Amide proton transfer imaging for differentiation of benign and atypical meningiomas

Abstract

Purpose

To investigate the difference in amide proton transfer (APT)-weighted signals between benign and atypical meningiomas and determine the value of APT imaging for differentiating the two.

Methods

Fifty-seven patients with pathologically diagnosed meningiomas (benign, 44; atypical, 13), who underwent preoperative MRI with APT imaging between December 2014 and August 2016 were included. We compared normalized magnetization transfer ratio asymmetry (nMTR_asym) values between benign and atypical meningiomas on APT-weighted images. Conventional MRI features were qualitatively assessed. Both imaging features were evaluated by multivariable logistic regression analysis. The discriminative value of MRI with and without nMTR_asym was evaluated.

Results

The nMTR_asym of atypical meningiomas was significantly greater than that of benign meningiomas (2.46% vs. 1.67%; P < 0.001). In conventional MR images, benign and atypical meningiomas exhibited significant differences in maximum tumour diameter, non-skull base location, and heterogeneous enhancement. On multivariable logistic regression analysis, high nMTR_asym was an independent predictor of atypical meningiomas (adjusted OR, 11.227; P = 0.014). The diagnostic performance of MRI improved with nMTR_asym for predicting atypical meningiomas.

Conclusion

Atypical meningiomas exhibited significantly higher APT-weighted signal intensities than benign meningiomas. The discriminative value of conventional MRI improved significantly when combined with APT imaging for diagnosis of atypical meningioma.

Introduction

With advances in imaging techniques and knowledge of tumour microenvironment, chemical exchange saturation transfer (CEST) was proposed as a novel magnetic resonance imaging (MRI) technique for detecting endogenous mobile proteins and peptides, even at relatively low molecular concentrations [1]. Amide proton transfer (APT) imaging is a specific form of CEST imaging based on chemical exchange of amide protons without exogenous contrast agents [2–5]. Here, exchangeable labile solute protons in amide bonds are selectively saturated by 3.5-ppm downfield off-resonance radiofrequency (RF) irradiation and transferred to bulk water by proton exchange, leading to a discernible decrease in signal intensity in bulk water and, eventually, to image contrast [4]. This technique has been successfully applied in human brain imaging for several diseases, including tumours and stroke [5–9]. Although the origin of APT signals in tumours is controversial, previous studies have suggested an increase in APT-weighted signals within brain tumours, especially in malignant tumours, due to increased mobile protein concentrations in malignant cells, associated with increased cellularity [2, 3, 5, 6, 9–11].

Intracranial meningioma is a common primary brain tumour in adults [12, 13]. A majority of meningiomas are benign (World Health Organization [WHO] grade I) and have favourable prognosis. However, nearly 10% of meningiomas are atypical (WHO grade II) or malignant (WHO grade III) tumours with aggressive biological behaviour, a tendency to recur [14], and, consequently, relatively worse prognosis, as evident from 5-year survival rates of 67.5% and 60% for patients with Grade II and III meningiomas, respectively [15]. In routine clinical practice, surgical resection followed by adjuvant radiotherapy is considered for treatment of atypical/malignant meningioma, especially in cases involving subtotal resection [16]. Preoperative prediction of meningioma grade is important because it influences treatment planning, including the surgical resection strategy. For differentiating between benign and atypical/malignant meningiomas, several conventional MRI findings have been suggested as indicators of aggressive tumours, including non-skull base location, T2 hyperintensity, peritumoural brain oedema, indistinct tumour-brain interface, irregular tumour margins, heterogeneous tumour enhancement, capsular enhancement, cystic changes, and skull invasion [17–23]. These findings, however, are not sufficiently reliable for grading meningiomas to ensure proper treatment planning. Although several studies have investigated the feasibility of APT imaging for grading glial tumours [9, 24], to the best of our knowledge, no study has employed APT imaging for grading meningiomas. Therefore, this study aimed to investigate the differences in APT-weighted signal between benign and atypical meningiomas and determine the value of APT imaging for differentiating the two.

Materials and Methods

Study Population

This retrospective study was approved by the institutional review board, and the requirement for informed consent was waived. Sixty consecutive patients with pathologically diagnosed meningiomas, who underwent preoperative MRI including APT and conventional imaging between December 2014 and August 2016, were enrolled. We excluded three patients because of poor image quality due to the susceptibility effect or patient motion (n = 2) and inadequate lesion size for image analysis (n = 1). Thus, 57 patients (male, 16; female, 41; mean age 54.4 ± 12.4 years) were included in this study. The interval between MRI and surgery was less than 3 days for all patients.

