Reproducibility of 3 T APT‐CEST in Healthy Volunteers and Patients With Brain Glioma

Abstract

Background

Amide proton transfer (APT) imaging is a chemical exchange saturation transfer (CEST) technique offering potential clinical applications such as diagnosis, characterization, and treatment planning and monitoring in glioma patients. While APT‐CEST has demonstrated high potential, reproducibility remains underexplored.

Purpose

To investigate whether cerebral APT‐CEST with clinically feasible scan time is reproducible in healthy tissue and glioma for clinical use at 3 T.

Study Type

Prospective, longitudinal.

Subjects

Twenty‐one healthy volunteers (11 females; mean age ± SD: 39 ± 11 years) and 6 glioma patients (3 females; 50 ± 17 years: 4 glioblastomas, 1 oligodendroglioma, 1 radiologically suspected low‐grade glioma).

Field Strength/Sequence

3 T, Turbo Spin Echo ‐ ampling perfection with application optimized contrasts using different flip angle evolution ‐ chemical exchange saturation transfer (TSE SPACE‐CEST).

Assessment

APT‐CEST measurement reproducibility was assessed within‐session (glioma patients, scan session 1; healthy volunteers scan sessions 1, 2, and 3), between‐sessions (healthy volunteers scan sessions 1 and 2), and between‐days (healthy volunteers, scan sessions 1 and 3). The mean APT_CEST values and standard deviation of the within‐subject difference (SD_diff) were calculated in whole tumor enclosed by regions of interest (ROIs) in patients, and eight ROIs in healthy volunteers—whole‐brain, cortical gray matter, putamen, thalami, orbitofrontal gyri, occipital lobes, central brain—and compared.

Statistical Tests

Brown‐Forsythe tests and variance component analysis (VCA) were used to assess the reproducibility of ROIs for the three time intervals. Significance was set at P < 0.003 after Bonferroni correction.

Results

Intratumoral mean APT_CEST was significantly higher than APT_CEST in healthy‐appearing tissue in patients (0.5 ± 0.46%). The average within‐session, between‐sessions, and between‐days SD_diff of healthy control brains was 0.2% and did not differ significantly with each other (0.76 > P > 0.22). The within‐session SD_diff of whole‐brain was 0.2% in both healthy volunteers and patients, and 0.21% in the segmented tumor. VCA showed that within‐session factors were the most important (60%) for scanning variance.

Data Conclusion

Cerebral APT‐CEST imaging may show good scan–rescan reproducibility in healthy tissue and tumors with clinically feasible scan times at 3 T. Short‐term measurement effects may be the dominant components for reproducibility.

Level of Evidence

2

Technical Efficacy

Stage 2

Novel imaging biomarkers in glioma patients may help diagnosis, prognosis, and treatment decisions by improving tumor characterization and delineation. ¹ , ² , ³ One such imaging biomarker is amide proton transfer (APT) imaging. APT is a subset of chemical exchange saturation transfer (CEST) imaging that allows for quantitative amide and peptide detection. ¹ , ⁴ , ⁵ Specifically, APT‐CEST may be especially useful in glioma imaging as these primary brain tumors express increased amounts of protein. ⁶ , ⁷ , ⁸ This magnetic resonance imaging (MRI) technique measures the intensity change in bulk water after the magnetization in amide protons is selectively saturated and transferred by chemical exchange to water protons. ⁶ This intensity change indirectly reflects protein concentration in the tissue, computed by the magnetization transfer ratio asymmetry (MTR_asym). ⁴ , ⁵ , ⁹

Multiple clinical applications have been proposed in recent years for neuro‐oncological APT‐CEST imaging and show promising results regarding diagnosis, characterization, and treatment planning and monitoring in glioma. ⁸ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ Also in other clinical contexts, APT‐CEST has demonstrated its potential clinical usability, eg, in the detection of stroke for highlighting ischemic areas, paving its way to a near‐future standard application. ¹⁵ , ¹⁶ , ¹⁷ However, particularly for routine clinical application, high reproducibility in healthy and tumorous tissue is crucial. Quantitative assessment of the reproducibility of differences between healthy and tumorous tissue is mandatory for the development of imaging biomarkers with established cut‐off values to non‐invasively differentiate between tissue conditions.

