Related Risk Factors And Artificial Intelligence Prediction Of Microvascular Invasion In Hepatocellular Carcinoma

Objectives: Hepatocellular carcinoma(HCC)has a high recurrence rate after surgery,and microvascular invasion(MVI)is believed to be an important risk factor for intrahepatic metastasis and postoperative recurrence of HCC.This study aims to:(1)Analyze the effect of MVI on the prognosis of HCC,and explore the related risk factors of MVI;(2)Establish MVI prediction model based on artificial intelligence(AI)and digital pathology technology.Methods:(1)Clinicopathologic data of 553 HCC patients who received radical hepatectomy at The Affiliated Hospital of Qingdao University from January 2014 to December 2018 and 89 HCC patients who underwent radical hepatectomy at Tsinghua Chang Gung Hospital in Beijing from October 2014 to October 2019 were retrospectively collected.The patients were divided into MVI(+)group(n=265)and MVI(-)group(n=377)according to the results of pathological examination.The collected data included: gender,age,neutrophils,lymphocytes,platelets(PLT),prothrombin time(PT),international normalized ratio(INR),albumin(Alb),total bilirubin(Tbil),alanine aminotransferase(ALT),aspartate aminotransferase(AST),γ-glutamyltransferase(GGT),serum alpha-feto protein(AFP),hepatitis B surface antigen(HBs Ag),neutrophil-lymphocyte ratio(NLR),platelet-lymphocyte ratio(PLR),surgical approach,tumor number,largest diameter,the Edmondson-Steiner grade,liver capsule invasion,satellite lesions,liver cirrhosis and bile duct tumor thrombus(BDTT).Kaplan-Meier curve and Log-rank test were used to analyze the effect of MVI on the prognosis of HCC patients.Univariate analysis(including independent sample t test,nonparametric Mann-Whitney U rank sum test and Pearson chi-square test)was used to evaluate the difference of each variable between the two groups.The variables with statistical differences were included in the Logistic regression analysis to explore the independent risk factors of MVI.(2)The clinicopathologic data and digital pathological images of 117 HCC patients who underwent liver surgery in The Affiliated Hospital of Qingdao University from January 2016 to July 2017 were retrospectively collected.The Densenet model was used for feature extraction and Lasso regression for feature selection,then we screened for pathological features related to MVI in order to develop and established a MVI prediction model and finally we achieved accurate assessment of MVI.Results:(1)Pathological examination confirmed the existence of MVI in 265patients(41.3%).Edmondson-Steiner grade(p < 0.001),liver capsule invasion(p =0.003),BDTT(p = 0.005),AFP(p = 0.041),tumor size(p = 0.005)and NLR(p = 0.015)were independent risk factors for MVI and the area under the receiver operating characteristic(ROC)curve was 0.743(p < 0.001),indicating that the model has significant clinical utility.(2)The accuracy rates of the MVI prediction model based on AI in the training cohort and the test cohort are about 87.65% and 86.11%,respectively,showing the great potential of this model in predicting MVI for HCC patients.Conclusions:(1)HCC combined with MVI indicated poor prognosis,and the independent risk factors for MVI were Edmondson-Steiner grade,liver capsule invasion,BDTT,AFP,tumor size and NLR.The application of these indicators can provide reference for the prediction of preoperative MVI,which has guiding significance for the formulation of treatment decisions for patients with high recurrence risk.(2)The MVI prediction model constructed by combining the advantages of AI and pathomics shows good prediction performance.The model can accurately identify patients with high MVI risk and provide the basis for individualized treatment.