MRI Protocol

Imaging studies were performed with a 3T MRI system (Achieva TX, Philips) using a 32-channel receiver head coil and a transmitter body coil. Conventional MR images were acquired according to the standard extra-axial brain tumour protocol in our hospital, which involves the following: (a) T2-weighted imaging (T2WI) in all three orthogonal directions; (b) axial T1-weighted imaging; and (c) three-dimensional (3D) contrast-enhanced T1WI. Additionally, APT images were acquired before contrast administration using the 3D gradient- and spin-echo approach [25] with the following imaging parameters: acquisition voxel size, 2.2 × 2.2 mm; slice thickness, 4.4 mm; TR/TE, 3000/17 ms; turbo spin-echo factor, 22; echo planar imaging factor, 7; and number of slices to cover the entire tumour, 15. Six saturation frequency offsets (±3.0, ±3.5 and ±4.0 ppm) were adopted, with four repetitions at ±3.5 ppm, to attain sufficient signal-to-noise ratio within the clinical time frame [5, 24]. Images were acquired with an RF saturation amplitude of 2 μT and total duration of 800 ms, with a four-block pulsed saturation scheme. Water-frequency shift due to field inhomogeneity was measured in a separate image acquired using the water-saturation shift referencing method with 21 offset frequencies ranging from -1.25 to 1.25 ppm, at intervals of 0.125 ppm (16 Hz), with one reference image acquired without a saturation RF pulse, resulting in a full Z-spectrum within the offset range [26]. The water-saturation shift reference image was acquired with a TR/TE of 1250/17 ms, RF saturation amplitude of 0.5 μT, and total duration of 400 ms, with a two-block pulsed saturation scheme. The total acquisition time for both APT and water-saturation shift reference images was 7 min 36 s.

APT Image Processing

After APT and water-saturation shift reference images were obtained, water frequency shift correction was performed. The full water-saturation shift Z-spectrum was fitted by a 12^th order polynomial on voxel-by-voxel basis for all offset frequencies. The fitted curve was then interpolated at a higher spectral resolution (1 Hz), where the lowest signal in the fit after interpolation was assumed to be an actual water resonance frequency. The water centre frequency offset was set as the measured displacement between the actual and ideal the water resonance frequency (0 Hz). The acquired APT-weighted data at the saturation frequency offsets (±3.0, ±3.5, ±4.0 ppm) were interpolated over the offset range and shifted using the estimated water centre frequency offset to obtain final shift-corrected APT-weighted image data [26]. Based on the shift-corrected data, magnetization transfer ratio asymmetry (MTR_asym) values at ± 3.5 ppm with respect to water frequency were calculated [4]:

$$\mathit{MT}{R}_{\mathit{asym}}\left(+3.5\mathit{ppm}\right)=\frac{S_{\mathit{sat}}\left(-3.5\mathit{ppm}\right)-S_{\mathit{sat}}\left(+3.5\mathit{ppm}\right)}{S_0}$$

where S₀ and S_sat (±3.5 ppm) are MRI signals without and with saturation RF pulse, respectively. APT-weighted images were then generated using MTR_asym at 3.5 ppm. All data processing was performed offline, using Matlab (MathWorks, Natick, MA).

Image Analysis

Circular regions of interest (ROIs) were drawn on APT-weighted images, which were independently evaluated by two neuroradiologists (with 8 and 3 years of experience in neuroradiology) blinded to clinical and histopathological data. First, three circular ROIs were manually placed in different slices on raw APT images which exhibited similar contrast with T2WI for anatomical landmarks in the brain. Each ROI was located in a place that best represented the entire tumour signal. We carefully attempted to exclude cystic or necrotic portions referring to both T2WI and post-contrast T1WI and included the area with the highest signal within tumours. Another ROI was placed in normal-appearing white matter (NAWM). Then, the ROIs were transferred to processed APT-weighted images (Fig 1). The measured signal intensities in the three ROIs within tumour on processed APT-weighted images were averaged to represent the tumour. Thus, mean MTR_asym values for solid enhancing tumours and NAWM were obtained in each patient. Finally, normalized MTR_asym (nMTR_asym) values were calculated as follows:

$$\mathit{nMT}{R}_{\mathit{asym}}=\mathit{MT}{R}_{\mathit{asym}}\left(\text{tumor}\right)-\mathit{MT}{R}_{\mathit{asym}}\left(\text{NAWM}\right)$$

where MTR_asym (tumour) is the mean MTR_asym of ROIs in enhancing tumours, and MTR_asym (NAWM) is the mean MTR_asym of ROIs in corresponding NAWM. The ROIs ranged in size from 518.4 to 691.2 mm². The mean MTR_asym values of the two readers were used for further analysis.