Previous preliminary findings in reproducibility studies were promising for clinical 3 T APT‐CEST applications. ¹⁸ , ¹⁹ , ²⁰ One study showed consistently higher within‐session than between‐sessions reproducibility for 3 T APT‐CEST with partial brain coverage and relatively long scanning time in healthy volunteers and glioma patients. ¹⁸ Another study found good reproducibility in brain tumors. ¹⁹ Spatially homogenous radiofrequency (RF) shimming and B0‐inhomogeneity correction could contribute to this reproducibility, but the authors of this study exclusively investigated mean tumor APT‐CEST values in a single slice. A third study found good reproducibility in only two healthy volunteers. ²⁰

The purpose of this study was to investigate the short‐ and long‐term reproducibility of 3 T APT‐CEST measurements in a comparatively large group of healthy volunteers and a pilot level group of glioma patients, with a clinically applicable scan protocol with full brain coverage and a short scan duration.

Methods

This prospective study was approved by the institutional review board (VUmc_2021‐0038) and written informed consent was obtained from all participants.

Study Design

Healthy volunteers were recruited non‐commercially through external advertisements. Inclusion criteria were 1) age >18 years and 2) neither clinical history nor MRI evidence of brain pathology. Prospective recruitment also included patients with suspected recurrent or de‐novo glioma on external MRI and referred for tumor resection between May 2021 and July 2021. Patients were included if 1) age >18 years, 2) they had a radiological diagnosis of de‐novo glioma or recurrent already confirmed glioma, 3) received clinically indicated MRI, and 4) had no other concurrent brain pathology at the time of diagnosis.

Participants with severe motion artifacts in T1‐weighted (T1w), contrast‐enhanced T1‐weighted (ceT1w), or APT‐CEST scans were excluded. No participants needed to be excluded due to compromised image quality.

Participants

Twenty‐one healthy volunteers (mean age ± SD: 39 ± 11 years, 11 women) and 6 glioma patients (mean age ± SD: 50 ± 17 years, 3 women) were scanned. Five patients subsequently received a histopathological confirmation according to the 2021 World Health Organization (WHO) criteria: four glioblastomas (GBMs; WHO grade 4) and one oligodendroglioma (WHO grade 2). ²¹ One patient opted against surgery due to the high radiological likelihood of a slow‐growing low‐grade glioma. Power analysis to determine group size to detect change in APT‐CEST for GBM and LGG is given in Table S1 in the Supplemental Material.

Protocol of Scan–Rescan Test

Healthy volunteers were subjected to three APT‐CEST scan sessions (Fig. 1). The first two sessions were separated by participant repositioning (between‐sessions reproducibility) and the last session occurred 4–14 days later (between‐days reproducibility). Patients underwent one scan session (Fig. 1). Each session, in both volunteers and patients, contained two consecutive APT‐CEST scans (within‐session reproducibility) with a B0‐map in between. All participants received a structural three‐dimensional (3D) T1‐weighted acquisition (T1w) while the patients additionally received a complete neuro‐oncological MRI protocol according to the European Organisation for Research and Treatment of Cancer standard including 3D T1‐weighted post‐contrast (ceT1w) and fluid‐attenuated inversion recovery. ²² APT‐CEST scanning was performed before the neuro‐oncological MRI protocol.

FIGURE 1.

(a) Scan protocol for the healthy volunteers. (b) Scan protocol for the patients. (c) Scan session. A session contains two consecutive APT‐CEST scans and one B0 map. (a) Scans: 1 = baseline, 2 = scan directly after the baseline—i.e., 1–2 = within‐session reproducibility—scan 4 is a scan 30 minutes after the first—i.e., 1–4 = between‐sessions reproducibility—scan 6 = another day—i.e., 1–6 = between‐day reproducibility. (b) Glioma patients underwent one session.

APT‐CEST Image Acquisition and Post‐Processing

All participants were scanned on a 3‐T MRI (Vida, Siemens Healthineers, Erlangen, Germany) using a 20‐channel head coil, including whole‐brain 3D MPRAGE T1w (repetition time (TR) 2300 msec, echo time (TE) 2.32 msec, inversion time (TI) 900 msec, refocusing flip angle 8°, field of view 240 mm, slice thickness 0.9 mm) and TSE SPACE APT‐CEST (TR 3000 msec, TE 17 msec, turbo factor 140, refocusing flip angle 120°, 2.8 × 2.8 × 2.8 mm³ resolution, 7 frequencies with saturation pulses [10 Gaussian pulses of 100 msec at 2.0 μT] at ±3.0 ppm, ±3.5 ppm, ±4.0 ppm, and 1560 ppm off‐resonance, scan duration: 4:36 minutes) plus a dual‐echo gradient recalled echo (GRE) B0 map for correction of spatial inhomogeneity of the resonance frequency. ²³

