Delta Histogram Analysis of Synthetic MRI for Evaluating Grade and Cellular Proliferation of Meningioma

Abstract

Purpose

To investigate the potential of delta histogram analysis of synthetic MRI (SyMRI) for predicting meningioma grade and cellular proliferation.

Materials and Methods

In this prospective study from February 2022 to October 2023, consecutive participants scheduled for conventional MRI and SyMRI were enrolled. Histogram parameters (HPs) were obtained with whole-tumor volume analysis. Delta histogram parameters, including absolute (dα = postcontrast HP – precontrast HP) and relative delta HPs (dα% = [postcontrast HP – precontrast HP]/precontrast HP), where α = T1, T2, or proton density (PD), were calculated. The Mann-Whitney U test was used to compare HPs between high- and low-grade groups. Receiver operating characteristic curve and multiple logistic regression analysis, the DeLong test, and integrated discrimination improvement were used to evaluate diagnostic efficacy of the HPs and models. Spearman rank correlation coefficients analyzed the correlations between HPs and Ki-67 expression.

Results

Ninety-five participants with meningioma (mean age, 58 years ± 13 [SD]; 28 male) were enrolled in the study. Participants with high-grade meningiomas had greater PD skewness; postcontrast T1 energy and total energy; postcontrast T2 IQR, mean absolute deviation (MAD), and robust MAD (rMAD); postcontrast PD skewness; dPD variance, IQR, MAD, and rMAD; and dPD% variance, IQR, MAD, and rMAD and lower PD median and 90th percentile (P < .05). These single SyMRI HPs demonstrated moderate diagnostic accuracy in grading meningioma (area under the receiver operating characteristic curve [AUC], range: 0.65–0.79). Among the combined models, the SyMRI model achieved the best performance (AUC = 0.88), outperforming single models (P < .05). Multiple SyMRI HPs were correlated with the Ki-67 index (P < .05).

Conclusion

Delta HPs from SyMRI demonstrate potential as biomarkers for meningioma grading and proliferation assessment.

Keywords: CNS, Brain/Brain Stem, Meningioma, Synthetic MRI, Delta Histogram Analysis, Neoplasm Grading, Ki-67 Antigen

Supplemental material is available for this article.

© RSNA, 2026

Summary

Delta histogram analysis of synthetic MRI data can noninvasively, quantitatively, and simultaneously grade meningioma and assess its proliferative activity; combined multiparametric models demonstrated optimal diagnostic performance.

Key Points

• ■ In this prospective study, histogram analysis revealed differences in synthetic MRI (SyMRI) parameters between low- and high-grade meningiomas (P < .05); the relative delta proton density mean absolute deviation performed best in tumor grading (area under the receiver operating characteristic curve [AUC] = 0.79).

• ■ The precontrast/postcontrast/delta feature multiparametric SyMRI model performed best in grading meningioma (AUC = 0.88, sensitivity = 100%).

• ■ Histogram analysis revealed correlations between multiple quantitative SyMRI parameters and the Ki-67 index (all P < .05), suggesting its potential for evaluating cellular proliferation.

Introduction

Meningiomas are the most common primary intracranial tumors and constitute 41.7% of all central nervous system tumors (1). The World Health Organization (WHO) classifies meningiomas into three pathologic grades (grades 1–3) (2), with distinct biologic behaviors and prognoses (3). Low-grade meningiomas (LGMs, WHO grade 1) are slow growing with favorable outcomes after resection, whereas high-grade meningiomas (HGMs, WHO grades 2 and 3) are aggressive with high mortality and recurrence rates (4). Standard treatment includes surgical resection, while adjuvant radiation therapy is added for high-grade tumors to reduce recurrence. Ki-67 is a proliferation marker associated with tumor progression and recurrence (5). Preoperative evaluation of meningioma grade and proliferative activity is essential for therapeutic strategies and predicting outcomes.

Conventional MRI is widely used for characterizing brain tumors but is qualitative and limited in grading and proliferation analysis. Quantitative MRI techniques, such as T1, T2, and proton density (PD) mapping, enable noninvasive tissue biologic evaluation (6). A previous study showed T1 mapping predicts meningioma grade and proliferation (7). However, conventional mapping requires separate sequences, prolonging scan time and causing misregistration (8).

Synthetic MRI (SyMRI) enables simultaneous multiparametric quantification in a single scan, substantially reducing the acquisition time compared with conventional methods (9). Its parameters reflect intrinsic tissue properties: T1 reflects water binding to macromolecules and paramagnetic ions, T2 reflects water molecular mobility and tissue microstructural features (cellular density, fibrosis), and PD closely reflects tissue water content (10–12). Recent studies have demonstrated the potential value of SyMRI for brain tumor grading and differentiation (7,9). High-grade tumors are characterized by increased vascularity and heterogeneity; thus, their postcontrast relaxation changes may be more obvious than those of low-grade tumors, suggesting postcontrast SyMRI could reveal changes in relaxation characteristics that are not detectable at precontrast imaging. However, the use of postcontrast SyMRI for meningiomas remains limited. The delta histogram method quantitatively analyzes contrast-enhanced radiologic feature changes (13). A recent study showed T1 delta histogram parameters (HPs) can predict meningioma grading and proliferation (7). However, to our knowledge, no studies have integrated SyMRI with delta histogram analysis for evaluating tumors.