Figure 1.

A region of interest was placed on a raw amide proton transfer (APT) image (c) with equivalent image contrast as a T2-weighted image (a) and a contrast-enhanced T1-weighted image (b). It was then transferred to the processed APT-weighted image (d). By referring to conventional MR images, cystic or necrotic areas could be excluded from ROIs.

Conventional MR images were analysed for tumour location, maximum tumour diameter, T2 hyperintensity, peritumoural brain oedema, irregular tumour margin, heterogeneous enhancement, capsular enhancement, cystic changes, hyperostosis of adjacent skull, skull invasion, and the dural tail sign by the same two neuroradiologists. Tumour location was categorized as “skull base” or “non-skull base.” T2 hyperintensity was defined as higher signal intensity of the tumour relative to grey matter on T2WI. Irregular tumour margin was defined by the lobulated appearance of tumour margins on contrast-enhanced T1WI. The mean values of maximum tumour diameters measured by the two neuroradiologists were determined for further analysis. Discrepancies were settled by consensus between the two radiologists.

Surgery and Pathological Evaluation

All patients underwent surgical tumour resection after MRI. Pathological diagnosis was made by neuropathologists according to the WHO criteria. Criteria for atypical meningioma (WHO grade II) was 4 to 19 mitoses per 10 high-power fields (HPF), the presence of brain invasion, or the presence of at least three of these features (‘sheet-like’ growth, hypercellularity, spontaneous necrosis, large and prominent nucleoli, small cells), while criteria for malignant (anaplastic) meningioma (WHO grade III) was frank anaplasia (histology resembling carcinoma, sarcoma, or melanoma) or elevated mitoses with 20 or more mitoses per 10 HPF [27–29]. Additionally, the Ki-67 labelling index was evaluated in all patients. In 42 of 57 patients, mitotic count was evaluated using the mitotic marker phosphohistone-H3.

Statistical Analysis

Based on the results of normality testing by the Kolmogorov–Smirnov test, MTR_asym values were compared between enhancing tumours and NAWM by the paired t-test, and nMTR_asym values and maximum tumour diameter were compared between patients with benign and atypical meningiomas by Student’s t-test. Correlation between nMTR_asym and the Ki-67 labelling index or mitotic count was evaluated by Pearson’s correlation coefficient analysis.

Differences in frequencies of specific imaging features between benign and atypical meningiomas were determined by the chi-square or Fisher’s exact test. Then, nMTR_asym and conventional imaging features found to be significant in univariable analysis were further evaluated to compare the diagnostic performance by receiver operating characteristic (ROC) curve. To determine whether inclusion of nMTR_asym with conventional MRI features improved the discriminative value of MRI for benign and atypical meningiomas, two different models comprising conventional imaging features which were identified as being significantly associated with atypical meningioma based on univariable analysis with and without nMTR_asym (models 1 and 2, respectively) were assessed based on the ROC curve using c-statistics (Harrell’s concordance index) after multivariable logistic regression analyses. The net reclassification index (NRI) and integrated discrimination improvement (IDI) values of the two models were also calculated to evaluate the increase in discriminative value [30].

Interobserver agreement on nMTR_asym in APT images and maximum tumour size in conventional MR images was evaluated by intraclass correlation coefficients (ICC), while that on conventional imaging features was evaluated by Cohen’s κ coefficient; ICC and κ values > 0.75 indicated excellent agreement. Statistical analyses were performed using SAS (version 9.2; SAS Institute Ind., Cary, NC, USA), MedCalc (version 9.3.6.0; MedCalc Software, Mariakerke, Belgium), and R (version 3.3.1; R Foundation for Statistical Computing, Vienna, Austria). Statistical significance was set at P < 0.05.