All off‐resonance scans were registered to the 1560 ppm off‐resonance scan to correct for head motion (MCFLIRT, FSL 5.0.9). ²⁴ APT‐CEST maps were then calculated as MTR_asym at 3.5 ppm off‐resonance from motion‐corrected images:

$${\mathrm{APT}}_{\text{CEST}}={\mathrm{MTR}}_{\text{asym}}\left(3.5\mathrm{ppm}\right)=\frac{S_{\mathrm{sat}}\left(-3.5\mathrm{ppm}\right)-S_{\mathrm{sat}}\left(3.5\mathrm{ppm}\right)}{S_0}.$$

The S _sat and S ₀ represent the imaging signal intensities measured with RF saturation at ±3.5 ppm and at 1560 ppm, respectively. The B0 map was expressed in ppm after unwrapping by FSL (Version 5.0.9) Prelude. The measured points on the Z‐spectrum (off‐resonance saturation spectrum) were interpolated to the actual frequency shift as measured by the B0 map. We extrapolated the measured points on the Z‐spectrum if the frequency shift was between ±0.5 ppm and ± 1 ppm. Voxels with a shift larger than 1 ppm were masked (set to 0). Next, the 1560 ppm off‐resonance APT‐CEST scan was registered to the non‐contrast 3D T1w image with co‐registration of the APT‐CEST map using FLIRT (FSL 5.0.9) for healthy volunteers and patients. ²⁴ The 3D T1w scan was spatially normalized to Montreal Neurosciences Institute (MNI) space using CAT12.7 (build 1615) ²⁵ and again with co‐registration of the APT‐CEST map, which was then smoothed by a 3.5 × 3.5 × 3.5 mm³ full width at half maximum (FWHM) Gaussian filter. No skull stripping was performed.

Image Analysis

Five anatomical regions of interest (ROIs) were available in MNI space using the Harvard‐Oxford atlas ²⁶ : the whole brain including cerebellum and internal cerebral spinal fluid spaces, cortical gray matter, deep white matter, putamen, and thalami. Three additional anatomical ROIs were manually delineated near the orbitofrontal gyri, the occipital lobes, and the central brain by (I.J.H.G. Wamelink) with 1 year of experience under supervision of a neuroradiologist with 10 years of experience (V.C. Keil) (Fig. 2). The orbitofrontal gyri represent a region that is susceptible to B0‐inhomogeneities. ²⁷ All three ROIs contained both white and gray matter.

FIGURE 2.

Axial, coronal, and sagittal slices of the three manually delineated regions of interest (ROIs).

In patients with glioma, the ROIs enclosing presumably vital glioma tissue were manually delineated, together with similar‐sized contralateral tissue ROIs, on the ceT1w for enhancing tumors and based on FLAIR for the non‐enhancing tumors by a neuroradiologist with 10 years of experience (V.C. Keil).

By multiplying the binary ROI atlases with the APT‐CEST scans the mean APT_CEST values were calculated for all scans. Next, the SD across all measurements was taken for the eight ROIs. The SD across subjects of the within‐subject difference (SD_diff) between the mean APT_CEST values was calculated for the eight ROIs and each of the three reproducibility time intervals—within‐session, between‐sessions, and between‐days. Additionally, voxel‐wise SD_diff maps were computed from non‐averaged APT_CEST values for the three reproducibility time intervals to visually identify locations of low reproducibility.

Statistics

The Brown‐Forsythe test (Python, Python Software Foundation, version 3.6.9, package: Scipy, version 1.5.2) was used to test the variance of the SD_diff among the three reproducibility time intervals of the eight anatomical ROIs. Bonferroni correction was performed by dividing the significance threshold (0.05) by the number of separate tests (8 ROIs × 2 tests = 16). Variance component analysis (VCA) (Rstudio Version: 1.4.1106; VCA Version: 1.4.3) was performed on the mean APT_CEST value of the eight anatomical ROIs to find the contributions of the different effects. The Brown‐Forsythe test and VCA were only performed on healthy volunteers.