As two quantitative imaging techniques, SyMRI measures tumor relaxation characteristics, while delta histogram analysis assesses tumor characteristic enhancement-related changes. Therefore, we hypothesized that combining these methods allows improved evaluations of tumor microstructures over the individual techniques. To this end, this study aimed to explore the utility of SyMRI-based delta histogram analysis in predicting meningioma grade and its association with Ki-67.

Materials and Methods

Study Design and Sample

This prospective study was approved by the institutional review board of Fujian Medical University Union Hospital (approval no. 2021KJCX055). All participants provided written informed consent. One author (J.Z.) is affiliated with GE HealthCare and contributed as an MRI research scientist, providing technical assistance with SyMRI sequence debugging under GE collaborative regulations, without financial or personal conflicts of interest.

From February 2022 to October 2023, consecutive individuals with suspected meningiomas were enrolled and underwent preoperative MRI. The inclusion criteria were as follows: (a) underwent both structural MRI and SyMRI before surgery, (b) underwent surgical resection performed at our hospital, and (c) had surgical pathology for confirming the diagnosis of meningioma on the basis of the 2021 WHO classification of central nervous system tumors (2). The exclusion criteria were the following: (a) underwent relevant treatments, such as intracranial radiation therapy, chemotherapy, or neurosurgery and (b) had MR images of insufficient quality.

Image Acquisition

MRI scans were performed with a 3.0-T MRI scanner (Signa Premier; GE HealthCare) with a 48-channel head coil.

The MRI protocol included axial T1-weighted fluid-attenuated inversion recovery (FLAIR), T2-weighted fast spin-echo, T2-weighted FLAIR, diffusion-weighted imaging, precontrast SyMRI (MAGnetic resonance image Compilation), postcontrast three-dimensional T1-weighted fast spoiled gradient-echo imaging, and postcontrast SyMRI. The contrast agent gadobutrol (Gadavist; Bayer Healthcare Pharmaceuticals) was administered intravenously at a dose of 0.1 mmol per kilogram of body weight and a flow rate of 1.2 mL/sec for postcontrast imaging. Postcontrast SyMRI was performed 5 minutes after contrast agent injection. The total scan protocol lasted 15 minutes. Detailed imaging parameters are provided in Appendix S1.

MR Image Processing and Histogram Analysis

The raw SyMRI data were processed using SyMRI version 8.0 software (SyntheticMR) to generate quantitative maps (T1, T2, PD, postcontrast T1 [T1C], postcontrast T2 [T2C], and postcontrast PD [PDC]). Rigid registration between these maps and postcontrast three-dimensional T1-weighted images was performed using the SPM12 toolbox (Wellcome Center for Human Neuroimaging, https://www.fil.ion.ucl.ac.uk/spm/) implemented in MATLAB (MathWorks). Two neuroradiologists (L.L. and X.C., with 15 and 8 years of experience, respectively) blinded to the pathology findings independently segmented the volumes of interest (VOIs) using the ITK-SNAP program (version 4.6.1; University of Pennsylvania, http://www.itksnap.org). With reference to structural anatomic MRI, the VOIs were delineated on axial three-dimensional contrast-enhanced T1-weighted images over solid tumor portions, avoiding cystic, necrotic, calcific, and hemorrhagic areas, and then projected onto the quantitative SyMRI maps. HPs were extracted from these maps using FeAture Explorer (version 0.4.1; https://github.com/salan668/FAE), including the mean, median, 10th percentile, 90th percentile, minimum, maximum, kurtosis, skewness, uniformity, variance, range, IQR, energy, total energy, entropy, root mean square, mean absolute deviation (MAD), and robust MAD (rMAD). Delta histogram analysis quantified relaxometry parameter distribution changes after contrast agent administration. For coregistered pre- and postcontrast SyMRI volumes, HP α (T1, T2, or PD) was initially extracted from manually delineated VOIs. Two types of delta HPs were then calculated: (a) absolute delta HPs (dα = postcontrast HP – precontrast HP) and (b) relative delta HPs (dα% = [postcontrast HP – precontrast HP]/precontrast HP).

Pathologic Analysis

Ki-67 expression was estimated via immunohistochemical staining. A neuropathologist (Hu Chen, PhD, with 10 years of experience) calculated the Ki-67 labeling index as the percentage of positive tumor nuclei among ≥1000 nuclei counted across five high-power fields (×400) in the “hottest region” (highest positivity density) (7,14,15). Any uncertainties in the assessment were resolved through consensus with a second neuropathologist (Meihong Yao, PhD, with 15 years of experience).

