A Study About The Value Of The Quantitative Radiomic Features Based On MRI Routine Sequences In Predicting The Grade Of WHO Ⅰ、Ⅱ Meningiomas

Objective:The purpose of the current study is to analyze retrospectively the MRI routine sequences of patients on meningiomas and to investigate the effect of the quantitative radiomic features in predicting the Grade of WHOⅠ、Ⅱ meningiomas.Method:We collected a total of 237 patients from the Neurosurgery Department of the first affilliated hospital of Xiamen University and Zhangzhou Affillated Hospital of Fujian Medical University who were treated by surgery and confirmed by postoperative histopathology,with 143 examples WHO Ⅰmeningioma group and 94 examples WHO Ⅱ meningioma group.Radiological features were extracted from T1 WI and T2 WI.The removing features with student’s t test,and least absolute shrinkage and selection operator(LASSO)were used to select radiomics features.Finally,30 features were selected from T1 WI,16 features were selected from T2 WI,and 27 features were selected from the fusion of T1 WI and T2 WI.Four classifiers were used to train the models(logistic regression,random forests,support vector machine,gradient boosting decision tree),and then 12 models were established using a cross verification method to differentiate WHO Ⅰ from WHO Ⅱmeningiomas.The performance was assessed by receiver-operating characteristic(ROC)analysis,sensitivity,specificity and accuracy.Result: The random forest-based model in T1 WI had better diagnostic performance for meningioma,with sensitivity of 0.62,specificity of 0.9,accuracy of 75%,test set AUC value of 0.798,and 95% confidence interval of 0.6745-0.9209.The model based on gradient boosting decision tree in T2 WI has better diagnostic performance for meningioma,with sensitivity 0.79,specificity 0.79,accuracy 66%,and AUC value of 0.836 in test set(95% confidence interval 0.7349-0.9370).The T1 WI and T2 WI fusion model based on random forest had better diagnostic performance for meningioma,with a sensitivity of 0.83,specificity of 0.85,accuracy of 74%,and AUC value of 0.98,95% confidence interval of 0.7349-0.9370 in the test set.Conclusion: Quantitative radiomics,which uses MRI routine sequences to extract imaging features from T1 WI and T2 WI,is of great value in the prediction of preoperative pathological grading of meningiomas.