Establishment And Evaluation Of Prediction Model Of Immunohistochemical Expression Of Hepatocellular Carcinoma By Radiomics And CT Signs

Objective: To investigate the value of imaging-based imaging in predicting the correlation between CT findings and expression of pathological immunohistochemistry CK19 hepatocellular carcinoma.Materials and methods: A retrospective analysis of 313 patients with surgical treatment and postoperative pathological confirmation of HCC in the first affiliated hospital of guangxi medical university from january 2015 to december 2016.Iclusions :2 weeks before operation CT enhancement examination,postoperative pathology contains pathological grade,Ki-67、CK19immunohistochemical examination and follow-up review data.Pathological group: CK19 expression was divided into CK19 positive groups(+),(2+),(3+)and CK19 negative groups(-),(+/-);Ki-67 results were divided into : <10%,0;10-25%,1(+);25-50%,2(2+);>50%,3(3+);The degree of tumor differentiation was graded according to Edmondson-Steiner pathological grading criteria.All the imaging features of the lesions were collected: the number of lesions,maximum diameter,capsule,margin,enhancement mode and CT value of each stage were observed.The enhancement pattern was divided into four groups :0(arterial phase enhancement),1(portal phase enhancement),2(arterial phase,portal phase enhancement),3(arterial phase,portal phase no obvious enhancement).Radiomics analysis: Application of radcloud3.0 software to extract 7 categories of a total of 107 imaging features,and through the LASSO algorithm to screen out the best correlation features,using SVM to provide modeling,using cross-validation method for testing,using ROC curves to evaluate the model prediction effectiveness,and add statistically significant imaging signs combined modeling to evaluate diagnostic efficacy.Results: There were 263 cases in male and 50 cases in female.CK19 positive group 65,CK19 negative group 248.There was no significant difference in gander and age between CK19 positive and negative groups(P >0.05).The relationship between follow-up CK19、Ki-67 and pathological grade and postoperative recurrence was statistically significant in CK19 positive group compared with CK19 negative group in one year and three years after operation(P =0.015;0.036).The recurrence rate within one year and three years after operation rises when the Ki-67 grade go higher,and the statistical difference was significant(P =0.028;P =0.042).The recurrence rate within one year and three years after operation rises when the pathological grade go higher,the difference in recurrence rate in one year was statistically significant(P=0.045),and the difference in recurrence rate in three years was not statistically significant(P =0.302).The relationship between CK19 expression and Ki67、pathological grade was analyzed,the statistical difference was not significant(p=0.093;P=0.052).The ROC curve was used to analyze the diagnostic efficacy of three pathological indexes in the diagnosis of recurrence within one year and three years was CK19(AUC 0.556 95% CI : 0.491-0.621 sensitivity 0.273 specificity 0.840),Ki-67(AUC 0.568 95% CI :0.504-0.633 sensitivity 0.477 specific 0.624),pathological grade(AUC 0.557 95% CI : 0.492-0.621 sensitivity 0.311 specificity 0.773),three indicators combined(AUC 0.60295%CI :0.538-0.665 sensitivity 0.492 specificity 0.680).Diagnostic efficacy of recurrence within three years was CK19(AUC 0.548 95%CI:0.485-0.612 sensitivity 0.250 specificity 0.847,Ki-67(AUC 0.563 95%CI : 0.499-0.627 Sensitivity 0.563 specific 0.466),two indicators combined(AUC 0.573 95%CI : 0.510-0.636 sensitivity 0.563 specific 0.540).The relationship between preoperative imaging findings and CK19 expression was not statistically significant between the CK19 positive group and the negative group in the number of lesions,lesion diameter,necrosis,boundary,and torticular thickening of tumor blood vessels(P > 0.05),but the differences in false capsule,hemorrhage,and enhancement pattern were statistically significant(P =0.000;P= 0.015;P =0.015),and ROC curve was used to analyze the pseudocapsule of diagnostic efficacy(AUC 0.614 95% CI :0.543-0.685 sensitivity 0.877 specific0.351),bleeding(AUC 0.540)95% CI :0.458-0.622 sensitivity 0.108 specific0.971),reinforcement(AUC 0.551):95% CI : 0.467-0.634 sensitivity 0.551 specific 0.875),combined pseudocapaculum,reinforcement method co-assessment(AUC 0.676 95% CI :0.604-0.748 sensitivity 0.738 specificity0.738),combination of three signs(AUC 0.686 95% CI : 0.613-0.759 sensitivity 0.738 specificity 0.532).The median CT values of plain scan,arterial phase,portal phase and equilibrium phase were lower than those of CK19 negative group,and the difference was statistically significant(P =0.000;P=0.007)in plain scan phase and portal phase(0.05).The LASSO algorithm was used to select 13 imaging features,construct model test,analyze diagnostic efficiency by ROC curve,and train set(AUC 0.684 95% CI : 0.61-0.76 sensitivity 0.66 specific 0.65),validation set(AUC 0.597 95% CI :0.44-0.76 sensitivity 0.54 specific 0.57).Using ROC curve to analyze diagnostic efficacy,training set(AUC 0.883 95% CI :0.82-0.94 sensitivity 0.80 specificity 0.80),validation set(AUC0.703 95% CI : 0.57-0.84 sensitivity 0.70 specificity0.77).Conclusion: CK19 、 Ki67 is HCC independent risk factor of recurrence within one year and three years after operation.The pathological grade is only valuable for predicting recurrence within one year after operation.CT enhancement scan without incomplete false capsule,bleeding,strengthening methods all have the prompt effect to the CK19 positive HCC,each stage CT value also has the certain prompt effect.A predictive model based on imaging features has some feasibility to differentiate CK19 expression,but the diagnostic efficacy is not obvious compared with imaging findings.Combined imaging features and imaging findings have better diagnostic value.