Genomic DNA was extracted from paraffin-embedded meningioma tissue, after which targeted next-generation sequencing and bioinformatic analysis were performed to detect single-nucleotide variants, insertions, and deletions, copy number variations, and diagnostic molecular markers (16,17).

Statistical Analysis

Statistical analyses were performed using SPSS version 27.0 (IBM), MedCalc software (version 15.2.2), and R software (version 4.2.2).

Shapiro-Wilk test was used to assess normality of continuous variables. Continuous variables with a normal distribution are presented as means ± SDs; otherwise, medians with IQRs are presented. Categorical variables are described using frequencies and percentages. Interobserver reliability of HPs was evaluated with intraclass correlation coefficient and qualified as follows: poor (0.00–0.20), fair (0.21–0.40), moderate (0.41–0.60), good (0.61–0.80), and excellent (0.81–1.00). Comparisons of the clinical data, molecular characteristics, MRI features, and SyMRI HPs were compared between HGMs and LGMs using the Mann-Whitney U test (continuous variables) and the Fisher exact test (categorical variables).

To increase diagnostic performance, multivariable logistic regression analysis with forward stepwise selection was used to construct models with different combinations of features (Appendix S2). Receiver operating characteristic (ROC) curves were used to analyze the diagnostic efficacy of the HPs and combined models, with optimal cutoff values for distinguishing HGMs from LGMs. The DeLong test was used to compare the areas under the ROC curve (AUCs) of each model and to calculate the 95% CIs for diagnostic parameters. The integrated discrimination improvement (IDI) was used to further compare the diagnostic performance of the models. The SyMRI model robustness was evaluated using least absolute shrinkage and selection operator regression with 10-fold cross-validation (Appendix S3).

Spearman rank correlation coefficients (ρ) were calculated to assess the correlations between SyMRI HPs and Ki-67 expression. Two-tailed P < .05 indicated statistical significance.

Results

Participant Characteristics

The participant flowchart is shown in Figure 1. Initially, 113 participants with histopathologically confirmed meningioma were included; of these, 18 participants were excluded because of severe motion artifacts (n = 5) and not having yet undergone intracranial radiation therapy (n = 7) or neurosurgery (n = 6). Finally, 95 participants (mean age, 58 years ± 13 [SD]; 28 male) were enrolled and categorized into the low-grade group (LGMs, 75 participants; mean age, 57 years ± 11; 17 male) or the high-grade group (HGMs, 20 participants; 52 years ± 11; 11 male). Among the 95 participants included in this study, 75 were diagnosed with LGMs, including 23 meningothelial meningiomas, 21 fibrous meningiomas, 27 transitional meningiomas, two angiomatous meningiomas, one lymphoplasmacyte-rich meningioma, and one metaplastic meningioma; the remaining 20 were diagnosed with HGMs, including 18 atypical meningiomas, one chordoid meningioma, and one anaplastic meningioma. The detailed clinical characteristics and conventional MRI features are presented in Table 1. Differences were observed between LGMs and HGMs in terms of age (P = .04), sex (P = .01), and Ki-67 index (P = .009), whereas neither tumor volume (P = .58) nor tumor location (P = .20) showed evidence of a difference between the two groups. In terms of conventional MRI features, compared with those with LGMs, participants with HGMs were more likely to present with peritumoral edema (11 of 20 [55%] vs 19 of 75 [25%], P < .001) and bone invasion (seven of 20 [35%] vs 20 of 75 [27%], P = .02). No evidence of differences was observed in the other conventional MRI features. The genetic and molecular markers of the participants are summarized in Table S1. Molecular pathologic data were available for 93 participants. Differences were observed between participants with LGMs and those with HGMs regarding chromosome 1p loss (P < .001) and chromosome 10 loss (P = .006). Conversely, no evidence of differences was found in terms of NF2 mutation, CDKN2A/2B loss, TERT mutation, BAP1 mutation, PIK3CA mutation, ARID1A mutation, chromosome 22q loss, or chromosome 14q loss.

Figure 1:

Participant enrollment flowchart. SyMRI = synthetic MRI.

