ADC histogram parameters differentiating atypical from transitional meningiomas: correlation with Ki-67 proliferation index

Background Preoperative differentiation of atypical meningioma (AtM) from transitional meningioma (TrM) is critical to clinical treatment. Purpose To investigate the role of apparent diffusion coefficient (ADC) histogram analysis in differentiating AtM from TrM and its correlation with the Ki-67 proliferation index (PI). Methods Clinical, imaging, and pathological data of 78 AtM and 80 TrM were retrospectively collected. Regions of interest (ROIs) were delineated on axial ADC images using MaZda software and histogram parameters (mean, variance, skewness, kurtosis, 1st percentile [ADCp1], 10th percentile [ADCp10], 50th percentile [ADCp50], 90th percentile [ADCp90], and 99th percentile [ADCp99]) were generated. The Mann–Whitney U test was used to compare the differences in histogram parameters between the two groups; receiver operating characteristic (ROC) curves were used to assess diagnostic efficacy in differentiating AtM from TrM preoperatively. The correlation between histogram parameters and Ki-67 PI was analyzed. Results All histogram parameters of AtM were lower than those of TrM, and the variance, skewness, kurtosis, ADCp90, and ADCp99 were significantly different (P < 0.05). Combined ADC histogram parameters (variance, skewness, kurtosis, ADCp90, and ADCp99) achieved the best diagnostic performance for distinguishing AtM from TrM. Area under the curve (AUC), sensitivity, specificity, accuracy, positive predictive value, and negative predictive value were 0.800%, 76.25%, 67.95%, 70.15%, 70.93%, and 73.61%, respectively. All histogram parameters were negatively correlated with Ki-67 PI (r = −0.012 to −0.293). Conclusion ADC histogram analysis is a potential tool for non-invasive differentiation of AtM from TrM preoperatively, and ADC histogram parameters were negatively correlated with the Ki-67 PI.


Introduction
Meningiomas are the most common primary central nervous system tumors in adults (1,2), accounting for 39.0% of all intracranial tumors (3).They are divided into three grades and 15 subtypes according to the 2021 World Health Organization (WHO) Central Nervous System (CNS) tumor classification (3).Atypical meningioma (AtM) is a common subtype of WHO grade 2 meningioma and accounts for approximately 5% of meningioma incidence, and its biological behavior is intermediate between benign and malignant, with a high recurrence rate (4,5).The 2016 WHO CNS tumor classification identified brain invasion as a criterion to diagnose AtM (2).Necrosis, peritumoral edema, and irregular shape are the common radiological features.A previous study found that AtM has a relatively high incidence of brain invasion (6).Transitional meningioma (TrM) is a mixture and/or transition between any two or more meningioma pathological types, with a variety of radiological features, including necrosis, calcification, and mild peritumoral edema.TrM belongs to WHO grade 1 meningiomas and is characterized by low malignancy and relatively long disease duration (7), which is different from AtM.The National Comprehensive Cancer Network (NCCN) Meningioma Guidelines (http://nccn.org/)pointed out that surgical resection is the treatment of choice for patients with low-grade meningioma (8), and that a good prognosis can be obtained after surgical resection, while patients with high-grade meningioma often receive adjuvant radiation therapy in order to reduce the recurrence rate and improve overall survival (9).In addition, the Ki-67 proliferation index (PI) is a nuclear antigen that can reflect tumor cell proliferation activity and is often used as an indicator to assess tumor recurrence and prognosis (10).There are significant differences between AtM and TrM in biological behavior, treatment modalities, and prognosis.Therefore, the accurate preoperative differentiation of AtM from TrM is essential for individualized treatment planning.
Magnetic resonance imaging (MRI) is the most commonly used examination modality for CNS tumors (11), and the value of conventional MRI in the study of meningioma grading and differential diagnosis is controversial.Some scholars believe that conventional MRI features rely on radiologists' work experience for the subjective evaluation of tumors, lack objective and quantitative support, and are of limited value in predicting meningioma grade and differential diagnosis (12).Histogram analysis is an objective and reproducible method based on texture analysis, which can quantitatively measure the microstructural information inside the tumor and more comprehensively reflect the physiological function and metabolic changes of the tumor as well as the heterogeneity of histological features (13).Apparent diffusion coefficient (ADC) values obtained by diffusion-weighted imaging (DWI) reconstruction can quantitatively reflect cell density and tumor microstructure.ADC is applied widely in the field of brain tumor assessment, including grading and staging, diagnosis, and prognosis or recurrence of brain tumors (14).Compared with ADC values measured alone, ADC histogram analysis more comprehensively evaluates the overall spatial heterogeneity of tumors by generating multiple parameters, which helps to reveal the deeper pathophysiological heterogeneity of tumors and provide guidance for the selection of clinical treatment options.
ADC histogram analysis, as a quantitative auxiliary diagnostic tool, has been used in the grading, typing, differential diagnosis, and prognosis of meningiomas (15); however, no studies have demonstrated whether ADC histograms can differentiate AtM from TrM.Therefore, the aim of the present study was to investigate the value of ADC histogram analysis in the differential diagnosis of AtM and TrM.We also tried to analyze the correlation between ADC histogram parameters and Ki-67 PI.

