Investigation of radiomics based on 18F-FDG PET/CT in predicting the COG risk stratification of neuroblastoma / 中华核医学与分子影像杂志

ABSTRACT

Objective:To explore the value of radiomics based on 18F-fluorodeoxyglucose (FDG) PET/CT in predicting the Children′s Oncology Group (COG) risk stratification of neuroblastoma (NB). Methods:From March 2018 to November 2019, the 18F-FDG PET/CT images of 125 NB children (51 males, 74 females, age 0.5-10.5 years) confirmed pathologically in Beijing Friendship Hospital were retrospectively analyzed. According to the COG classification, patients were divided into high-risk group and non-high-risk group (including low- and intermediate-risk). Imaging radiomics features were extracted from PET and CT images and screened. Logistic regression was used to build the first model based on radiomics features (R_model) and calculate radiomics score (Rad_score), then build the second model (RD_model) based on Rad_score and demographic features and at last build the third model (RDC_modle) based on Rad_score, demographic features and clinical features. The receiver operating characteristic (ROC) curve was used to evaluate the predictive efficacy of these models. Results:The training set contained 94 NB cases (63 high-risk cases, 31 non-high-risk cases), and the validation set contained 31 NB cases (21 high-risk cases, 10 non-high-risk cases). Four radiomics features were obtained by screening, of which two features were based on CT images and the other two features were based on PET images. The area under the curves (AUCs) of the R_model, RD_model and RDC_model in training or validation set were 0.91, 0.94, 0.98 or 0.86, 0.92, 0.95, respectively. The accuracies of the R_model, RD_model and RDC_model in training or validation set were 86%(81/94), 89%(84/94), 93%(87/94) or 84%(26/31), 84%(26/31), 87%(27/31), respectively.Conclusions:Radiomics based on 18F-FDG PET/CT can accurately predict the COG risk stratification of NB. Prediction model of radiomics features combined with demographic and clinical characteristics can further improve the accuracy of predicting NB COG risk stratification, which can help personalized and precise therapy protocol management in NB.