Table 1:

Clinical Characteristics of Participants with LGMs and HGMs

Characteristic	LGM (n = 75)	HGM (n = 20)	P Value
Age (y)	58 (49–67)	50 (47.00–55.75)	.04*
Male sex	17 (22.67)	11 (55.00)	.01*
Tumor volume (mm3)	21.63 (10.8–60.0)	36.00 (8.40–91.08)	.57
Ki-67 (%)	2 (1–3)	4 (1.25–7.25)	.009†
Tumor location			.20
Falx cerebri	17/75 (23)	4/20 (20)
Convexity	19/75 (25)	9/20 (45)
Other locations	39/75 (52)	7/20 (35)
Conventional MRI feature
Heterogeneous enhancement	45 (34/75)	50 (10/20)	.80
Indistinct margins	47 (35/75)	50 (10/20)	.81
CSF cleft loss	19 (14/75)	60 (12/20)	.58
Peritumoral edema	25 (19/75)	55 (11/20)	<.001‡
Bone invasion	27 (20/75)	35 (7/20)	.02*
Venous invasion	36 (27/75)	20 (4/20)	.28

Note.—Data are expressed as medians, with quartiles (Q1, Q3) in parentheses, or numbers or numerators and denominators, with percentages in parentheses. Statistical testing (P) was performed using the Mann-Whitney U test for comparisons of age, tumor volume, and Ki-67 expression and the Fisher exact test for comparisons of sex, tumor location, and conventional MRI features. CSF = cerebrospinal fluid, HGM = high-grade meningioma, LGM = low-grade meningioma.

^*P < .05.

^† P < .01.

^‡ P < .001.

Comparisons of HPs between Low- and High-Grade Meningiomas

Good interobserver agreement was obtained for the HPs, with intraclass correlation coefficients ranging from 0.98 to 1.00. The SyMRI HPs that differed between HGMs and LGMs are listed in Table 2, while the complete set of parameters is provided in Table S2. Participants with HGMs showed greater T1C energy, T1C total energy, T2C IQR, T2C MAD, T2C rMAD, PD skewness, PDC skewness, dPD variance, dPD IQR, dPD MAD, dPD rMAD and dPD% variance, dPD% IQR, dPD% MAD, and dPD% rMAD than LGMs did (P < .05). Conversely, the median and 90th percentile of the PD of participants with HGMs were lower than those of participants with LGMs (P < .05). Representative parametric maps of participants with LGMs and HGMs are shown in Figure 2.

Table 2:

Significantly Different Synthetic MRI Histogram Parameters between LGMs and HGMs

Parameter	LGM (n = 75)	HGM (n = 20)	Z Value	P Value
T1C
Energy (×106)	2231.70 (782.23–5699.09)	6074.76 (1369.59–13302.20)	−1.981	.048*
Total energy (×106)	1961.45 (687.51–5008.97)	5339.14 (1203.74–11691.38)	−1.981	.048*
T2C
IQR	8.90 (7.30–11.9)	11.90 (7.98–15.55)	−2.159	.03*
MAD	6.71 (5.17–9.29)	8.74 (5.76–11.66)	−2.118	.03*
rMAD	3.82 (3.13–5.12)	5.15 (3.37–6.66)	−2.273	.02*
PDC skewness	−0.42 (−1.12 to 0.09)	0.11 (−0.47 to 0.49)	−2.739	.006*
PD
Median	86.60 (84.30–89.30)	84.70 (82.50–86.27)	−2.273	.02*
90th percentile	93.00 (91.10–96.50)	90.60 (88.95–94.33)	−2.168	.03*
Skewness	−1.09 (−1.59 to −0.33)	−0.32 (−0.82 to 0.21)	−3.423	.001†
dPD
Variance	7.58 (−5.5 to 26.94)	21.20 (3.02–41.48)	−2.027	.04*
IQR	1.15 (0.10–2.10)	2.55 (0.63–4.23)	−2.255	.02*
MAD	1.68 (0.20–3.10)	3.57 (2.64–6.46)	−3.715	<.001‡
rMAD	0.41 (0.04–0.92)	1.09 (0.26–1.75)	−2.273	.02*
dPD%
Variance	0.18 (−0.09 to 0.59)	0.65 (0.10–1.09)	−2.428	.02*
IQR	0.16 (0.01–0.32)	0.36 (0.07–0.78)	−2.446	.01*
MAD	0.36 (0.05–0.70)	0.96 (0.57–1.64)	−3.898	<.001‡
rMAD	0.15 (0.01–0.30)	0.34 (0.10–0.73)	−2.373	.02*

Note.—Data are expressed as medians, with quartiles (Q1, Q3) in parentheses. dPD = absolute delta PD histogram parameter, dPD% = relative delta PD histogram parameter, HGM = high-grade meningioma, LGM = low-grade meningioma, MAD = mean absolute deviation, PD = proton density, PDC = postcontrast PD, T1C = postcontrast T1, T2C = postcontrast T2, rMAD = robust MAD.

^*P < .05.

^† P < .01.

^‡ P < .001.

