MRI Based Brain-tumor Interface Radiomic Features With 4mm Width Have The Advantage In Meningioma Invasiveness Diagnosis

Purpose: Meningioma invasion has a great influence on postoperative recurrence and prognosis which may greatly influences the judgment of clinical decision making.Studies have shown that conventional magnetic resonance imaging based radiomics can be used for preoperative diagnosis of meningioma invasion,and the role of preoperative inflammatory biomakers in the histological grading of meningioma has also been reported in relevant literatures.Inspired by histopathological characteristics of invasive meningioma,we supposed to evaluate nine radiomic-based models constructed from different regions(braintumor interface2-5mm,whole tumor and consolidation of the two regions above)to further exploring whether brain-tumor interface radiomic features have the overall preponderance in meningioma invasion differenciation.Clinical risk factors such as peritumoral edema volume and peripheral blood inflammatory biomarkers were evaluated to determine whether they have the power for tumor invasion distinguishing.Besides,tumor volume growth affects the choice of treatment strategy for meningioma,in this paper,we evaluated the accuracy of several simple volumetric algorithms for estimating meningioma volume.Method: In this study,505 cases of meningioma in our hospital cohort and 214 cases in taihe hospitals cohort were enrolled to collect clinical parameters and original magnetic resonance images.Tumor segmentation was performed on the Contrast Enhancement-T1 images.After image processing,1015 extracted radiomic features were then filtrated using LASSO and RF algorithm.Then,the discrimination efficiency was compared among the nine regions using overall area under curve(AUC)values.Besides,principal component analysis(PCA)of radiomic features was performed to build the most predictive power model for each region.Our research also evaluated the predictive performance of clinical indicators.Result: Compared with the other eight regions,the model constructed based on brain-tumor interface 4mm region has the highest ROC-AUC in the internal training set,the internal validation set and the external validation set,which are as follows:(0.891,0.85,0.932)(0.851)(0.743,0.96))and(0.881,0.833,0.928).The Precision Recall area under the curve(PRAUC)was also the highest one for brain-tumor interface,and there was a significant statistical difference FOR ROC-AUCs and PR-AUCs.The most discriminative indexes came from GLCM and GLDM datasets.Combined with peritumoral edema volume feature,the differentiation ability of brain-tumor interface 4mm modelwas improved.The 2/3SH and1/2ABC algorithms were more accurate in the estimation of meningioma volume,to be specific,the 2/3SH method showed a more accurate performance for the small volume subset meningioma compared with 1/2ABC method.Conclusion: Radiomic features based on 4mm brain-tumor interface area showed a overall preponderance in the identification of invasive meningioma,and the combination of peritumoral edema volume is helpful in improving the discrimination ability.There was no statistic differences in peripheral blood inflammatory markers between invasive and noninvasive meningiomas.The simple 2/3SH and 1/2ABC algorithms can be used to accurately evaluate the volume of meningioma,and when it come to small meningioma the 2/3SH method is more suitable in the measurement.Others: For small meningiomas,4mm width of tumor brain interface may not represent the optimal width for identifying invasive meningiomas,and whether the optimal width of tumor brain interface is correlated with tumor volume remains to be further explored.