Objective:To investigate the application value of clinical imaging characteristics and multi-parameter MRI radiomics features in evaluating the H3K27M mutation status of diffuse midline glioma(DMG)in children.Methods:98 patients with diffuse midline glioma(DMG)were retrospec-tively included,including 74 patients with H3K27M mutant and 24 with wild-type H3K27M.They were divided into training set(n=68)and testing set(n=30)according to the ratio of approximately 7∶3.The radiomics features were extracted from T2WI and contrast-enhanced T1WI sequences(cT1WI).The minimal redundancy maximal relevance(mRMR)and least absolute shrinkage and se-lection operator(LASSO)were used to select the best features in the training set to calculate the Rad-Score.Clinical image features and Rad-Score were incorporated into multivariate logistics regression to screen independent risk factors.The screened clinical imaging features and radiomics features were used to construct a combined model to predict the H3K27M status of DMG.A nomogram was used to visualize the combined model.The receiver operating characteristic curve(ROC)was used to evaluate the diagnostic efficacy of the models.Decision curve analysis(DCA)was used to assess the clinical ap-plication value of the model.Results:1648 radiomics features were extracted based on T2WI and cT1WI sequences,and five radiomics features were finally selected to construct the radiomics model.The mod-el showed good prediction ability on the training and testing sets,and the area under the curve(AUC)was 0.844 and 0.758,respectively.Multivariate logistics regression showed that ring enhancement and minimum apparent diffusion coefficient(ADCmin)were risk factors of clinical imaging characteristics associated with H3K27M status(all P values＜0.05).The clinical imaging models constructed by ring enhancement and ADCmin exhibit moderate efficacy in predicting H3K27M status,and the AUC of the training set and testing set were 0.802 and 0.720,respectively.The combined model constructed by ring enhancement,ADCmin,and Rad-Score shows the optimal prediction performance for H3K27M sta-tus,with AUC of 0.863 and 0.851 in the training and testing sets,respectively.Conclusion:The com-bined model constructed based on clinical imaging features and radiomics features extracted from multi-parameter MRI has great clinical application value in evaluating the H3K27M status of DMG in children.
维普
万方数据