Figure 2:

Representative synthetic MR (SyMR) images and histograms from a low-grade meningioma (LGM, World Health Organization [WHO] grade 1, fibroblastic meningioma) in a 54-year-old female participant (blue) and a high-grade meningioma (HGM, WHO grade 2, atypical meningioma) in a 44-year-old male participant (pink). SyMR images were acquired in the axial plane. (A, B) T1 maps, (C, D) T2 maps, (E, F) proton density (PD) maps, (G, H) postcontrast T1 (T1C) maps, (I, J) postcontrast T2 (T2C) maps, and (K, L) postcontrast PD (PDC) maps. Comparisons of the histograms of different quantitative maps of the two participants are on the right side of the maps (LGM with a blue border and HGM with a pink border).

Evaluation of the Diagnostic Performance of HPs for Differentiating High- and Low-Grade Meningiomas

Table 3 shows representative ROC curve analysis results of the HPs that differed between the groups, while the complete results are listed in Table S3. The pre- and postcontrast and absolute and relative delta SyMRI HPs of the tumors differentiated HGMs and LGMs, with AUCs ranging from 0.66 to 0.75, 0.65 to 0.70, 0.65 to 0.77, and 0.67 to 0.79, respectively. Among these parameters, dPD% MAD showed the best diagnostic performance in differentiating high- and low-grade meningiomas, with an AUC of 0.79 (95% CI: 0.69, 0.86), an accuracy of 71% (67 of 95), a sensitivity of 85% (17 of 20), and a specificity of 67% (50 of 75), with a cutoff value of 0.50 for grading meningiomas.

Table 3:

Diagnostic Performance of Representative Synthetic MRI Histogram Parameters for Differentiating Low-Grade and High-Grade Meningiomas

Parameter	AUC	Cutoff	Sens (%)	Spec (%)	PPV (%)	NPV (%)	Acc (%)
PD skewness	0.75 [0.65, 0.83]	−0.93	85 (17/20) [62, 97]	60 (45/75) [48, 71]	36 (17/47) [29, 44]	94 (45/48) [84, 98]	65 (62/95) [55, 74]
T1C
Energy	0.65 [0.54, 0.74]	7 068 606 395	50 (10/20) [27, 73]	80 (60/75) [69, 88]	40 (10/25) [26, 56]	86 (60/70) [79, 90]	74 (70/95) [65, 82]
Total energy	0.65 [0.54, 0.74]	6 212 642 339	50 (10/20) [27, 73]	80 (60/75) [69, 88]	40 (10/25) [26, 56]	86 (60/70) [79, 90]	74 (70/95) [65, 82]
T2C rMAD	0.67 [0.56, 0.76)	4.70	65 (13/20) [41, 85]	68 (51/75) [56, 78]	35 (13/37) [26, 46]	88 (51/58) [80, 93]	67 (64/95) [58, 76]
PDC skewness	0.70 [0.60, 0.79]	0.05	65 (13/20) [41, 85]	71 (53/75) [59, 81]	37 (13/35) [27, 49]	88 (53/60) [80, 93]	69 (66/95) [60, 78]
dPD MAD	0.77 [0.67, 0.85]	2.58	80 (16/20) [56, 94]	71 (53/75) [59, 81]	42 (16/38) [33, 52]	93 (53/57) [85, 97]	73 (69/95) [63, 81]
dPD% MAD	0.79 [0.69, 0.86]	0.50	85 (17/20) [62, 97]	67 (50/75) [55, 77]	40 (17/42) [32, 50]	94 (50/53) [85, 98]	71 (67/95) [61, 79]

Note.—Data in brackets are 95% CIs. Data in parentheses are the numerator and denominator of the obtained parameters. Acc = accuracy, AUC = area under the receiver operating characteristic curve, dHP = absolute delta PD histogram parameters, dHP% = relative delta PD histogram parameters, MAD = mean absolute deviation, NPV = negative predictive value, PD = proton density, PDC = postcontrast PD, PPV = positive predictive value, rMAD = robust mean absolute deviation, Sens = sensitivity, Spec = specificity, T1C = postcontrast T1, T2C = postcontrast T2.

Performance Evaluation of Different Diagnostic Models

Among the combined models, only PD skewness was selected to construct the precontrast SyMRI model, which yielded an AUC of 0.75 (95% CI: 0.65, 0.88) for predicting the tumor grade. Three histogram features—T1C total energy, T2C MAD, and PDC skewness—were selected to construct the postcontrast SyMRI model, which achieved an AUC of 0.81 (95% CI: 0.71, 0.88). The absolute delta SyMRI model was constructed using dPD MAD and dPD rMAD, yielding an AUC of 0.78 (95% CI: 0.69, 0.86). For the relative delta SyMRI model, two histogram features, dPD% IQR and dPD% MAD, were selected, and the AUC of the resulting model was 0.79 (95% CI: 0.69, 0.87).

For the combined SyMRI model integrating pre- and postcontrast and absolute and relative delta SyMRI parameters, three histogram features, including PD skewness, dPD% IQR, and dPD% MAD, were incorporated. Compared with the other models, the combined SyMRI model achieved the best diagnostic performance, with an AUC of 0.88 (95% CI: 0.79, 0.93), an accuracy of 68% (65 of 95), a sensitivity of 100% (20 of 20), and a specificity of 60% (45 of 75). The ROC curve analysis results of these models are shown in Table 4.

