Establishment And Validation Of Liver Parenchyma Features Based Prognostic Models Of Hepatectomy

Post-hepatectomy liver failure(PHLF)represents a postoperative dysfunction in the liver’s synthesis,excretory and detoxification functions.It increases morbidity and mortality after liver resection,especially in patients with advanced liver fibrosis and cirrhosis.Cirrhosis is not only related to the occurrence of PHLF,but causes a coagulation disorder that increases the risk of increased blood loss and the need for blood transfusion during surgery.In China,over 80% of hepatocellular carcinoma(HCC)cases are related to hepatitis B virus(HBV)infections that can cause liver dysfunction and chronic fibrosis.In preoperative evaluations for hepatectomy,liver function and the degree of liver fibrosis are commonly considered.Liver fibrosis biomarkers like serum biomarkers and fibrosis models are used alone or in combination to determine the degree of liver fibrosis,and predict PHLF.Currently,liver stiffness measurements(LSM)using imaging methods including transient elastography(TE)and magnetic resonance elastography(MRE)have been reported as good tools to diagnose liver fibrosis,and have a predictive effect on the occurrence of PHLF.Two-dimensional shear wave elastography(2D-SWE)has been reported to have better fibrosis diagnosis accuracy than real-time tissue elastography.However,the predictive accuracy of 2D-SWE for PHLF has not been discussed yet.According to principle,LSM results reflect the hardness of the liver.However,in clinical work,we found that the LSM results of some patients were inconsistent with the findings of palpation by surgeons.When we reviewed the stiffness measured by 2D-SWE and fibrosis pathology in hepatectomy patients at our center,we found that a group of patients had high-grade fibrosis but a low level of stiffness.This led to situations in which cases were converted to open surgery after visualization under laparoscopy.For patients with a high degree of stiffness but a low fibrosis grade,open surgery may be avoided.Liver hardness measured by the durometer has been shown to be correlated with the degree of liver fibrosis,and to be a potential parameter for predicting PHLF.However,these studies did not analyze the predictive accuracy of hardness for PHLF.Furthermore,the correlation between liver hardness measured by the durometer and liver stiffness measured by 2D-SWE has not been discussed yet.Delayed recovery of liver function is defined as "serum bilirubin> 50 μmol / L or PT> 20 s at any time point within 1 to 5 days after surgery",which increases mortality after liver resection.However,the correlation between liver hardness and postoperative delayed recovery has not been discussed yet.In order to evaluate the correlation between liver hardness and postoperative liver dysfunction,predictive accuracy of 2D-SWE-measured stiffness and durometermeasured liver hardness for PHLF and postoperative delayed liver function recovery was analyzed.A preoperative linear regression model was constructed to improve 2D-SWE’s ability to predict liver hardness.The model’s predictive ability for postoperative liver dysfunction was confirmed in the validation cohort.As a new method of medical image analysis,radiomics provides new ideas for tumor treatment and prognosis.Therefore,this study used machine learning to establish a prediction model for HCC recurrence by delineating the liver parenchyma,tumor and spleen in the preoperative enhanced CT images of HCC patients and extracting features.In conclusion,2D-SWE showed a good correlation with durometer-measured liver hardness.The hardness scale model based on liver stiffness,HBs Ag and albumin could predict liver hardness before surgery.Additionally,the model and liver hardness demonstrated PHLF predictive ability.At the same time,a predictive model of early recurrence of liver cancer after surgery was constructed based on CT image features.These models enable surgeons to have a more intuitive understanding of the liver before surgery,providing a reference for perioperative management.Part I Predictive effect of two-dimensional shear wave elastography on post-hepatectomy liver function recoveryAims:To further study and evaluate the significance of 2D-SWE in preoperative evaluation and prognosis of liver resection to better guide perioperative treatment and plan surgery.Methods:A total of 162 patients with hepatectomy