The agreement of the APT_CEST values in tumor patients was assessed by the intraclass correlation coefficient (ICC) (Python, Python Software Foundation, version 3.6.9, package: Pingouin, version 0.3.12). The within‐session ICC was computed for the tumor and contralateral ROI. The ICCs were calculated with a 95% confidence interval (CI) using an average random raters model, and given the following classifications: 0.00–0.39, poor; 0.40–0.59, fair; 0.60–0.74, good; and 0.75–1.00 excellent. ²⁸

Bland–Altman plots were created to show differences between the three reproducibility time intervals in healthy volunteers. The Mann–Whitney U test (Python, Python Software Foundation, version 3.6.9, package: Scipy, version 1.5.2) was performed to compare the mean APT_CEST values of the different structures in healthy volunteers with the mean tumor values in patients. Statistical significance was determined at P < 0.006 with Bonferroni correction (0.05/8).

The effect size (E) was only computed for patients:

$$\text{Effect}\text{size}\left(E\right)=\frac{\text{Tumor}‐\text{to}‐\text{normal}\text{tissue}\text{difference}}{{\mathrm{SD}}_{\text{diff}}}.$$

Results

APT_CEST Descriptive Statistics in Volunteers and Patients

The mean and SD for the anatomical and tumor ROIs are shown in Table 1. The mean APT_CEST difference between ROIs in GBMs and contralateral tissue was 1.11% (N = 4, Table 1, Fig. 3). Intratumoral mean APT_CEST (1.59 ± 0.67%, N = 6) was significantly higher than contralateral similar‐sized ROI APT_CEST in healthy‐appearing tissue in patients (0.5 ± 0.46%), and also significantly higher than mean APT_CEST in anatomical ROIs in healthy volunteers (between 0.42 and 1.02, N = 21).

Table 1.

Mean Regional APT_CEST Values in Healthy Volunteers and Glioma Patients

Dataset	Whole Brain	Cortical Gray Matter	Deep White Matter	Putamen	Thalami	Orbitofrontal Gyri	Occipital Lobes	Central Brain	Tumor ROI	Contralateral ROI
Healthy volunteers (N = 21)	0.88 ± 0.93	0.82 ± 0.85	0.42 ± 0.53	0.95 ± 0.45	1.39 ± 0.46	0.26 ± 1.81	1.02 ± 0.65	0.87 ± 0.54
Patients (N = 6)	0.93 ± 0.95								1.59 ± 0.67	0.5 ± 0.46
GBM, WHO 4 (N = 4)	1.02 ± 0.17								1.93 ± 0.51	0.83 ± 0.34
Oligodendroglioma, WHO 2 (N = 1)	0.66 ± 0.04								0.57 ± 0.29	0.14 ± 0.05
LGG, radiologically WHO 2 (N = 1)	0.93 ± 0.06								1.24 ± 0.04	0.14 ± 0.02

APT_CEST = amide proton transfer chemical exchange saturation transfer; GBM = glioblastoma; LGG = low grade glioma; ROI = region of interest.

FIGURE 3.

Single transversal APT‐CEST slices of four different glioblastoma patients. APT_CEST value in color scale projected on the post‐contrast T1‐weighted image, within the tumor ROI only. Red and green boxes are positioned around the tumor and contralateral ROIs, respectively. Note the heterogeneous hyperintensity in the tumorous regions that show hyperintensity on the contrast‐enhanced T1‐weighted image when compared to the contralateral region (a–c). (d) The APT‐CEST value of a non‐enhancing glioblastoma. APT = amide proton transfer; CEST = chemical exchange saturation transfer; ROI = region of interest.

Reproducibility in Healthy Volunteers

Figure 4 shows high within‐session reproducibility and only slightly higher between‐sessions and between‐days voxel‐wise SD_diff. The highest within‐session voxel‐wise SD_diff variance was found in the orbitofrontal gyri, as also shown by the higher ROI SD_diff (Table 2). The SD_diff was consistent within each of the three time intervals and only showed a small increase (maximum of 0.11%) between them.

FIGURE 4.

Voxel‐wise SD maps of the within‐subject difference between scan sessions 1–2, 1–4, and 1–6 (rows a, b, and c, respectively), for all healthy volunteers (N = 21). Color scale represents the SD of the difference in APT_CEST value (%) between sessions (SD_diff), where higher SD_diff means lower reproducibility.

Table 2.

The SD_diff of the Eight ROIs and Each of the Three Reproducibility Time Intervals

	Whole Brain	Cortical Gray Matter	Deep White Matter	Putamen	Thalami	Orbitofrontal Gyri	Occipital Lobes	Central Brain
Within‐session (%)	0.20	0.20	0.19	0.22	0.26	0.61	0.22	0.20
Between‐sessions (%)	0.24 P = 0.35	0.25 P = 0.28	0.26 P = 0.39	0.28 P = 0.22	0.28 P = 0.31	0.67 P = 0.76	0.26 P = 0.33	0.27 P = 0.33
Between‐days (%)	0.22 P = 0.17	0.22 P = 0.20	0.24 P = 0.11	0.33 P = 0.01	0.33 P = 0.05	0.48 P = 0.58	0.24 P = 0.14	0.26 P = 0.03

The P‐value is the comparison between the within‐session difference and either the between‐sessions and or between‐days difference for the brain structures in healthy volunteers. P‐values less than 0.003 were considered significant.