Table 4:

Diagnostic Performance of Different Diagnostic Models for Differentiating Low-Grade and High-Grade Meningiomas

Model	AUC	Cutoff	Sens (%)	Spec (%)	PPV (%)	NPV (%)	Acc (%)
Precontrast model	0.75 [0.65, 0.88]	0.16	85 (17/20) [62, 97]	60 (45/75) [48, 71]	36 (17/47) [30, 44]	93 (45/48) [84, 97]	65 (62/95) [55, 74]
Postcontrast model	0.81 [0.71, 0.88]	0.21	80 (16/20) [56, 94]	79 (59/75) [68, 87]	50 (16/32) [38, 62]	94 (59/63) [86, 97]	79 (75/95) [70, 86]
Absolute delta model	0.78 [0.69, 0.86]	0.19	80 (16/20) [56, 94]	73 (55/75) [62, 83]	44 (16/36) [34, 55]	93 (55/59) [85, 97]	75 (71/95) [65, 83]
Relative delta model	0.79 [0.69, 0.87]	0.38	55 (11/20) [32, 77]	95 (71/75) [87, 99]	73 (11/15) [55, 89]	89 (71/80) [83, 93]	86 (82/95) [79, 92]
SyMRI model	0.88 [0.79, 0.93]	0.09	100 (20/20) [83, 100]	60 (45/75) [48, 71]	40 (20/50) [34, 47]	100 (45/45) [94, 100]	68 (65/95) [59, 77]

Note.—Data in brackets are 95% CIs. Data in parentheses are the numerator and denominator of the obtained parameters. Acc = accuracy, AUC = area under the receiver operating characteristic curve, NPV = negative predictive value, PPV = positive predictive value, Sens = sensitivity, Spec = specificity, SyMRI = synthetic MRI.

Although the diagnostic performance of the combined SyMRI model was better only to that of the precontrast SyMRI models as assessed with the DeLong test (P < .03), the IDI results demonstrated that the combined SyMRI model improved upon the predictive capacities of all four other models (P < .05). The results of the IDI analysis and the DeLong test are shown in Table S4.

Robustness Validation of the SyMRI Model

Least absolute shrinkage and selection operator regression analysis yielded a model with high concordance with the primary logistic regression model (AUC = 0.88 [95% CI: 0.80, 0.95]) and identified the same three key SyMRI features (PD skewness, dPD% IQR, and dPD% MAD) with consistent positive coefficients, thereby supporting the robustness of the main findings (Appendix S3; Tables S6, S7; Figs S1, S2).

Correlation of HPs with the Ki-67 Labeling Index

Ki-67 expression data were available for 80 participants. The Spearman correlation coefficients between the HPs and the Ki-67 index that were considered statistically significant are presented in Table 5, while the complete results of the analysis are listed in Table S5. Correlations were found between the Ki-67 index and multiple SyMRI HPs, including T1 energy, T1 total energy, T2 energy, T2 total energy, PD energy, PD skewness, PD total energy, T2C energy, T2C total energy, PDC skewness, PDC energy, PDC total energy, dPD energy, dPD total energy, and dPD% MAD (r = 0.233–0.295, P < .05), whereas PD variance, PD IQR, PD MAD, PD rMAD, dT1 energy, dT1 total energy, dT2 energy, dT2 total energy, and dT1% minimum (r = −0.250 to −0.223, P < .05) were negatively correlated with the Ki-67 labeling index.

Table 5:

Histogram Parameters with Significant Spearman Rank Correlations with the Ki-67 Labeling Index

Parameter	Correlation Coefficient (ρ)	P Value
T1
Energy	0.235	.04*
Total energy	0.235	.04*
T2
Energy	0.250	.03*
Total energy	0.250	.03*
PD
Skewness	0.295	.008†
Variance	−0.227	.04*
IQR	−0.223	.046*
Energy	0.251	.03*
Total energy	0.251	.03*
MAD	−0.233	.04*
rMAD	−0.223	.046*
T2C
Energy	0.249	.03*
Total energy	0.249	.03*
PDC
Skewness	0.239	.03*
Energy	0.251	.03*
Total energy	0.251	.03*
dT1
Energy	−0.250	.03*
Total energy	−0.250	.03*
dT2
Energy	−0.233	.04*
Total energy	−0.233	.04*
dPD
Energy	0.260	.02*
Total energy	0.260	.02*
dT1% minimum	−0.250	.03*
dPD% MAD	0.233	.04*

Note.—dPD = absolute delta PD histogram parameters, dPD% = relative delta PD histogram parameters, dT1 = absolute T1 delta histogram parameters, dT1% = relative delta T1 histogram parameters, dT2 = absolute delta T2 histogram parameters, MAD = mean absolute deviation, PD = proton density, PDC = postcontrast PD, rMAD = robust MAD, T2C = postcontrast T2.