were retrospectively included in this study.2D-SWE was used to measure liver stiffness and assess liver fibrosis before surgery.The receiver operating curve(ROC)and the area under the curve(AUROC)were used to evaluate the predictive accuracy of 2D-SWE for postoperative liver dysfunction.Univariate and multivariate analyses by logistic regression were used to determine liver function risk factors for PHLF and delayed recovery.Results:The AUROC of stiffness measured by 2D-SWE for PHLF was 0.737(p=0.002).However,2D-SWE showed no predictive ability for delayed recovery of liver function.The optimal cutoff value of the stiffness was 11.5 kPa(sensitivity=0.737,specificity= 0.629).Patients with a stiffness>11.5 kPa had significantly higher risks of PHLF with hazard ratios of 3.710(95% CI 1.389-9.910,p=0.009).Conclusion:Liver stiffness assessed by 2D-SWE is an independent parameter to predict PHLF.Reducing the volumes of removed liver is considerable to decrease the morbidity of PHLF in patients with stiffness>11.5 kPa.Part II Model for liver hardness using two-dimensional shear-wave elastography,durometer,and preoperative biomarkersAims:Liver hardness measured by durometer has been proved that correlated with the degree of liver fibrosis and was a parameter for predicting post-hepatectomy liver failure(PHLF).We aim to measure correlations between 2D-SWE and durometer-measured objective liver hardness and construct a preoperative liver hardness linear regression model.Methods:Seventy-four hepatectomy patients with liver hardness were enrolled in a derivation cohort.Hardness was measured using a durometer and 2D-shear wave elasticity(SWE).Tactile-based liver hardness scores(0–100)were given by surgeons after palpating the liver tissue of the surgical field.Correlation coefficients for durometer-measured hardness and preoperative parameters were calculated.Multiple linear regression models were constructed to select the best predictive durometer scale.Receiver operating characteristic(ROC)curves were used to calculate the best model’s prediction of PHLF and post-hepatectomy delayed liver function recovery;univariate and multivariate analyses by logistic regression were used to determine risk factors for postoperative liver function efficiency.A separate validation cohort(n=162)was used to evaluate the model.Results:The stiffness measured using 2D-SWE had good linear correlation with durometer-measured hardness(Pearson rank correlation coefficient 0.704,p<0.001).The best multiple linear regression model for durometer scale(hardness scale model)was based on stiffness,hepatitis B virus surface antigen,and albumin level and had an r2 value of 0.580.The area under the ROC for the durometer and hardness scale for PHLF prediction were 0.807(p=0.002)and 0.785(p=0.005),respectively.The optimal cut off value of the durometer and hardness scale was 27.38(sensitivity=0.900,specificity=0.660)and 27.87(sensitivity=0.700,specificity=0.787),respectively.Patients with a hardness scale>27.87 had significantly higher risks of PHLF with hazard ratios of 7.835(95% CI 1.486-41.306,p=0.015).The model’s PHLF predictive ability was confirmed in the validation cohort.Conclusion:Liver stiffness assessed by 2D-SWE correlated well with durometer hardness values.The multiple linear regression model predicted durometer hardness values and postoperative liver function efficiency.Part III CT-based radiomics signatures to predict early recurrence in HBV related hepatocellular carcinoma after curative tumor resectionAims:To study the correlation of radiomics signatures analysis based on contrastenhanced CT and post-hepatectomy early recurrence in hepatocellular carcinoma(HCC).Methods:Seventy-two patients with HBV related HCC were retrospectively analyzed.Radiomics features was extracted from the portal-phase CT images of HCC,spleen and liver parenchyma.Random Forest(RF)algorithm was used to select features.Support Vector Machine(SVM)predictive model was developed in training set according to radiomics signatures of porta-phase.Receiver operating characteristic curve(ROC)was used to determine the performance of predicting model.Results:The predicting model yielded an area under ROC of 0.998 with sensitivity and specificity of 0.982,0.9 and 1,respectively.Conclusion:The machine learning predictive model developed by portal-phase CT radiomics feature can effectively predict of early recurrence in HBV related HCC.