Purpose Explore the predictive value of nomograms based on enhanced CT radiomics for invasiveness of thymic epithelial tumor.Materials and Methods The clinical and imaging data from 155 cases confirmed with thymic epithelial tumors at the First Affiliated Hospital of Wannan Medical College from January 2015 to January 2023 were retrospectively collected.All cases were randomly divided into training(n=108)and validation(n=47)groups in a 7∶3 ratio.The radiomics features from venous phase images were extracted.The least absolute shrinkage and selection operator algorithm for dimensionality reduction were utilized to establish radiomics labels and calculate the Rad-score.Univariate and multivariate regression analyses were conducted to identify independent risk factors.Imaging feature models,Rad-score and imaging omics clinical combined model were constructed to plot the corresponding nomograms.The diagnostic performance and clinical benefits of the models were evaluated via receiver operating characteristic curves and decision curves.The DeLong test was applied to compare area under the curve differences between models and used calibration curves to assess nomograms calibration.Results 16 optimal image omics features were selected by dimensionality reduction.Logistic regression analysis showed that tumor morphology(OR=2.932,P=0.025),peripheral tissue invasion(OR=11.461,P=0.005)and Rad-score(OR=255.27,P=0.002)were independent risk factors.The area under the curve in the training set and the verification set were 0.852 and 0.831,respectively.Compared with the image feature model and Rad-score in the training set,the differences were statistically significant(Z=3.607,2.270,P<0.05).The threshold probability of the column chart model training set was between 0.08 and 0.88 for clinical benefit.Conclusion The combined model nomograms based on enhanced CT radiomics and clinical features can effectively predict thymic epithelial tumor invasiveness and assist clinicians in formulating precise treatment plans before surgery.
维普
万方数据