^*P < .05.

^† P < .01.

Discussion

The preoperative prediction of the histologic characteristics and biologic behaviors of meningiomas is clinically important. This study demonstrated the utility of delta histogram analysis as a reliable noninvasive method for evaluating the grade and cellular proliferation of meningioma at SyMRI. Notably, the diagnostic efficiency of the dPD% MAD was superior to that of the other single HPs (AUC = 0.79). Moreover, the combined SyMRI model demonstrated greater diagnostic performance than the individual parameter alone (AUC = 0.88). HPs obtained from SyMRI correlated with the proliferative activity of meningioma (all P < .05). This approach offers distinct advantages over conventional multiparametric protocols, including reduced scan times and improved diagnostic performance.

MR relaxometry enables the assessment of tumor tissue composition. However, previous quantitative studies mainly focused on native tumor relaxation times (18). In contrast, we obtained quantitative T1, T2, and PD values before and after contrast agent injection and calculated dα and dα% changes to quantify tissue alteration relaxation. Compared with LGMs, HGMs presented significantly greater T1C (energy, total energy) and T2C (IQR, MAD, rMAD) values. Tumor microcirculation variability determines the difference in relaxation times before and after contrast agent injection. HGMs exhibit a hyperpermeable, disorganized angiogenic network that accelerates contrast agent accumulation, shortens T1, and markedly increases contrast enhancement (10,11). PD, meanwhile, reflects the tissue water content (12). The median and 90th percentile of PD were significantly lower in HGMs, consistent with findings in other high-grade malignancies (19–21). Rapid proliferation and elevated cellular density in HGMs reduce extracellular space and restrict water mobility (11,22), causing a leftward histogram shift. Moreover, HGMs had significantly greater PD (skewness), PDC (skewness), dPD, and dPD% (variance, IQR, MAD, and rMAD) than LGMs. The increases in these parameters may be explained by their high sensitivity to edema and structural damage within HGMs (12,23,24). Notably, PD-related parameters accounted for the majority of the significant features.

To our knowledge, no prior studies have evaluated the utility of HPs derived from pre- and postcontrast SyMRI for meningioma grading. The dα retains physical units and highlights clinically relevant small increments. The dα%, meanwhile, normalizes increments to individual baselines, minimizing interpatient variability and hardware-related bias. Our results demonstrate that the dPD% MAD achieved the highest diagnostic performance in differentiating HGMs and LGMs, outperforming all the other individual HPs. This finding regarding the relative delta HPs of PD is noteworthy, as this area remains largely unexplored. PD provides information on the water content in tissue, making related parameters sensitive to edema and structural damage (24). High microvessel density and fragmented basement membranes in HGMs increase vascular permeability, thereby increasing contrast media extravasation, interstitial edema, and microstructural injury, resulting in elevated PD values. Accordingly, the postcontrast increase in PD makes the observed histogram changes biologically more meaningful than a mere surrogate of vascularity. Normalizing these changes in the form of dPD% eliminates interlesion baseline bias and enables quantification of the extent of edema and structural injury (23). The MAD reflects the dispersion of lesion pixel values and represents tumor heterogeneity (23). High-grade tumors are more prone to necrosis and cystic degeneration, resulting in a mixture of solid, liquefied, and hemorrhagic components. This pathologic mixture directly broadens the pixel-value distribution and increases the MAD. Consistent with our findings, a previous study (11) reported that HPs correlate with tumor stage. Notably, the standalone diagnostic efficacy of dPD% MAD is moderate, limiting its utility as an independent criterion; it is more suitable as a complementary indicator in multiparametric diagnostic models.

In addition, we further developed combined models to evaluate the differential diagnostic performance. Among all the models, the SyMRI model showed encouraging performance. Notably, this model minimized missed HGM diagnoses and improved grading reliability, with 100% sensitivity (20 of 20). Another interesting finding is that the parameters obtained from dPD% constitute a greater proportion of all the parameters in the combined SyMRI model. Moreover, the IDI results demonstrate that the relative delta model outperforms the absolute delta model in terms of diagnostic performance, highlighting the critical role of relative delta HPs in meningioma grading.

