| Objective:This study was conducted in order to investigate the value of computed tomography(CT)-based radiomics signatures for the preoperative prediction of the degree of pathological differentiation of hepatocellular carcinoma(HCC).Methods: A retrospective study of the complete data from 110 patients diagnosed with HCC by surgical pathology in the First Affiliated Hospital of Wannan Medical College from January 2014 to December 2020,According to the 2010 edition WHO standard,the patients were divided into well differentiated group and moderately-poorly differentiated group(46 cases in well differentiated group and 64 cases in moderately-poorly differentiated group),the patients were randomly divided into training group(78 cases)and test group(32 cases)at a ratio of 7:3.The Radi Ant DICM Viewer(64-bit)software was used to screen and save the plain scan and enhanced CT images of the upper abdomen or the whole abdomen of the patients,the ITK-SNAP(version 3.6.0)software was used to manually drawn on the region of interest(ROI)layer by layer and generated the region of interest volume(VOI),imported the images of each period and the corresponding VOI information into the Artificial Intelligence Kit(A.K.)software developed by GE company in the United States,and extracted the feature parameters of each period.The maximal relevance and minimal redundancy(m RMR)、least absolute shrinkage and selection operator(LASSO)methods were used for data dimension reduction to choose the best radiomics signatures subset,logistic regression method was used to establish clinical feature model,radiomics model of each phase,combined radiomics model and combined diagnostic model.The reliability of the model was tested with 100-folds leave group out cross validation(LGOCV),receiver operating characteristic(ROC)curves were drawn in the training group and the test group to evaluate the predictive ability of the model,and the goodness of fit of the model was tested with Hosmer-Lemeshow.Decision curve analysis(DCA)was used to evaluate the clinical application value of the model.Results: 1.In clinical characteristics,only AFP has statistically significant between the well differentiated group and moderately-poorly differentiated group in the training group and the test group.The AUC(95%CI)of the clinical feature AFP logistic regression prediction model in the training group and test group are respectively0.649(0.528-0.771)and 0.761(0.578-0.944).2.Based on CT unenhanced、arterial phase、portal phase、delay phase and combined,10、11、11、12、13 features were screened out,and radiomics models of each phase were established respectively.The area under the curve(AUC)(95%CI)of the degree of hepatocellular carcinoma pathological differentiation predicted by the radiomics model of unenhanced,arterial phase,portal phase and delayed phase in the training group and test group were 0.833(0.744-0.922)and 0.806(0.641-0971)、 0.892(0.822-0.962)and0.777(0.616-0.938)、0.872(0.807-0.957)and 0.846(0.699-0.994)、0.863(0.780-0.945)and 0.826(0.674-0.978).The efficiency of the combined radiomics signatures of each phase was better than that the individual phases,the area under the curve(AUC)(95%CI)in the training group and the test group were 0.905(0.836-0.974)and0.858(0.731-0.985)respectively.The degree of integration was better,and the decision curve analysis shows that the clinical application value is high.3.The predictive performance of the combined diagnostic model is further improved.The area under the curve(AUC)(95%CI)in the training group and the test group were0.915(0849-0.982)and 0.874(0.753-0.996),respectively.Conclusion: Based on CT arterial phase radiomics model is better than portal phase、delayed phase and unenhanced.The combined diagnostic model better than the combined radiomics model.Based on CT radiomics is expected to provide clinicians with a new solution for rapid and non-invasive diagnosis the degree of pathological differentiation in hepatocellular carcinoma before surgery. |
PDF Full Text Request