Results

Among the 57 included patients, 13 were pathologically diagnosed with atypical meningioma and 44 with benign meningioma (transitional, 18; meningothelial, 15; fibrous, 8; angiomatous, 2; and secretory, 1). None of the patients exhibited malignant meningiomas. While 39 patients exhibited non-skull base tumours (cerebral convexity, 19; parasagittal area, 18; and lateral ventricle, 2), 18 exhibited skull base tumours (posterior fossa, 8; sphenoid ridge, 5; parasellar, 4; and anterior skull base, 1). No patient showed tumour recurrence until February 2017.

There was a significant difference in MTR_asym between enhancing tumours (1.84 ± 0.69%) and NAWM (0.10 ± 0.48%; P < 0.001). There was also a significant difference in nMTR_asym between patients with benign and atypical meningiomas (1.67% vs. 2.46%; P < 0.001; Table 1). Figures 2 and 3 present representative cases of benign and atypical meningiomas, respectively. Atypical meningiomas exhibited higher APT-weighted signal intensities than benign meningiomas.

Table 1.

Comparison of Demographic and Clinical Characteristics and Imaging Findings between Patients with Benign and Atypical Meningiomas

	Benign Meningioma (n = 44)	Atypical Meningioma (n = 13)	P value
Age (mean ± SD), years	53.0 ± 11.95	60.0 ± 13.26	0.130
Sex (male)	12 (27.3%)	4 (30.8%)	0.805
nMTRasym (mean ± SD)	1.67 ± 0.64	2.46 ± 0.48	< 0.001
Maximum tumour diameter (mean ± SD), mm	43.4 ± 14.7	56.4 ± 23.1	0.030
Non-skull base location	27 (61.4%)	12 (92.3%)	0.044
T2 hyperintensity	21 (47.7%)	6 (46.2%)	1.000
Peritumoural brain oedema	19 (43.2%)	9 (69.2%)	0.123
Irregular tumour margin	8 (18.2%)	4 (30.8%)	0.440
Heterogeneous enhancement	20 (45.5%)	10 (76.9%)	0.048
Capsular enhancement	8 (18.2%)	1 (7.7%)	0.668
Cystic change	2 (4.5%)	1 (7.7%)	0.547
Reactive hyperostosis	15 (34.1%)	4 (21.1%)	1.000
Skull invasion	10 (22.7%)	3 (23.1%)	1.000
Dural tail sign	29 (65.9%)	6 (46.2%)	0.215

Note—Unless otherwise indicated, data are presented as number of patients (%). P values were calculated using Student’s t-test for continuous variables and the chi-square or Fisher’s exact test for categorical variables. nMTR_asym = normalized magnetization transfer ratio asymmetry value.

Figure 2.

A 52-year-old female patient with benign meningioma exhibiting low signal intensity on a T2-weighted image (a) and heterogeneous enhancement at the centre on a contrast-enhanced T1-weighted image (b). A ROI was drawn on solid enhancing portion of the tumour on a raw APT image (c) and then transferred to the processed APT-weighted image (d). The normalized magnetization transfer ratio asymmetry value within the enhancing tumour was 1.69 (c).

Figure 3.

A 57-year-old female patient with atypical meningioma exhibiting intermediate signal intensity on a T2-weighted image (a) and homogeneous enhancement on a contrast-enhanced T1-weighted image (b). A ROI was drawn on area which showed the highest signal within the tumour on a raw APT image (c) and then transferred to the processed APT-weighted image (d). The normalized magnetization transfer ratio asymmetry value within the enhancing tumour was 3.12 (c).

Both Ki-67 labelling index (3.25 ± 2.67) and mitotic count (2.45 ± 2.70) were significantly and positively correlated with nMTR_asym (Pearson’s correlation coefficients, 0.440 and 0.426, respectively; P = 0.002 and 0.017, respectively; Fig 4).

Figure 4.

Correlation of normalized magnetization transfer ratio asymmetry value (nMTR_asym) with Ki-67 labelling (a) and mitosis (b).

The overall mean maximum tumour diameter was 46.4 ± 17.6 mm, with atypical meningiomas exhibiting significantly greater maximum tumour diameters than benign meningiomas (56.4 ± 23.1 vs. 43.4 ± 14.7 mm; P = 0.030). Heterogeneous enhancement and non-skull base location were observed significantly more often in atypical meningiomas than in benign meningiomas. There were no significant differences in the frequencies of other imaging features between the two tumours (Table 1).