General information
This study was approved by the Medical Ethics Committee of the Second Hospital of Lanzhou University (approval no.2020A-109) and the requirement for informed consent was waived.
In this retrospective study, we collected the data of patients with AtM and TrM who underwent MRI examinations and surgical resections in our hospital between January 2019 and June 2022 according to the inclusion and exclusion criteria.The inclusion criteria were as follows: (i) histopathologically confirmed diagnosis of AtM or TrM, and solitary lesion; (ii) meningioma resection within one week after MR examination; and (iii) complete clinical and preoperative imaging data.The exclusion criteria were as follows: (i) chemoradiotherapy, targeted therapy, or other therapy administered to the patient before the preoperative MRI scan; and (ii) poor image quality or incomplete MRI sequence.A total of 78 AtM and 80 TrM were included in this study.

Image analysis
The MRI scans of all enrolled patients were adjusted for window width and level from the PACS workstation and exported to a mobile hard disk for histogram analysis by two experienced neuroradiologists (radiologists 1 and 2 with 8 and 10 years of experience in diagnostic neuroradiology, respectively).MaZda software (http://www.eletel.p. lodz.pl/mazda/) was used by two attending radiologists.Delineation and histogram analysis of regions of interest (ROIs) along the edge of the whole tumor were performed on axial ADC images using a double-blind method.The ROIs of the tumor were first outlined on the T1C image, and the ROIs were then copied to the corresponding sections of ADC images for further analysis (16).In this study, ROIs were placed as far as possible to avoid necrosis and the cystic and hemorrhagic areas of the tumor (17,18); manually delineated ROIs were filled in red, and the software automatically measured and generated the following histogram parameters: mean, variance, skewness, kurtosis, 1st percentile (ADCp1), 10th percentile (ADCp10), 50th percentile (ADCp50), 90th percentile (ADCp90), and 99th percentile (ADCp99).

Pathological examination
This study retrospectively collected patients with meningiomas between January 2019 and June 2022 and reclassified them according to the 2021 WHO CNS tumor classification.Hematoxylin and eosin (H&E) staining was performed on 158 patients with meningiomas after surgical resection.Immunohistochemical analysis of Ki-67 protein was performed using a monoclonal mouse anti-human Ki-67 antibody with brownish-yellow nuclei judged as Ki-67 positive cells and Ki-67 PI expressed as the number of Ki-67 monoclonal antibody positive cells per 1000 tumor cells in the most mitotically active area.

Statistical methods
All data were analyzed using MedCalc version 19.1 (MedCalc, Mariakerke, Belgium), SPSS version 25.0 (IBM Corp., Armonk, NY, USA), and R software (https:// www.r-project.org/) for analysis.Tests for normal distribution were performed using the Shapiro-Wilk test, and those with continuous variables conforming to normal distribution were presented as mean ± standard deviation; otherwise, the median (lower and upper quartiles) and histogram parameters between AtM and TrM were compared using the independent samples t-test or Mann-Whitney U test.Categorical variables were compared using the chi-square test.Reproducibility of all histogram parameters was evaluated with the intraclass correlation coefficient (ICC), and an ICC >0.75 indicated a good result.Statistically significant histogram parameters were compared and analyzed by plotting receiver operating characteristic (ROC) curves, and the area under the ROC curve (AUC), sensitivity, specificity, accuracy, positive predictive value, and negative predictive value were obtained to assess their value in differentiating AtM from TrM preoperatively.The relationship between histogram parameters and Ki-67 PI was analyzed using Pearson's correlation.Differences were considered statistically significant at P < 0.05.

