| Purpose: To analyze the correlation between the radiomics features of intratumoral and peritumoral zones and the response to concurrent chemoradiotherapy(CCRT)of cervical squamous cell carcinoma,and to explore the predictive value of two-dimensional(2D)and three-dimensional(3D)radiomics models for the response to CCRT of cervical squamous cell carcinoma.Methods: A total of 132 patients with cervical squamous cell carcinoma who were examined by dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI)in our hospital from January 2020 to August 2022 were collected retrospectively.According to the proportion of 7∶3,the patients were randomly divided into the training set(n=92)and validation set(n=40).General clinical data such as age,the state of menstruation,FIGO stage,maximum tumor diameter(MTD)and lymph node state reported by MRI were collected.Based on the axial images of DCE-MRI before CCRT,2D intratumoral region of interest(ROI),3D intratumoral-ROI and 3D peritumoral-ROI were sketched respectively,and Py Radiomics was used to extract the radiomics features.The extracted radiomics features parameters were standardized by the Z-score,and the correlation analysis and least absolute shrinkage and selection operator(LASSO)algorithm were used for dimensionality reduction and screening.According to the selected features and their corresponding coefficients,the radiomics scores(Rad-score)were calculated and the radiomics models(2D-intratumoral model,3Dintratumoral model,3D-peritumoral model and 3D combined model)were established based on the Logistic regression.The receiver operating characteristic(ROC)curve was applied to evaluate the distinguishing performance of the model,and the internal validation of 1,000-iteration Bootstrap analysis was used to evaluate the stability of performance for the four radiomics models,and the calibration degree of the models were evaluated by the Brier score.The calibration curve of the best radiomics model was drawn,and the decision curve analysis(DCA)was also carried out.The integrated discrimination improvement(IDI)was calculated to quantitatively describe the overall improvement in the predictive efficiency of 3D-intratumoral radiomics model compared with 2D-intratumoral model.Results: Finally,132 patients were enrolled in the study,there were 74 cases in the complete response(CR)group,and 58 cases in the non-complete response(non-CR)group.In the training set and validation set,there was no statistically significant difference in age,the state of menstruation,FIGO stage,MTD and the lymph node state reported by MRI between CR group and non-CR group(P>0.05).The number of radiomics features extracted from 2D intratumoral-ROI,3D intratumoral-ROI and 3D peritumoral-ROI was all 1037,and the number of radiomics features of 3D intra-peritumoral combination was 2074.Finally,5,6,8 and 8 radiomics features were screened respectively.The difference of Rad-score in the training set and validation set was statistically significant(P<0.05).In the training set,the ROC curve showed that the values of area under the curve(AUC)of the four models ranged from 0.774(95%CI: 0.675 to 0.885)to 0.893(95%CI: 0.812 to 0.948).After internal validation of 1,000-iteration Bootstrap analysis,the AUC values of 2D-intratumoral model,3D-intratumoral model,3Dperitumoral model and the 3D-combined model were 0.772(95%CI: 0.680 to 0.869),0.860(95%CI: 0.779 to 0.935),0.847(95%CI: 0.771 to 0.925)and0.888(95%CI: 0.832 to 0.955),while in the validation,the AUC values were0.757(95%CI: 0.596 to 0.878),0.849(95%CI: 0.701 to 0.942),0.824(95%CI:0.671 to 0.926)and 0.887(95%CI: 0.747 to 0.965),respectively.And the Brier score showed that the four models had good calibration in the training set and validation set.The predictive efficacy of the combined model reached the highest.The F1-score,sensitivity,specificity,positive predictive value(PPV)and negative predictive value(NPV)in the training set were 0.875,84.3%,78.1%,82.7% and 80.0% respectively,and that in the validation set were 0.884,82.6%,94.1%,95.0% and 80.0%,respectively.The DCA showed that the combined model had high clinical application value.The IDI values were 0.155(P=0.027)and 0.179(P=0.039)in the training set and validation set,respectively.Compared with the 2D-intratumoral radiomics model,the predictive performance of 3D-intratumoral model was improved.Conclusion: The radiomics analysis based on the tumor space volume can fully explore the tumor heterogeneity.The radiomics features of intratumoral zones and peritumoral zones were correlated with the response to CCRT of cervical squamous cell carcinoma.The predictive model combined with intratumoral and peritumoral radiomics features can noninvasively quantitatively predict the response to CCRT,and provide guidance for clinicians to optimize the diagnosis and treatment strategies and risk stratification of patients with cervical cancer. |
PDF Full Text Request