This retrospective study was carried out following the relevant guidelines and regulations, and the methods used in this study were approved by the National Ethics Committee. The MRI images of patients were used in this study without any intervention in the diagnostic or treatment procedures. In addition, informed consent was waived because of the retrospective nature of the study.
Overall, preoperative MRI images of 45 patients with the diagnosis of meningioma tumors at different grades (9 patients with grade III, 16 patients with grade II, and 20 patients with grade I tumors) were used in this study. The participants consisted of 15 men and 30 women with a mean age of 55.6 ± 9.1 years, ranging from 34 to 75.
The images of the patients were collected from the database of the Imaging Department, Tajrish Hospital (one of the biggest hospitals in Tehran, Iran). The imaging procedures of all the images used in the present study were performed during 2019 and 2020. Tumor grading according to the WHO classification  was obtained from the patients’ health documents archived in the imaging department.
The patients did not have any history of brain surgery and the MRI images were taken before the biopsy procedure. In addition, participants with a history of chemotherapy or radiotherapy were excluded from this study. Notably, there were several inclusion criteria for the patients’ MRI images; the assessed tumors had to be primarily bigger than 10 mm in size (in the smallest dimension), and the data related to tumor grading had to be available based on WHO classification . Furthermore, MRI images had to contain transversal T1 (pre- and post-contrast), T2, T2-FLAIR-weighted images along with ADC maps data.
Imaging procedures were performed using 1.5T scanners (Siemens MAGNETOM Avanto, Siemens Healthcare, Germany) using a head coil. In this study, the following sequences were analyzed: axial T1-weighted (pre- and post-contrast images), fast spin-echo sequence (TR 400 ms, TE 14 ms, flip angle 90°, 4-mm slice thickness, and 280-mm field of view), axial T2-weighted (TR 2820 ms, TE 95 ms, flip angle 90°, 4-mm slice thickness, and 280 mm field of view), and axial T2- FLAIR (TR 8000 ms, TE 125 ms, flip angle 90°, inversion time 2000, 4-mm slice thickness, and 280-mm field of view). The ADC images were acquired by applying diffusion gradients in three independent directions. All images were taken by three gradient b-values of 0, 500, and 1000 mm2/s respectively. DICOM (digital imaging and communication on medicine) software (DicomWorks v1.3.5 2000, 2002; License: Freeware Free; Publishers: Phillippe Puech and Loic Bousse) was used to obtain the ADC values (ADC map) in different regions of diffusion-weighted MRI images.
Analyzing the image
All the MR images and ADC maps were saved in DICOM format for better prevention of image information loss. The images were imported to Ray station treatment planning software (Version 8.A, Raysearch Laboratories, Sweden) and the tumor region (the whole lesion based on the tumor boundaries), as well as normal brain tissue, were then contoured in the post-contrast T1-weighted images (slice by slice) and subsequently mapped (copied in the same region) on the other image sequences. The treatment planning software reconstructed the volume by connecting the contoured ROIs in all the MRI slices. The details of volume reconstruction using ROIs and contoured in series of two-dimensional images with known slice thickness were explained in a previous report . Two independent radiation oncologists contoured the ROIs of tumor and brain, and an experienced radiation oncologist (with more than 20 years of experience) reviewed the contours of tumors in all the patients to ensure that the correct tumor region was selected.
The MR images alongside the related structures (in DICOM format) were imported to the CERR (Computational Environment in Radiotherapy Research) , a MATLAB-based application (MATLAB Ver.2019b, MathWorks company, MA, USA). In summary, CERR allowed us to use DICOM structures delineated in the treatment planning system on MR images and obtain the three-dimensional intensity histogram in each of the image sequences. After obtaining the histogram of the tumor and normal brain tissue for every image, the following parameters were calculated: mean; maximum value; minimum value; median; mode; and 10th, 25th, 50th, 75th, and 90th percentiles as well as kurtosis and skewness.
The parameters derived from three-dimensional histograms from each of the image sets were compared between different tumor grade groups utilizing the non-parametric statistical Kruskal-Wallis test. Furthermore, the multi-parametric linear regression analyses were used to find a model from selected histogram parameters having higher predictive value (i.e., having lower P values) for differentiating the meningioma grades in different MRI image series. In addition, Spearman’s correlation was performed to find the correlation coefficients between the histogram parameters with significant differences (P value < 0.05) using all the patients’ data (without considering the tumor grade) in different MRI image series. The receiver operating characteristic (ROC) curve was used to evaluate the diagnostic ability of the histogram parameter values to differentiate between high-, intermediate-, and low-grade meningioma tumors. The cutoff value was chosen to maximize the Youden index. Furthermore, sensitivity and specificity (at cutoff points), and also the AUC (area under the curve) values, were calculated using the obtained ROC curves. The level of statistical significance was set at P < 0.05, and all the statistical tests were performed using SPSS software package, V18 (SPSS Inc., Chicago, USA).