基于多序列MRI影像组学特征联合模型对胶质母细胞瘤总生存期的预测价值OA

Objective:To evaluate the potential of multi-sequence MRI radiomics features in predicting the overall survival with glioblastoma patients.Methods:A total of 352 patients with pathologically confirmed glioblastoma were included retrospectively(211 cases in the training set,141 cases in the validation set).The high-dimensional radiomics features of the images were extracted using the PyRadiomics software and screened layer-by-layer using the univariate Cox regression and least absolute shrinkage and selection operator(LASSO)algorithm.The filtered optimal features were used to construct four sequence models,and a combined model was built together with clinical features.Results:Across all sequence models,patients in the high-risk subgroup had a significantly worse clinical prognosis compared with the low-risk subgroup.In the validation set,the AUC values of the combined model predicting 1-,2-,and 3-year overall survival were 0.740(95%CI 0.636-0.840),0.730(95%CI 0.619-0.841),and 0.790(95%CI 0.669-0.915).The calibration curves showed good agreement between the survival rates predicted by the combined model and the actual survival rates.Conclusions:The combined model based on multi-sequence MRI radiomics features can accurately predict the overall survival and facilitate personalized treatment in glioblastoma patients.