ROC curve analysis with nMTR_asym, heterogeneous enhancement, maximum tumour diameter, and non-skull base location, the nMTR_asym showed significantly higher area under the curve (AUC) value than the others. The optimal cutoff value of nMTR_asym for predicting atypical meningiomas was >2.19, with sensitivity of 69.2%, and specificity of 84.1%. The values and 95% confidence intervals [CI] of the AUC of the evaluated parameters were as follows: nMTR_asym, 0.825 (0.701–0.949); heterogeneous enhancement, 0.657 (0.517–0.798); maximum tumour diameter, 0.671 (0.514–0.828), and non-skull base location, 0.655 (0.517–0.776; Fig 5).

Figure 5.

Receiver operating characteristic curve analysis. Normalized magnetization transfer ratio asymmetry value (nMTR_asym; area under the curve [AUC], 0.825; 95% confidence interval [CI], 0.701–0.949); heterogeneous enhancement (AUC, 0.657; 95% CI, 0.517–0.798); maximum tumour diameter (AUC, 0.671; 95% CI, 0.514–0.828); non-skull base location (AUC, 0.655; 95% CI, 0.517–0.776).

On multivariable logistic regression analysis of the four variables, nMTR_asym was an independent imaging biomarker for predicting atypical meningioma (odds ratio, 11.227; P = 0.014; Table 2). On comparison of the predictive power for atypical meningioma of the two multivariable models (model 1 with covariates as maximum tumour diameter, heterogeneous enhancement, and non-skull base location; model 2 with covariates as maximum tumour diameter, heterogeneous enhancement, non-skull base location, and nMTR_asym) using c-statistics analysis, model 2 exhibited a significantly better diagnostic performance than model 1 (AUC: 0.879 vs. 0.764; P = 0.039). Model 2 also exhibited better performance with NRI of 0.811 (95% CI, 0.272–1.350) and IDI of 0.172 (95% CI, 0.059–0.286) values (Table 2).

Table 2.

Performance Comparison of Multivariable Logistic Regression Models with and without nMTR_asym for Prediction of Atypical Meningioma

	Model 1			Model 2
Variables	Adjusted OR	95% CI	P value	Adjusted OR	95% CI	P value
Maximum tumour diameter	1.036	0.998–1.084	0.238	1.004	0.958–1.052	0.876
Heterogeneous enhancement	2.581	0.571–14.789	0.571–14.789	1.618	0.282–9.293	0.589
Non-skull base location	5.262	0.577–47.948	0.141	6.953	0.607–1.052	0.119
nMTRasym				11.227	1.632–77.241	0.014
AUC*	0.764 (95% CI, 0.633–0.866)			0.879 (95% CI, 0.772–0.954)
NRI	reference			0.811 (95% CI: 0.272–1.350)
IDI	reference			0.172 (95% CI: 0.059–0.286)

Model 1 = maximum tumour diameter + heterogeneous enhancement + non-skull base location

Model 2 = maximum tumour diameter + heterogeneous enhancement + non-skull base location + nMTR_asym

AUC = area under the receiver operating characteristic curve, CI = confidence interval, IDI = integrated discrimination index, nMTR_asym = normalized magnetization transfer ratio asymmetry value, NRI = net reclassification index, OR = odds ratio.

^*P = 0.039

Interobserver agreement on nMTR_asym and maximum tumour diameter was excellent (ICC values, 0.911 [95% CI, 0.840–0.959] and 0.989 [95% CI, 0.982–0.994], respectively). Interobserver agreement on specific conventional imaging features was also excellent, with κ values of 0.755 to 1.000 for all parameters.

Discussion

Although surgical resection is the main treatment for meningiomas, monitoring or gamma-knife surgery is considered in patients with asymptomatic small meningiomas without pathological diagnosis [31]. Therefore, it is important to identify atypical meningiomas preoperatively. In the present study, we investigated the ability of APT imaging to differentiate between benign and atypical meningiomas. Although preliminary studies have employed APT imaging for evaluation of brain tumours [6, 32], to the best of our knowledge, the present results are the first to describe the feasibility of APT imaging for grading meningiomas. The MTR_asym values of both tumours and normal brain tissue are affected by various tissue microenvironment and technical factors. Therefore, we evaluated the diagnostic value of APT imaging based on nMTR_asym—the difference in MTR_asym between tumours and normal brain tissue—because it is minimally affected by non-physiologic factors. Atypical meningiomas exhibited significantly higher nMTR_asym than benign meningiomas (2.46% vs. 1.67%), which is in accordance with results of previous studies on APT imaging for grading gliomas that also reported significant increases in signal intensity according to tumour grade [9, 24].

