In this research we retrospectively collected a total of 191 patients who received surgery and were diagnosed with primary VS in the First Affiliated Hospital of Zhengzhou University between December 2013 and August 2020. All patients underwent a preoperative plain and gadolinium-enhanced MRI in our center, the MRI sequences were collected including T1-weighted images, T2-weighted images, T2-weighted-Fluid-Attenuated Inversion Recovery (T2-FLAIR) images, and T1-weighted gadolinium-enhanced (T1-CE) imags. Tumors in Stage I (tumor confined to the internal auditory canal, diameter 1-10mm) by Koos classification were excluded because the blood supply of such small tumors may be difficult to define through the evaluation criteria of this study, and there was a possibility that insufficient radiomics information could be extracted. All surgical procedures were performed by the same highly qualified neurosurgeon. Patients were placed in a lateral position under general anesthesia during the whole procedure. All procedures were performed via a suboccipital retrosigmoid approach, and after revealing the sigmoid sinus and transverse sinus, the cerebrospinal fluid was fully released and the tumor was resected in pieces under the microscope. The blood supply of the tumor was recorded and confirmed by the senior neurosurgeon and the assistant together. Those who had less bleeding during tumor resection, the bleeding was easily removed by an aspirator, and the field under the microscope could always be kept clean, were recorded as poor blood supply; those who had more bleeding during tumor resection, and the blood in the field under the microscope was difficult to remove completely by using only one set of aspirator, were recorded as rich blood supply. The amount of intraoperative bleeding, which estimated by neurosurgeon at the end of the surgery, was also recorded to assess the reliability of the grouping.
MRI data acquisition
MRI was performed with a MAGNETOM Skyra 3.0T scanner (Siemens Medical Solutions, Erlangen, Germany) and standard head coil. The imaging sequences in our study included: axial T1-CE (repetition time, 434 ms; echo time, 2.5 ms; slice thickness, 5 mm) with gadopentetate dimeglumine (Magnevist, Bayer Healthcare) was administered by injection through a peripheral venous catheter at a dose of 0.2 mmol/kg, axial T1 data (repetition time, 434 ms; echo time, 2.5 ms; slice thickness, 5 mm), Axial T2 data (repetition time, 434 ms; echo time, 2.5 ms; slice thickness, 5 mm), and axial T2-Flair data (repetition time, 434 ms; echo time, 2.5 ms; slice thickness, 5 mm). Fig. 1 shows the radiomics workflow in this study.
Imaging preprocessing and ROI segmentation
The preprocessing started with the registration (brain) module of the 3D-Slicer program (version 4.11.0, Windows 64bit) to co-registration T1WI, T2WI, and T2-Flair to the T1-CE sequence (Percentage Of Samples 0.002, B-spline Grid Size 14, 10, 12), and the image correction was performed by the N4ITK MRI field bias correction module with (1,1,1) B-spline grid resolution as the confusion matrix. Next, the pyradiomics package was applied in Python environment to perform Image Normalization (normalizeScale = 100), and resampling the image (ResamplePixelSpacing = [3, 3, 3], Interpolator = sitkBSpline).
The region of interest (ROI) was drawn on T1-CE separately by two neurosurgeons with over 5 years of clinical experience using 3D-Slier software. Neurosurgeon-A drew all ROIs manually; and Neurosurgeon B applied the Nvidia AI-Assisted Annotation (Nvidia AIAA) segmentation module to semi-automatically draw ROIs by placing boundary-points and manually correct misdraw.
Radiomic feature extraction
Feature extraction was performed with the python pyradiomics package, using the 8 Filter Classes built into the package, including Original, Wavelet, LoG (sigma [3.0, 5.0]), Square, SquareRoot, Logarithm, Exponential, and Gradient, using 7 feature classes, including First Order Statistics, Shape-based (3D), Shape-based (2D), Glcm: Gray Level Cooccurence Matrix, Glrlm: Gray Level Run Length Matrix, Glszm: Gray Level Size Zone Matrix, Gldm: Gray Level Dependence Matrix, and using the ROI drawn by each of doctors A and B as masks to obtain two sets of feature for A and B, respectively.
The intraclass correlation coefficient (ICC) of feature sets A and B was calculated using the Pingouin package in Python, and only features with high stability (ICC > 0.8) were retained. To minimize human factors, we selected feature set B (ROI drawn by semi-automatic method) to proceed to the next step of feature selection. Feature selection was performed using 4 methods: student t-test, Least Absolute Shrinkage and Selection Sperator (LASSO), ANOVA, and t-test + LASSO. The optimal λ value in LASSO was automatically selected by bootstrap methods.
Feature classificationand model validation
The radiomic features selected by different methods were used to establish the model by the scikit-learn package (Version 0.23.0) in python by entering them into Multivariable Linear Regression model (MLR), Support Vector Machines (SVM), Random Forest (RF), and Tree Models, respectively. The 5-repeats-3-fold cross-validation method was used, which divided the case samples into the training set and validation set at the ratio of 2:1, and repeated the validation 5 times to obtain a total of 15 predictions results from each, and plotted the receiver operating characteristic curve (ROC) graphs of cross-validation for different feature selection methods and classifier combinations of validation sets respectively, and calculated their average area under curve (AUC). The best combination was selected by its performance (AUC). The accuracy, sensitivity, and F1-score, which considered both accuracy and sensitivity of the best model, were also calculated.
Artificial judgment was performed by two other neurosurgeons with over 5 years of clinical experience, they were blinded to the intraoperative recording, and MRI image features of both cohorts were identified by visual observation. Higher signal of the tumor on T1-CE sequences, the finding of multiple flow voids on the tumor surface or in the tumor parenchyma, and solid tumors with less cystic, marked the tumor prediction as rich in blood supply. Lower signal of the tumor on T1-CE sequences, no finding of obvious flow voids related to the tumor, and tumors with multiple cystic, marked the tumor prediction as poor in blood supply. The prediction results of two neurosurgeons were recorded and compared with the gold standard (intraoperative record), and the accuracy, sensitivity, and F1-score were calculated and compared with the prediction results of the machine-learning model.
The statistical analysis of baseline data was performed using IBM SPSS Statistics 21. The quantitative data was analyzed using Student’s t-test, and the qualitative data was analyzed using Pearson’s Chi-square test, p-value < 0.05 was considered as statistical significance.