SD_diff = standard deviation across subjects of the within‐subject difference; ROI = region of interest.

The between‐days SD_diff was consistently slightly lower than the between‐sessions SD_diff. However, the Brown‐Forsythe test showed a non‐significant difference between the between‐sessions and between‐days reproducibility (P = 0.75, N = 21). The SD_diff increase between sessions was small and the reproducibility did not significantly differ between the within‐session and between‐sessions and the within‐session and between‐days (Table 2). No statistically significant differences were found after Bonferroni correction. The within‐session SD_diff of the whole brain for all patients was 0.2% (N = 6).

The VCA showed that total variance (0.03–0.07) was relatively low compared to the difference between GBMs and contralateral tissue (1.11%). The variance appears to be predominated by the factor within‐session rather than any other factor (Table 3). This is consistent with the SD_diff maps (Fig. 4). Figures 5 and 6 show Bland–Altman plots of the mean APT_CEST value against the APT_CEST value session differences of the brain ROIs.

Table 3.

Percentage of Total Variance Computed by the Variance Component Analysis

	Whole Brain	Cortical Gray Matter	Deep White Matter	Putamen	Thalami	Orbitofrontal Gyri	Occipital Lobes	Central Brain
Total variance (APTCEST %)	0.033	0.039	0.034	0.069	0.058	0.205	0.041	0.043
Factor subject	16%	28%	11%	34%	13%	42%	12%	22%
Factor day	19%	16%	21%	16%	4%	14%	20%	8%
Factor session	5%	2%	13%	14%	22%	1%	8%	24%
Factor within‐session	60%	54%	55%	36%	61%	43%	60%	46%

Largest contributor to total variance is marked in bold.

APT_CEST = amide proton transfer chemical exchange saturation transfer.

FIGURE 5.

Bland–Altman plots. Amide proton transfer chemical exchange saturation transfer (APT_CEST) differences of the five different brain structures are plotted against their mean APT_CEST value. Continuous and broken lines indicate mean difference and 95% limits of agreement (mean difference ± 1.96 SD of the paired difference) respectively.

FIGURE 6.

Bland–Altman plots. Showing differences of reproducibility in the orbitofrontal gyri, central brain, and occipital lobes. Note the larger limits of agreement in the orbitofrontal gyri and the narrower limits of agreement in the occipital lobes and central brain regions.

Reproducibility in Patients

The within‐session reproducibility of the tumors was slightly lower (SD_diff, 0.21%, N = 6) than in the contralateral ROIs (SD_diff, 0.11%), but without a statistical significance (P = 0.73). The mean whole‐brain APT_CEST values did not significantly differ between healthy volunteers and patients (0.88 ± 0.93%, 0.93 ± 0.95% respectively; P = 0.13). Mean tumor values did however differ significantly with the eight ROIs in healthy volunteers. Within‐session scan agreement was the highest in contralateral ROIs (ICC = 0.99; 95% CI, 0.92–1.00; N = 21) followed by glioma (ICC = 0.97; 95% CI, 0.82–1.00; N = 6). The effect size of the APT_CEST in GBM was 5 when considering the ROI‐wise within‐session SD_diff of the putamen.

Discussion

We found high reproducibility across the three reproducibility time intervals in healthy volunteers. Within‐session measurement agreement in glioma, contralateral‐to‐glioma, and healthy brain ROIs was excellent. Furthermore, within‐session reproducibility in whole‐brain APT‐CEST scans of glioma patients was equal to the reproducibility in healthy volunteers and even better for tumor ROIs. The variance thus seems to be predominated by general error contributors such as thermal noise, subject motion, B0 inhomogeneity, and stress on scanner hardware, rather than long‐term instrumental errors, physiological variance, or between‐subject differences. These results may suggest that 3 T APT‐CEST provides sufficient reproducibility for clinical application, especially given the larger APT_CEST value changes between glioma and normally appearing tissue.

