Application Of Different Radiomics Diagnostic Models Based On Conventional MRI Sequence In Preoperative Grading Of Grade Ⅱ And Ⅲ Glioma

Objective:Based on multiparametric MRI radiomics features to establish radiomics model for the preoperative grading of lower-grade gliomas(LGG).Exploring the feasibility of radiomics model in LGG preoperative classification.Materials and Methods:A retrospective collection of grade Ⅱ and grade Ⅲ gliomas that were resected by neurosurgery from December 2014 to December 2020 in my Hospital and confirmed by pathology.And all patients underwent routine head MRI examinations before surgery(T2WI and T1 WI enhanced).After certain entry and exclusion criteria,98 glioma patients meeting our experimental needs were screened out,43 cases were grade Ⅱ gliomas and55 cases were grade Ⅲ gliomas.The DICOM format images of the T2 WI sequence and T1 WI enhanced sequence of these 98 patients were imported into the “Darwin Scientific Research Platform”.The region of interest(ROI)of T2 WI and T1 WI enhanced were delineated layer by layer on this platform.A total of 98 glioma patients were randomly divided into two groups with a 7:3 ratio: training group(n=69)and test group(n=29).1688 radiomics features are extracted from T2 WI and T1 WI enhanced images respectively,including 18 first-order statistics,14 Size-and shape-based features,75 texture features and 1581 high-order statistics.Then,F test and Least absolute shrinkage and selection operator(LASSO)algorithm are used to reduce dimensionality and feature selection of all data extracted.7 related features were screened out from T2 WI,and 6related features were screened out from T1 WI enhanced.Finally,the two functions of support vector machine(SVM)and logistic regression(LR)were selected to build four machine learning models based on the related features after reduction and screening.SVM model and LR model based on T2 WI.SVM model and LR model based on T1 WI enhanced.Med Calc software is used to draw receiver operator characteristic(ROC)curve.And calculate the area under curve(AUC),specificity and sensitivity of these four models according to the ROC curve.Results:Among 43 grade Ⅱ gliomas,the most common pathologic type was astrocytoma(24,55.8%),followed by oligodendroglioma(14,32.6%),and oligoastrocytoma(5,11.6%).Among 55 grade Ⅲ gliomas,the most common pathologic type was astrocytoma(27,49.1%),followed by oligodendroglioma(21,38.2%),and oligoastrocytoma(7,12.7%).The results showed that astrocytoma was the most common pathological type,and there was no statistical significance in pathological type between grade Ⅱ and grade Ⅲ gliomas(P>0.05).A total of 3376 radiomics features were extracted from T2 WI and T1 WI enhanced,By calculation,the ICC range within the group is between 0.751-0.959,and the ICC range between the groups is between 0.760-0.944.ICC results show that the extracted features have high reproducibility.The average age of patients with grade Ⅱ glioma is 49.6±11.3 years,and the average age of patients with grade Ⅲ glioma is 53.2±12 years.Although the average age of patients with grade Ⅱ glioma is lower than that of patients with grade Ⅲ glioma,there was no statistically significant difference(P>0.05).Among patients with grade Ⅱ glioma,21 were male(48.8%)and 22 were female(51.2%).Among patients with grade Ⅲ glioma,34 were male(61.8%)and 21 were female(38.2%).The gender composition of patients with grade Ⅱ glioma and grade Ⅲ glioma the difference was not statistically significant(P>0.05).In the radiomics model based on T2 WI using SVM and LR classifiers,the AUC value of the SVM model in the training group is 0.855,the specificity is 87.50%,the sensitivity is 81.58%,and the AUC value of the LR model is 0.876,the specificity is 87.50%,and the sensitivity is 84.21%;the AUC value of the SVM model in the test group is 0.842,the specificity is 69.23%,and the sensitivity is 94.12%.The AUC value of the LR model is 0.869,the specificity is 69.23%,the sensitivity is 94.12%.In the radiomics model based on T1 WI enhanced using SVM and LR classifiers,the AUC value of the SVM model in the training group is 0.922,the specificity is 96.77%,the sensitivity is 79.49%,and the AUC value of the LR model is0.913,the specificity is 96.77%,and the sensitivity is 79.49%.The AUC value of the SVM model in the test group is 0.888,the specificity is 78.57%,and the sensitivity is93.75%.The AUC value of the LR model is 0.884,the specificity is 92.86%,the sensitivity is 75.00%.Conclusions:This study shows that the four radiomics diagnostic models established from the important features extracted from T2 WI and T1 WI enhanced images have high diagnostic value for the pathological tissue classification of grade Ⅱ and grade Ⅲ gliomas.The SVM model based on T1 WI enhanced has the highest diagnostic efficiency.