Prediction Of Lymph Node Metastasis In Patients With Epithelial Ovarian Cancer Based On Enhanced CT Radiomics Combined With Clinical Indicators

Objective: To explore the value of enhanced CT radiomics combined with clinical indicators in preoperatively predicting lymph node metastasis of epithelial ovarian cancer(EOC).Methods: The clinical data,pathological reports and image files of 195 patients with epithelial ovarian cancer who underwent surgery in Shengjing Hospital of China Medical University from January 2015 to January 2021 were analyzed retrospectively,including 56 cases with lymph node metastasis and 139 cases without lymph node metastasis.According to the date of surgery,the patients were divided into a training set(n=145)and a test set(n=50).The region of interest(ROI)was segmented in 3D Slicer software.The radiomics features were extracted by the Pyradiomics package in Python and then preliminarily screened by the Mann?Whitney U test.Finally,the most relevant features were selected by Lasso regression to construct the Radscore and then combined with the statistically significant clinical indicators to build a combined mode.The goodness of fit of the model in the training set was validated by Bootstrap method and Hosmer-Lemeshow test,and the performance of the models was evaluated and compared by Receiver operating characteristic curve(ROC),Kappa test,Decision curve analysis(DCA)and Delong test.Results: 1153 radiomics features were extracted from preoperative enhanced CT images of epithelial ovarian cancer patients.Ultimately 13 features were identified to construct the Radscore.The AUC of Radscore in the training set was 0.923(95%CI:0.880～0.966),sensitivity 0.925,specificity 0.800,accuracy 0.834,Kappa 0.636,and0.884 in the test set(95%CI: 0.790～0.979),sensitivity 0.812,specificity 0.853,accuracy0.840,Kappa 0.644.CA125,human epididymis protein 4(HE4)and albumin(ALB)were significantly correlated with lymph node metastasis in epithelial ovarian cancer.CA125 and ALB were independent risk factors for lymph node metastasis.The AUC of the combined prediction model composed of Radscore,CA125 and ALB was 0.931 in the training set(95%CI: 0.891～0.972),sensitivity 0.875,specificity 0.848,accuracy0.855,Kappa 0.666,and 0.888 in the test set(95%CI: 0.794～0.982),sensitivity 0.938,specificity 0.765,accuracy 0.820,Kappa 0.629.Conclusion: The combined prediction model based on enhanced CT radiomics and clinical indicators to predict lymph node metastasis of epithelial ovarian cancer before surgery is a feasible and accurate noninvasive prediction tool.