While the between‐sessions reproducibility was not evaluated previously, the within‐session and between‐days reproducibility of APT‐CEST scans of this study are in line with previous studies in healthy volunteers ¹⁸ , ²⁰ and brain tumor patients. ¹⁸ , ¹⁹ , ²⁰ The within‐session, between‐sessions, and between‐days reproducibility presented in the current study showed a trend of decrease—although not statistically significant—for the SD_diff and the VCA. Between‐days reproducibility showed higher consistency than between‐sessions reproducibility, albeit not to a statistically significant degree. The first two studies used scan protocols and acquisition parameters similar to ours with an exclusive focus on mean APT_CEST values. ¹⁸ , ¹⁹ Our current study pursues a different perspective by also performing a voxel‐wise reproducibility analysis in healthy tissue of patients and healthy participants to focus on the reproducibility of relative APT_CEST values. Both studies found similar within‐session scan agreement for glioma. ¹⁸ , ¹⁹

Our results show that the different imaging sessions had no statistically significant effect on APT_CEST measurement values, which is in agreement with a previous similar analysis of variance in healthy volunteers and glioma patients. ¹⁸ Our findings are comparable to another study with two healthy volunteers. ²⁰ Furthermore, our difference between glioma measurements and healthy appearing contralateral tissue at all time points was in agreement with previous studies. ¹³ , ¹⁸ , ²⁹ While APT‐CEST reproducibility in the brain has not yet been investigated extensively, our reproducibility was similar to that of studies in other body parts such as breast (7 T) and prostate (3 T) tissue and different CEST sequences such as glutamate‐CEST. ³⁰ , ³¹ , ³² , ³³

We found the lowest reproducibility in the orbitofrontal gyri, which can be explained by the susceptibility of these regions to B0‐inhomogeneity that impacts the Z‐spectrum. ¹³ Moreover, regions with large veins, such as areas close to the superior sagittal sinus, also showed high variance consistent with the literature. ²⁹ We corrected for the B0‐inhomogeneity by creating a B0‐map between both APT‐CEST scans. ³⁴ The B0‐map was scanned once for each session in order to limit the session duration. We thus implicitly assumed that the measurement error of the B0‐map itself does not contribute to APT_CEST variability. However, the within‐session variability does include subject motion between CEST and B0 scans. Other potential causes are subject movement in general, and the pre‐ and post‐processing steps, which is why we applied the same sequence parameters during the entire study. Finally, low reproducibility could be caused by physiological variation of amide concentrations.

With good reproducibility, APT‐CEST imaging shows potential for clinical implementation. The SD_diff was consistent within each of the three time intervals and only slightly increased between the different time intervals. Moreover, the SD_diff was small compared to the tumor signal change, denoting a detectable difference between healthy and tumorous tissue, in particular for GBMs. This opens the door for potential clinical biomarker applications such as percentage differences in APT‐CEST values that can be used to differentiate tumors from viable tissue. A potential benefit of APT‐CEST over conventional tumor imaging, more specifically ceT1w imaging, is that APT‐CEST might be able to identify non‐contrast‐enhancing parts of high‐grade gliomas. ²⁹ Previous studies have also shown that APT imaging is able to differentiate between solitary brain metastases and GBMs or radiation necrosis and viable tumor tissue. ³⁵ , ³⁶ , ³⁷ These findings are particularly important for quick and non‐invasive disease staging and tumor classification, which may avoid unnecessary biopsies, but also open potential alternatives to gadolinium contrast administration. ³⁸

Limitations

We only investigated at intra‐vendor reproducibility in this study. It is thus unclear whether the results can be generalized to different scanners and vendors. Inter‐vendor reproducibility is important to study for clinical implementation as scanning a patient on the same scanner for follow‐up assessment is not always clinically feasible. Another limitation is that our sample size for glioma patients remains on a pilot study level. Finally, we could have scanned more offset frequencies to improve correction for off‐resonance effects but this would have prolonged the scan duration to clinically unfeasible times.

Conclusion

The findings of this study indicate that whole‐brain APT‐CEST imaging with clinical scan time has a sufficiently high short‐ and long‐term reproducibility in tumors and healthy tissue at clinically feasible scan times at 3 T. The tumor–normal tissue contrast was larger than the APT‐CEST scan–rescan errors. Instrumental noise seems to be the largest component, while long‐term physiological variance and between‐subject variability were smaller. Reproducibility might be lower near regions with B0‐inhomogeneity.

Supporting information

Table S1 Power analysis to determine group size for statistical tests. Alpha = 0.05, power = 0.8, and sigma = SD_diff/sqrt(2).