Objective To explore the value of the radiomics model based on conventional multi-sequence MRI in pre-operative prediction of Ki67 expression in meningiomas.Methods A retrospective analysis was performed on 305 pa-tients with meningioma confirmed by surgery and pathology in China-Japan Union Hospital of Jilin University from March 2013 to November 2021.All patients underwent MRI scanning before surgery to obtain T1 weighted imaging(T1WI)、T2 weighted imaging(T2WI)、T2-fluid attenuated inversion recovery(T2-FLAIR),and T1WI enhanced ima-ging(T1C).ITK-SNAP software was used to delineate the edge of meningioma.The resulting EnHROI was then inflat-ed by 3 mm and 5 mm on the uAI Research Portal radiomics platform to obtain the EnH3mmROI and EnH5mm ROI.The imaging was preprocessed before extracting radiomics,and then the obtained radiomic features were screened and reduced in dimension.The statistical features of radiomic were used to establish models by Quadratic Discriminant A-nalysis and Logistic Regression.To evaluate the performance of different prediction models by receiver operating char-acteristic curve(ROC)and the area under the ROC curve(AUC)in the training and testing sets.Results The models established by two machine learning methods have good prediction performance,and the quadratic discriminant analysis model exhibited higher and more stable diagnostic efficacy,the AUC values of EnH model,EnH3mm model and EnH5mm model in training set data were 0.806、0.841、0.773 while in validation set data were 0.776、0.818、0.757 re-spectively.Comparison between the combined models showed that the model based on EnH3mm model had the best performance,the specificity in the training and testing sets were 0.882 and 0.857,and the accuracy were 0.796 and 0.777,respectively.Conclusion The radiomics model based on conventional multi-sequence MRI has some value in pre-dicting the Ki67 expression status of meningiomas,and EnH3mm model has better diagnostic efficiency.
NETL
NSTL
万方数据
