Value of ultrasound radiomic model to the prediction of the positive expression of glypican-3 in patients with hepatocellular carcinoma
Objective To screen the ultrasound radiomic features of patients with hepatocellular carcinoma(HCC),to construct a predictive model for the positive expression of glypican 3(GPC3),and to explore its predictive value.Methods A total of 290 HCC patients underwent surgical treatment in Henan Provincial People's Hospital and the First Affiliated Hospital of Zhengzhou University from May 2019 to February 2023,among whom the postoperative immunohistochemical examination showed positive GPC3 in 219 patients and negative GPC3 in 79.According to the ratio of 8∶2,290 patients were randomly divided into the training set(n=232)and the testing set(n=58).The gender,age,Child-Pugh liver function classification,maximal tumor location,tumor diameter,alpha-fetoprotein,alanine aminotransferase,aspartate aminotransferase,alkaline phosphatase,glutamyl transpeptidase,albumin,total bilirubin,conjugated bilirubin,serum creatinine,prothrombin time,fibrinogen,international normalized ratio,and proportions of liver cirrhosis,splenomegaly,ascites,positive hepatitis B surface antigen/hepatitis C virus antibody,and positive GPC3 were compared between two sets.All patients received ultrasound examination to outline the area of liver lesions.A total of 1 046 ultrasound radiomic features were extracted from the sonographic images after normalization and standardization.Lasso regression was used to screen out 10 radiomic features which were highly related with the positive expression of GPC3.Machine learning random forest algorithm was used to establish a model for predicting the positive expression of GPC3 in HCC patients.ROC curve was plotted to evaluate the predictive efficiency of the prediction model on the positive expression of GPC3 in two sets.Results There were no significant differences in the gender ratio,age,Child-Pugh liver function classification,tumor location,maximal tumor diameter,alpha-fetoprotein,alanine aminotransferase,aspartate aminotransferase,alkaline phosphatase,glutamyl transpeptidase,albumin,total bilirubin,conjugated bilirubin,serum creatinine,prothrombin time,fibrinogen,international normalized ratio,and proportions of liver cirrhosis,splenomegaly,ascites,positive hepatitis B surface antigen/hepatitis C virus antibody and positive GPC3 between two sets(P＞0.05).The AUC of the prediction model in the training set for predicting positive GPC3 was 0.820(95％CI:0.758-0.883,P＜0.05),and the accuracy,specificity and sensitivity were 59.5％,84.2％and 51.4％,respectively.The AUC of the prediction model in the testing set was 0.700(95％CI:0.567-0.832,P＜0.05),and the accuracy,specificity and sensitivity were 63.8％,85.7％and 56.8％,respectively.Conclusion The ultrasound radiomic model constructed by machine learning random forest algorithm has a good predictive value for the positive expression of GPC3 in HCC patients.
hepatocellular carcinomapositive glypican-3ultrasound radiomic model
NETL
NSTL
万方数据