Comparison of ADC histogram parameters between AtM and TrM
All ADC histogram parameters have a good to excellent inter-observer agreement (ICC = 0.823-0.991).Between the AtM and TrM groups, all histogram parameters in the AtM group were smaller than those in the TrM group (Table 2); the variance, skewness, kurtosis, ADCp90, and ADCp99 were significantly different between the two groups (P < 0.05) (Fig. 3).However, the mean (P = 0.116), ADCp1 (P = 0.121), ADCp10 (P = 0.090), and ADCp50 (P = 0.079) were not significantly different.

ROC curve analysis of ADC histogram parameters
The ROC curve analysis of histogram parameters with statistical differences between the AtM and TrM groups is shown in Table 3 and Fig. 4. ROC analysis showed that the variance, skewness, kurtosis, ADCp90, and ADCp99 could differentiate AtM from TrM preoperatively and noninvasively.When combined, the ADC histogram parameters had the best differential diagnostic efficacy and the AUC increased to 0.800 (range = 0.729-0.859).Values in parentheses are 95% CI.

Discussion
AtM and TrM are the two common subtypes of meningiomas.They have different prognosis, survival, and clinical treatment options.Therefore, accurate preoperative differentiation of AtM and TrM has important clinical value.Our findings suggest that ADC histogram analysis may be helpful in differentiating AtM from TrM as an auxiliary diagnostic tool, with combined ADC histogram parameters achieving the best diagnostic performance.To our knowledge, this is the first study to differentiate between AtM and TrM using ADC histogram analysis.
DWI is a non-invasive method of MRI that quantitatively analyzes the diffusion of water molecules in the body (19).ADC values can reflect cellularity and/or proliferation activity in tumors (20), and then evaluate the biological characteristics of tumor cells.In other words, compared to benign tumors, malignant tumors with dense tumor cells and large nuclei have more restricted diffusion of water molecules and lower ADC values.In a meta-analysis of the correlation between ADC values and benign and high-grade meningiomas, the authors suggested that ADC values may reflect the level of meningioma aquaporin 4 and correlate with progesterone receptor status in meningiomas (15).Several previous studies have found that high-grade meningiomas have lower ADC values than low-grade meningiomas (21,22).ADC histogram analysis has been shown to be a potentially useful tool in the study of the differential diagnosis, grading, and prognosis of brain tumors by extracting features from the entire tumor and automatically generating multiple histogram parameters to achieve a comprehensive and quantitative assessment of tumor heterogeneity and other features (22)(23)(24)(25).
ADC histogram analysis obtains certain ADC histogram parameters and traditional ADC values were similar, which could reflect tumor cell density and distribution.Low ADC percentile values correspond to low ADC values and higher cell density; conversely, higher ADC percentile values correspond to high ADC values, lower cell density, and unrestricted diffusion of water molecules (14).Our study showed that the mean, ADCp1, ADCp10, ADCp50, ADCp90, and ADCp99 of AtM were less than those of   TrM; ADCp90 and ADCp99 were significantly different between the two groups (P < 0.05 for all).The most plausible explanation is closely related to the histopathological features of AtM and TrM.Compared to TrM, AtM is mitotically active, has a higher tumor cell density, and water molecule diffusion is easily restricted.Tumor cell composition (tumor cell density and plasma ratio) is a key factor affecting the diffusion of water molecules.The TrM is mainly composed of spindle cells and arranged in bundles or whorls in a relatively sparse manner, which widens the extracellular space and leads to a reduction in the degree of water molecule diffusion restriction.A multicenter study of ADC histograms to predict meningioma grade (26) found that WHO grade 2/3 meningiomas showed significantly lower ADC histogram analysis parameters, especially median ADC, compared with WHO grade 1 tumors, similar to the results of this study.Liu et al. (14) used ADC histograms to identify solitary fibrous tumor (SFT) and AtM, and mean, ADC1, ADC10, ADC50, and ADC90 of the SFT/hemangiopericytoma (HPC) were greater than those of the AtM group, and significant differences were observed (P < 0.05).Gihr et al. (27) found that the entire ADC profile (p10, p25, p75, p90, mean, median) was significantly lower in high-grade versus lowgrade meningiomas.Several previous studies on the identification of SFT with meningioma yielded similar findings.Liu et al. (28) found that SFT showed a significantly lower mean, ADC1, ADC10, ADC50, ADC90, and  ADC99 than microcystic meningioma (McM) (all P < 0.05); Wang et al. ( 29) compared the differences between the ADC of intracranial SFT and TrM and found that SFT exhibited significantly lower AP1 and AP10 (all P < 0.05) than TrM.
