Assessment Of CT-based Radiomics And Nomogram In The Classification Of Common Ovarian Epithelial Tumor Subtypes

[Objective]To investigate the value of enhanced CT-based radiomic features and nomogram combined with risk factors in the preoperative subtype classification of common ovarian epithelial tumors.[Materials and methods]A total of 255 patients,from January 2016 to June 2021 at Shandong University Qilu Hospital,with pathology-confirmed ovarian serous and mucinous cystadenoma,borderline tumor and cancer were retrospectively enrolled.All of these patients underwent contrastenhanced CT preoperatively,which including pelvic.Additionally,age,CA125,and number of pregnancies were collected.The preoperative CT images were exported in DICOM format.According to pathologic results,patients were divided into two groups:serous cystadenoma and mucinous cystadenoma,borderline tumors and cancer,all randomized in a 7:3 ration into training set and validation set.Appropriate significance analysis methods were selected according to the type of data,and variables with statistical differences were included in univariate and multivariate logistic regression,then risk factors were obtained.3D Slicer software(V 4.11)was used to delineate the region of interest(ROI)from venous phase CT images.After that,data was uploaded to the Radcloud platform(Huiying Medical Technology Co.,Ltd)to extract radiomic features,and a total of 1409 features were obtained.Then,the variance threshold,selectKbest and least absolute shrinkage and selection operator(LASSO)were used to select the optimal features step by step.Risk factor models(Risk)were developed based on risk factors,and radiomic models(Radscore)were developed using the selected radiomic features,and the two were combined to construct nomogram.The calibration curve and decision curve analysis(DCA)were used to evaluate nomograms.And other parameters,such as receiver operator characteristic(ROC)curve,sensitivity and specificity were used to evaluate the predictive power of models.[Results]For differentiating serous cystadenoma and mucinous cystadenoma,8 radiomic features and 2 CT-based risk factors were ultimately selected.In the training set,the Risk model predicted mucinous cystadenoma with an area under the ROC curve(AUC)of 0.797(95%confidence interval[CI]:0.683 to 0.911),a sensitivity of 0.750 and a specificity of 0.867,the Radscore model had an AUC of 0.846,(95%CI:0.743 to 0.948),a sensitivity of 0.875 and a specificity of 0.800,and Nomogram had an AUC of 0.900,(95%CI:0.814 to 0.986)and a sensitivity 0.875 and specificity 0.867.In the validation set,the Risk model predicted mucinous cystadenoma with an AUC of 0.786(95%CI:0.615 to 0.956),a sensitivity of 0.733 and a specificity of 0.786,and the Radscore model with an AUC of 0.800(95%CI:0.637 to 0.963),a sensitivity of 0.867 and specificity 0.643,Nomogram had an AUC of 0.886,(95%CI:0.742 to 1.000),sensitivity 0.867 and specificity 0.929.In addition,the calibration curve and decision curve analysis showed that Nomogram has better differential diagnostic value.For borderline tumors and cancer,11 radiomic features and 3 risk factors were ultimately selected.In the training set,the AUC of the Risk model predicted cancer was 0.781(95%CI:0.691 to 0.872),a sensitivity of 0.647 and a specificity of 0.862,the Radscore model had an AUC of 0.875,(95%CI:0.805 to 0.945),a sensitivity of 0.812 and a specificity of 0.862,and Nomogram had an AUC of 0.914,(95%CI:0.850 to 0.977)and a sensitivity of 0.871 and specificity 0.862.In the validation set,the Risk model predicted cancer with an AUC of 0.820(95%CI:0.701 to 0.940),a sensitivity of 0.568 and a specificity of 1.000,and the Radscore model with an AUC of 0.736(95%CI:0.582 to 0.890),a sensitivity of 0.514 and specificity 0.923,Nomogram had an AUC of 0.825,(95%CI:0.696 to 0.954),sensitivity 0.811 and specificity 0.692.In addition,the calibration curve and clinical decision curve analysis showed that Nomogram has better differential diagnostic value.[Conclusion]1.The number of cysts can be used to differentiate ovarian serous cystadenoma from mucinous cystadenoma.Mucinous cystadenoma is more common with multiple cysts compared to serous cystadenoma.2.Age,ascites and CA125 are of diagnostic value in differentiating ovarian serous and mucinous borderline tumors from ovarian cancer.Patients with ovarian cancer are older than those with borderline tumors,and ovarian cancer is more likely to be combined with ascites and elevated CA125(≥95 U/mL).3.Enhanced CT-based radiomic features can be used to subclassify ovarian serous and mucinous tumors with better efficacy than risk factor models.4.A nomogram combining the radiomic features and risk factors is a convenient and effective way to predict the type of ovarian serous and mucinous tumor preoperatively.