Among the conventional MR features evaluated in the present study, greater maximum tumour diameter, heterogeneous enhancement, and non-skull base location were associated with atypical meningiomas. A recent study also reported the association between tumour volume and histopathological grade of meningioma, which may be attributed to the relatively high proliferative potential of high-grade tumours [18]. Heterogeneous enhancement is associated with heterogeneous distribution of tumour cells, which reflects intratumoural ischemic necrosis, calcification, haemorrhage, and/or cystic changes [18–20]. We also found that non-skull base meningiomas were more likely to be atypical meningiomas, which is in line with previous findings [21–23]. Furthermore, previous studies have reported the association of anatomical location of meningiomas with not only WHO classification but also higher MIB-1 and Ki-67 labelling indexes [22, 33].

In the present study, however, APT imaging exhibited a better diagnostic performance than conventional MRI in diagnosis of atypical meningiomas. The results of multivariable regression analysis with conventional MRI features and nMTR_asym as covariates revealed only high nMTR_asym as being a statistically significant predictor of atypical meningiomas. Inclusion of APT imaging parameters with conventional imaging features improved the discriminative value of MRI for benign and atypical meningiomas. Considering that it is relatively difficult to demonstrate a significant difference in AUC analysis of performance models despite an added biomarker being a clinically significant factor, our results indicate the superior discriminative ability of APT imaging for benign and atypical meningiomas. The diagnostic improvement in MRI upon inclusion of APT imaging was also validated by NRI and IDI, which exhibited values greater than zero.

Recent studies on gliomas have demonstrated the positive correlation of APT-weighted signal intensity with Ki-67 labelling and cell proliferation index [9, 10]. In the present study, we also found that APT-weighted signal intensity is positively correlated to Ki-67 labelling index and mitotic count. These results might reflect that high-grade tumours, which exhibit higher proliferative and mitotic activities, have higher densities of cells and, hence, higher concentrations of intracellular proteins and peptides than low-grade tumours.

There are several limitations to the present study. First, there are several previous studies attempting to differentiate typical and atypical/malignant meningiomas by means of diffusion-weighted imaging, which have reported contradictory results [17, 18, 34]. In addition, there are previous studies investigating perfusion MRI in meningioma patients, which reported perfusion MRI findings were correlated with grade of meningioma or Ki-67 labelling index in meningioma [35, 36]. Further studies are required to compare the diagnostic performances of diffusion-weighted imaging, perfusion MRI, and APT imaging in differentiating benign and atypical meningiomas. Second, instead of segmenting entire tumours, ROIs were placed manually in the solid portions of tumours which we thought represented tumour characteristics the most. We chose this method because meningiomas occur near the skull, which is susceptible to field inhomogeneity and consequently may affect APT signals. In addition, the resolution of APT imaging must be improved to achieve adequate coregistration with conventional MRI images and segmentation of the entire tumour. Third, because pathological specimens lose location information during surgical resection, there might have been mismatches between ROIs on images and pathological specimens from resection. Lastly, there were various subtypes of benign meningioma in this study, which might require further investigation to clarify relation between each subgroup and diagnostic performance. On the other hand, there was limited number of atypical meningioma patients and lack of malignant meningioma patient. However, considering the relative rarity of these types of tumours, this result was inevitable.

In summary, atypical meningiomas exhibited significantly higher APT-weighted signal intensities than benign meningiomas. Therefore, APT imaging is a useful imaging biomarker that adds value to conventional MRI for differentiation of atypical and benign meningiomas.

Key points.

1. APT imaging is useful for differentiating between atypical and benign meningiomas.

2. Atypical meningiomas exhibited high APT-weighted signal intensity than benign meningiomas.

3. The diagnostic performance of MRI improved with nMTR_asym for predicting atypical meningiomas.

Abbreviations and acronyms

MRI

Magnetic resonance imaging

CEST

Chemical exchange saturation transfer

APT

Amide proton transfer

MTR_asym

Magnetization transfer ratio asymmetry

nMTR_asym

Normalized magnetization transfer ratio asymmetry

ROC

Receiver operating characteristic

AUC

Area under the curve

NRI

Net reclassification index

IDI

Integrated discrimination improvement

ICC

Intraclass correlation coefficients