Variance refers to the change relative to the mean gray value and is used to reflect the heterogeneity of the tumor (30); skewness reflects the symmetry of the distribution of variables from the curve shape characteristics (16,30), and the greater the absolute value of skewness, the greater the morphological deviation of the distribution; kurtosis describes the flatness of the overall numerical distribution (16), and these histogram parameters are associated with the heterogeneity of the tumor and play an important role in the differential diagnosis and prognostic evaluation of the tumor.A previous study (31) showed that ADC variance can help to distinguish meningoepithelial from fibroblastic meningiomas.Another study (32) on differentiating between meningioma and schwannoma showed that schwannomas had significantly higher ADC values than meningiomas and the kurtosis of ADC had the best AUC among all histogram measures (AUC = 1.0).In this study, variance, skewness, and kurtosis were significantly different between the two groups.Contrary to the expected results, variance, skewness, and kurtosis of TrM were greater than those of AtM, which may relate to why the ROIs were delineated to not include cystic degeneration, necrosis, and hemorrhages in the tumor and could not more objectively reflect the heterogeneity of the tumor.Alternatively, this may be attributed to the sample size.Further studies may be warranted.
Ki-67 is a nuclear protein that is an integral component of the mitotic chromosome periphery.Ki-67 PI can be detected in cells by immunohistochemistry and is considered a biomarker reflecting tumor cell proliferation and predicting prognosis (33).A previous study (34) found that Ki-67 PI showed a significant increase from WHO grade 1 to WHO grade 2, which may provide useful prognostic information, which is similar to the findings that the Ki-67 PI of AtM was significantly higher than that of TrM in this study, and tumor cell proliferation activity increased with Ki-67 PI.Ki-67 PI was statistically significantly negatively associated with ADC (r = −0.63,P < 0.001) in a multicenter analysis of using DWI differentiating between malignant and benign meningiomas (35).This study showed that all histogram parameters and Ki-67 PI were negatively correlated and the correlation coefficient between Ki-67 and ADC histogram parameters were relatively low.One possible explanation is that the more complex composition of atypical meningiomas, the significant tumor heterogeneity, and the possible presence of microhemorrhages and microliquefaction necrosis within the lesion and other components may have an impact on the ADC value.Surov et al. (36) performed ADC whole-lesion histogram analysis of meningiomas  31) also found similar results-that Ki-67 PI is correlated with the following histogram parameters: C90, mean of T1, kurtosis, C10, and variance of ADC (all P < 0.05).Georg et al. (37) found that skewness was significantly associated with Ki-67 PI in meningiomas and is a valuable molecular marker for differentiating low-and high-grade meningiomas and assessing the risk of tumor recurrence.
The present study has some limitations.First, it was a single-center retrospective study, and further multicenter prospective studies with large sample sizes are warranted.Second, only ADC images are analyzed using histograms and manual segmentation is time-consuming.Third, the study of multi-sequence MRI images based on artificial intelligence (deep learning) needs to be explored.Finally, all the imaging sequences were 2D acquisitions in this study.
In conclusion, the results of this study showed that ADC histogram analysis may be helpful in differentiating AtM from TrM.In addition, ADC histogram parameters were negatively correlated with Ki-67 PI.Therefore, ADC histogram analysis may be used as a non-invasive imaging method for the preoperative assessment of Ki-67 PI in AtM and TrM.

Fig. 4 .
Fig. 4. ROC curves for ADC histogram parameters in discriminating between TrM and AtM, and the combined ADC histogram provided the best diagnostic performance with an AUC of 0.800.ADC, apparent diffusion coefficient; AtM, atypical meningioma; AUC, area under the ROC curve; ROC, receiver operating characteristic; TrM, transitional meningioma.

Table 1 .
Comparison of baseline clinical and radiological features between AtM and TrM.
Values are given as n (%) or mean ± SD, unless otherwise indicated.AtM, atypical meningioma; TrM, transitional meningioma

Table 2 .
Comparison of ADC histogram parameters between AtM and TrM groups.Values are given as mean ± SD, unless otherwise indicated.Values in parentheses are upper and lower quartiles, outside the parentheses is the median, ADC, apparent diffusion coefficient; AtM, atypical meningioma; CI, confidence interval; SD, standard deviation; TrM, transitional meningioma.

Table 3 .
ROC curve analysis of ADC histogram parameters distinguishing AtM from TrM.
Varianceand found that the histogram parameters were negatively correlated with Ki-67 PI except ADC min .Cao et al. (