Establishment And Validation Of Predictive Model Of Microvascular Invasion In Hepatocellular Carcinoma By Immune And Radiomics

Objective: Microvascular invasion(MVI)plays an important role in the treatment and prognosis of hepatocellular carcinoma(HCC),but there is a lack of effective methods to predict MVI.In this study,we propose to integrate the Tumor Immune Microenvironment(TIME),imaging characteristics and clinical data of HCC patients to establish a prediction model of MVI.Methods: A retrospective study method was used to select 135 patients with HCC,randomly divided into two cohorts according to the ratio of 7:3,the training set(n=95)and the validation set(n=40),and 42 patients with HCC were selected as the external validation set at an external hospital.CT scans were performed in all patients to assess tumor diameter,number,intratumoral arteries,peritumoral enhancement,tumorliver differences,internal arteries,tumor margins,tumor capsule,enhancing tumor capsule and hypoattenuating halo.The 3D-Slicer software was used to outline region of interest in the arterial and venous phases,and the radiomics features were extracted by this software.The positive expression of CD68,CD163,S100A12,CD33 and CD1 a in HCC was detected by immunohistochemical staining,and the number of corresponding immune cells was calculated by in Form2.4.0 software.Independent risk factors for MVI were screened by multifactorial regression analysis.Three models for predicting MVI based on independent risk factors(model 1: immune molecules,CT signs;model 2:immune molecules,CT signs,intratumoral CT radiomics features;model3: immune molecules,CT signs,intratumoral and peritumoral CT radiomics features)were established and validated in internal and external cases.The predictive performance of the models was assessed by subject operating characteristic(ROC)curves,calibration curve analysis,and decision curve analysis(DCA),and the AUC was compared by Delong test.Results: Among all patients,immunohistochemical results showed that CD68,CD163 and S100A12 had high expression in MVI with statistical differences(p<0.05).In addition,the degree of differentiation of HCC,AFP,ALT,AST,platelet sorting count to lymphocyte sorting count ratio,CT signs of tumor diameter greater than 5 cm,presence of internal arteries in the portal venous stage,unsmooth tumor margins,presence of capsule and radiological enhancement of capsule were different in patients with MVI(p<0.05).Further,multifactorial regression analysis showed that CD68,CD163,S100A12,tumor diameter,capsule enhancement,2 intratumoral features,and 3 peritumoral features were independent risk factors for MVI.The AUCs of model 1,model 2 and model 3 were 0.864(95% CI: 0.797-0.923),0.927(95% CI: 0.875-0.969)and 0.930(95% CI: 0.881-0.969)in the training set,0.755(95% CI :0.623-0.875),0.902(95% CI : 0.819-0.968)and 0.900(95% CI : 0.816-0.966)in the validation set.The AUCs of model 2 and model 3 in external cases were 0.905(95% CI : 0.805-0.981)and 0.952(95% CI :0.880-1.000)in the training set and 0.667(95% CI : 0.375-0.929)and0.905(95% CI : 0.722-1.000)in the validation set.Conclusion: Combining immune molecules,CT signs and intratumoral and peritumoral CT radiomics features to establish a predictive model of HCC MVI has high predictive performance,which can provide a basis for the selection of treatment options for liver cancer patients.