residual contrast enhancement on MRI. differentiation of glioma progression and radiation necrosis as early as 3 mo after the completion of radiotherapy. [12][13][14] Others have reported in prospective clinical studies at 1.5, 3, and 7 T with 19 -71 patients that the APTw asym contrast was associated with PFS before treatment onset, whilst the prognostic performance of this technique rapidly decreased in patients after the start of therapy. [12][13][14]16,[22][23][24] We only observed an association of the APTw asym with PFS but not with therapy response at the first follow-up, as assessed by RANO criteria.
In prospective clinical studies at 7 T with 20 and 26 patients with newly diagnosed gliomas, Regnery et al. and Paech et al. reported that the MTR Rex APT could predict therapy response and survival, with higher contrast values for patients with early disease progression. 19,23 However, Meissner et al. did not find any association of MTR Rex APT contrast values with treatment response after the start of radiotherapy. 18 Mirroring the findings by Meissner et al., we also did not find any association of MTR Rex APT contrast values with therapy response and PFS at the first follow-up after the completion of radiotherapy at 3 T.
For the MTR Rex MT, we only observed an association with therapy response but not with PFS. To our knowledge, this is the first report on the association of the MTR Rex MT with clinical outcomes in patients with glioma following radiotherapy.
In a prospective clinical study with 19 patients, Mehrabian et al. have demonstrated that MT const could prognosticate therapy response, whereas MT const after the start of therapy and PeakAreaAPT did not show any association with clinical outcome. 16 In our study with 61 evaluable participants, MT const and PeakAreaAPT demonstrated the highest association with therapy response and PFS. Furthermore, the MT const could also differentiate between participants assessed as having progressive and pseudoprogressive disease.
Generally, for the interpretation of these study findings it is essential to consider well-known diagnostic limitations of the RANO criteria. RANO compares volumetric changes of contrast enhancement on T 1w and abnormal T 2w -FLAIR signal hyperintensities on consecutive MRI exams under consideration of clinical status and application of antiangiogenic drugs. However, in clinical practice blood-brain-barrier disruption and edema caused by disease progression or radiotherapy are in most cases not easily distinguishable and frequently occur to variable degrees simultaneously. Furthermore, other factors such as therapeutically induced granulation tissue and inflammatory processes might also contribute to respective changes on MRI. [7][8][9]25 Regarding CEST, the Lorentzian-fit-based PeakAreaAPT and MT const contrasts have several other contributions, such as T 1 , spill-over, and unwanted ssMT in case of the PeakAreaAPT. 16,36 The asymmetry-based APTw includes, in addition to those, contributions from the relayed nuclear Overhauser effect (rNOE). 32,36 In relaxation-compensated CEST contrasts the APT and ssMT peaks are corrected for spill-over effects, even though also these contrasts have certain T 1 contributions. 21,36 In this regard, it can be speculated that the superior performance of the Lorentzian-fit-based contrasts in this study might be explained by a summation of several biomolecular changes which influence T 1 relaxation, spill-over, and CEST effects as APT, rNOE, and ssMT to varying degrees. T 1 is heavily influenced by the water content of examined volumes. Therefore, the higher MT const values in tumor tissue compared to NAWM, which is reflected by higher T 1 ( Figure S3), might be linked to the tumor-associated disruption of the blood brain barrier and resulting edema. However, the fact that in this study only a weak association of T 1 with therapy response according to RANO but not with PFS was observed underscores the added clinical value of the investigated Lorentzian-fit-based CEST contrasts. Nevertheless, the finding that the MTR Rex MT values in tumor volumes were lower compared to NAWM, which is in line with previously published results for contrasts of the ssMT, 22,37 indicates that the investigated Lorentzian-fit-based contrasts have stronger remaining T 1 contributions compared to the investigated relaxation-compensated contrasts.
This study has several limitations. Since true disease progression is secured by histology only in a fraction of cases, relying on RANO assessment for determination of clinical endpoints is a major weakness of this study due to well-known shortcomings of the RANO criteria. Therefore, we additionally included PFS analysis and long-term follow-up (median 9.2 mo) to enable best possible ground truth assessment in the clinical routine setting. Furthermore, the included study participants had different types of diffuse gliomas which were resected to varying extents and treated with varying types and doses of radiation. To overcome such limitations, future studies should aim for even larger patient cohorts (e.g., in multicenter trials) with longitudinal CEST MRI data acquisition. In this context, leading experts in the field have recently published consensus recommendations for APTw asym CEST imaging in neuro-oncologic applications in order to provide a rationale for optimized APTw asym imaging at 3 T, including specific recommendations for pulse sequences, acquisition protocols, and data processing methods. 33

CONCLUSIONS
In patients with glioma at the first follow-up 4 -6 wk after the completion of radiotherapy, MT const , PeakAreaAPT, and APTw asym can predict clinical outcomes by means of PFS, whereas MT const can distinguish pseudoprogression from disease progression, as assessed according to RANO criteria. These findings indicate that Lorentzian-fit-based and asymmetry-based CEST metrics of the APT and ssMT early after the completion of radiotherapy might have synergistic potential for supporting clinical decision making in patients with glioma.

SUPPORTING INFORMATION
Additional supporting information may be found in the online version of the article at the publisher's website. fit for the 4-pool Lorentzian-fit and constant ssMT model with the added amine pool, which was fitted in the range of −6-6 ppm (C). Figure S2. Receiver operating characteristic (ROC) curves testing for the capacity of T 1 to differentiate between study participants with progressive disease (PD) and stable disease (SD) according to the response assessment in neuro-oncology (RANO) criteria. Participants with pseudoprogression (PP) are included in SD, given the fact that PP is formally counted as SD according to the response assessment in neuro-oncology (RANO) criteria.  Figure S3. Violin plots of T 1 for study participants with residual contrast enhancement on MRI (n = 42). The graph shows absolute T 1 values for contrast-enhancing tumor volume (CE), whole tumor volume (WT) and normal appearing white matter (NAWM). Red columns show cases with progressive disease (PD, n = 19), green columns with stable disease (SD, n = 23) and blue columns with pseudoprogression (PP, n = 8). Green columns (SD) also include cases with PP, given the fact that PP is formally counted as SD according to the response assessment in neuro-oncology (RANO) criteria. There were no significant differences of T 1 between the tumor volumes of participants with PD, SD and PP. Figure S4. Kaplan-Meier analyses testing for the association of progression-free survival (PFS) with T 1 of whole tumor volume (WT) and contrast-enhancing tumor volume (CE) in study participants with residual contrast enhancement on MRI (part. w. CE, number at risk n = 37). However, no significant association of T 1 with PFS could be observed. For PFS analyses, participants with midline glioma location (n = 5) where additionally excluded from the subgroup of participants with residual contrast enhancement on MRI.