Previous studies have investigated the predictive value of conventional MRI features for meningioma grading, but no conventional MRI features have been shown to reliably grade meningiomas (25). Although only peritumoral edema and bone invasion showed discriminatory value in our findings, these features are unreliable for grading due to interobserver variability. While several functional MRI sequences, including perfusion and diffusion MRI, have been applied to the preoperative evaluation of meningiomas, inconsistencies remain across studies (25–28). Similarly, conventional MRI-based histogram analysis for meningioma grading has also been explored, but the conclusions are heterogeneous (29,30). These conflicting results highlight the need to further validate conventional MRI histogram analysis for meningioma grading. Additionally, previous studies have relied on traditional quantitative mapping for meningioma grading, including T1 and T2 mapping (7,18). Our SyMRI model achieved a higher diagnostic AUC than both T1 and T2 mapping. Notably, the traditional mapping generated only a single parametric map and required prolonged acquisition, limiting their clinical practicality. Moreover, these studies were restricted to a single time point, with changes in HPs before and after contrast agent administration having long been overlooked.

Unlike conventional MRI or traditional mapping methods, SyMRI simultaneously quantifies in a single scan, enabling rapid acquisition and accurate metrics measurement. Built-in corrections for B1 inhomogeneity, coil sensitivity, and scanner drift ensure that pre- and postcontrast signal differences reflect true tumor pharmacodynamics rather than technical noise (10–12,19). SyMRI has proven excellent intra- and interscanner repeatability and linearity, supporting reliable quantitative comparison before and after contrast agent administration (36,37). In this study, we combined SyMRI with delta histogram analysis to predict the meningioma grade. Additionally, SyMRI exhibited notable workflow efficiency. Pre- and postcontrast SyMRI sequences were completed in less than 7 minutes, representing a 30%–40% reduction compared with that of conventional multisequence protocols. Importantly, SyMRI also enabled the automatic generation of all conventional sequences in a single scan. Postprocessing procedures were equally efficient. The VOI segmentation required less than 2 minutes per case, and histogram extraction and analysis were completed within 3–5 seconds. By applying standardized scanner parameters, a fixed contrast agent dosage, a 5-minute postcontrast delay, and consistent postprocessing, this study minimized technical variability and ensured robust quantitative comparisons. Preoperative evaluation confirmed normal circulation in all patients, reducing nonpathologic physiologic confounding effects. Together with the superior temporal efficiency and robust technical reliability of SyMRI, this rigorous standardization underscores the clinical applicability of the technology. Ultimately, this reproducible, noninvasive imaging tool holds promise for facilitating accurate preoperative meningioma grading and guiding personalized therapeutic strategies.

We further analyzed the relationships between SyMRI HPs and the Ki-67 proliferation index, revealing significant associations between various HPs and Ki-67 index. Consistent with previous studies, delta HPs represent biomarkers reflecting tumor heterogeneity and proliferation status, with greater potential than conventional metrics (13,29). It is observed that the energy and total energy of most SyMRI HPs were significantly correlated with the Ki-67 index. These metrics have rarely been studied in meningioma but are associated with proliferation in other tumors (38), although the underlying mechanism remains unclear. Notably, the correlation strength between SyMRI parameters and Ki-67 index was modest, warranting clinical caution when using these parameters to assess proliferative activity. Because aggressiveness and proliferation reflect multifactorial processes, delta HPs capture quantitative changes in tissue relaxometry and contrast enhancement that indirectly reflect these pathologic processes, rather than revealing causal biologic mechanisms. This association is most likely attributed to common pathologic features that influence both imaging phenotypes and tumor biologic behavior. Future studies combining in vitro experiments, longitudinal follow-up, and molecular profiling are needed to clarify the biologic basis of these associations.

There were several limitations to our study. First, this was a single-center, single-scanner study with a small and imbalanced cohort, which may have led to model overfitting and introduced selection bias and may limit the generalizability of our findings. Further multicenter studies with larger, balanced sample sizes and standardized protocols are therefore warranted to validate our results. Second, despite manual delineation of the tumor VOI currently being the reference standard, this process is time-consuming and operator dependent. Future studies should consider the application of automated methods for region delineation to facilitate practical applications. Third, this study lacked an analysis of correlations between SyMRI parameters and genetic/molecular markers. As molecular features are increasingly recognized for predicting meningioma prognosis (39), future studies should investigate the association between imaging characteristics and molecular pathologic markers of meningiomas. Fourth, the inherent heterogeneity of Ki-67 expression within tumors is unavoidable, which may contribute to the difficulty in establishing a robust correlation between HPs and the proliferative index. Finally, the complex quantitative parameters of SyMRI may limit its routine clinical application, especially with heavy workload and limited training in quantitative imaging. Simplified and automated postprocessing algorithms are needed to facilitate clinical translation.

In conclusion, delta HPs derived from SyMRI demonstrate considerable potential as imaging biomarkers for the classification of meningioma grades. Furthermore, the SyMRI model significantly improves diagnostic performance. This approach presents promising prospects for preoperative treatment planning and prognostic assessment in participants, with a shorter scan time. Further studies with larger cohorts are required to validate this quantitative method and enhance its clinical application.

Disclosures of conflicts of interest

Please see ICMJE form(s) for author conflicts of interest. These have been provided as supplemental materials.