Establishment of CT Radiomics Model for Predicting Microvascular Invasion,Postoperative Outcome And Recurrence in Patients with Liver Cancer
Objective To establish CT radiomics models for predicting microvascular invasion,postoperative outcome and recurrence and to evaluate its predictive value in patients with liver cancer.Methods 360 patients from January 2015 to January 2017with early liver cancer who were diagnosed and underwent hepatectomy in our hospital were divided into training group(240 cases)and validation group(120 cases)according to the random number table method.The pre-operation CT image data and pathological data of the two groups were analyzed,the three-dimensional region of interest were obtained by outlining layer by layer,and the image features were extracted.The lasso(least absolute shrinkage and selection operator)regression was used to reduce the dimension and ten fold cross validation was conducted to obtain the CT features and establish the prediction model,Receiver operating characteristic curve(ROC)was used to determine the predictive value of the prediction model.Results There was no significant difference between the training group and the validation group in gender,age,hepatitis B surface antigen,alt,AST,GGT,TBIL,DBIL,AFP,maximum tumor diameter,child Pugh grade,tumor microvascular invasion(P＞0.05).For microvascular invasion,a total of 10 specific characteristic parameters were obtained.The area under the ROC curve(AUC)of the corresponding model in the training group was 0.705(95％CI:0.640～0.770),with a sensitivity of 71.42％and a specificity of 58.97％;The AUC in the validation group was 0.745(95％CI:0.679～0.810),with a sensitivity of 75.00％and a specificity of 66.17％.For the recurrence of 3 years after operation,a total of 13 specific characteristic parameters were obtained.The AUC of the corresponding model in the training group was 0.720(95％CI:0.629～0.812),with a sensitivity of 72.22％and a specificity of 70.49％;In the validation group,the AUC was 0.753(95％CI:0.666～0.839),the sensitivity was 70.27％,and the specificity was 69.86％.Conclusion CT radiomics model has high predictive value for microvascular invasion and postoperative recurrence of liver cancer.
万方数据
维普
