CT-based Radiomics Models For Predicting Portal Veins Pressure And Gastroesophageal Varices Bleeding In Cirrhotic Patients

Background and purpose:Gastroesophageal varices bleeding(GVB)is the most severe complication of portal hypertension(PH)in cirrhosis,accurate diagnosis and monitoring of portal hypertension is difficult in clinical practice.The aim of the study is to develop a noninvasive predictive model of portal vein pressure(PVP)values based on contrast-enhanced CT radiomics features and predict the risk of first spontaneous GVB in cirrhotic patients.Methods:From November 2014 to November 2019,204 consecutive patients with cirrhosis and portal hypertension underwent transjugular intrahepatic portosystemic shunt(TIPS)procedure in our liver center were included for the development of a predictive model.Radiomics PVP(rPVP)models based on liver,spleen and combined features were established on the above subjects.Radiomics features were extracted from each ROI and reduced via the least absolute shrinkage and selection operator(LASSO)regression to achieve an optimal predictive formula.A validation cohort of 100 patients diagnosed cirrhosis with or without first GVB were used to predict the risk of GVB.The association between rPVP and first GVB were observed.Results: Three separate predictive formulas forPVP were derived from radiomics features.In the training cohort,the rPVP calculated by the formulas were close to the real PVP measured during the TIPS procedure of 41.4cm H2O(P>0.19).In the validation cohort,the rPVP of patients with GVB were greater than that of patients without GVB in all three models.However,only the model based on liver features showed statistically significant between the two groups(P=0.042),while the rPVP calculated by the spleen based model and the combined model showed no statistically significant between the two groups with or without GVB(P=0.120 and P=0.809).Conclusion: PVP can be accurately predicted by a noninvasive,contrast-enhanced CT derived radiomics model.rPVP serves as a noninvasive and an auxiliary indicator for predicting the risk of first GVB for the patients with cirrhosis.