Preoperative Prediction Of Microvascular Invasion Of Single Hepatocellular Carcinoma With Deep Learning Model&CT Features From Contrast Enhanced CT

Part1 Preoperative prediction of microvascular invasion for single hepatocellular carcinoma with features of CT from enhanced CTObjective:To assess the value of CT feature from contrast enhanced CT in predicting micro'vascular invasion(MVI)of single hepatocellular carcinoma(HCC).Materials and Methods:The clinical and imaging data of 1116 patients with single hepatocellular carcinoma who undergoing radical resection were retrosp ectively collected,including sex,age,albumin,lymphocyte,neutrophil,platelet,erythrocyte,leukocyte,alkaline phosphatase,aspartate aminotransferase,glutama te transaminase,total bilirubin,y-glutamyltransferase and hepatitis B virus DN A levels,prothrombin time,alpha-fetoprotein level,carbohydrate antigen 199;I maging features included tumor diameter,margin,capsule,internal arteries,ne crosis,hypodense halo,peritumoral enhancement,satellite nodules and arterial enhancement modes.Univariate analysis was used to incorporate the data with statistical significance into multivariate logistic regression model.The model w as analyzed by multiple factors,and the ROC curve was constructed to explo re its diagnostic value.Result:Univariate analysis showed that tumor diameter,internal arteries,hypod ense halo,capsule,margin,peritumoral enhancement,necrosis,arterial enhance ment modes,satellite nodules,PLT,AFP,hepatitis B DNA(P<0.05)were r isk factors for MVI.Multivariate logistic regression analysis showed that necr osis,internal arteries,hypodense halo,tumor margin,capsule and satellite nodul es(P<0.05)were independent risk factors for predicting MVI.ROC curve an alysis showed that the sensitivity,specificity and AUC were 0.777,0.704 and 0.803 with combined signs of above features.Conclusion:Preoperative contrast-enhanced CT is valuable in predicting micro vascular invasion of single hepatocellular carcinoma.Part2 Preoperative prediction of microvascular invasion for single hepatocellular carcinoma with deep learning model from enhanced CTObjective:To assess the value of deep learning model from contrast enhance d CT in predicting microvascular invasion(MVI)of single hep-atocellular carci noma(HCC).Materials and Methods:CT imaging results of 1116 patients with singlehepa tocellular carcinoma who undergoing radical resection were retrospectivey colle cted.The images of arterial phase and portal vein phase of dynamic enhanced CT were preprocessed,and the required focus layers and coordinateswere ma rked.The deep learning neural network was constructed according to the path ological MVI as the gold standard.The average probability obtained was the probability of predicting MVI.Result:80%of cases were used for training model,20%for testing model,AUC value of arterial phase model was 0.810 in testing group,AUC value o f portal phase model was 0.814 in testing group,and AUC value of combine d two-phase model was 0.832.Conclusion:Deep learning neural network model based orn preoperative contra st enhanced CT is valuable in predicting MVI of single hepatocellular carcino ma,the diagnostic efficiency of two-phase combined model is higher than any other phase.