CT-based Radiomics Model For Evaluating Graft Function After Liver Transplantation

Objective:Liver transplantation is a key treatment for end-stage liver disease.However,due to the occurrence of poor graft function after liver transplantation,the patient’s prognosis is poor,which affects the success rate of liver transplantation and the waste of liver resources.To explore the predictive value of CT imaging features in transplant function prognosis after liver transplantation,and to establish a prediction model and radiomics nomogram.To accurately predict transplant function based on clinical and radiomics model,and to provide evidence for treatment.Methods:The clinical data and imaging data of 136 recipient patients who underwent classic orthotopic liver transplantation at the Third Medical Center of the PL A General Hospital from January 2013 to December 2018 were retrospectively analyzed.The patients received enhanced CT scans after the operation,and ROI delineation and image feature extraction were performed using ShenRui medical software.The R 3.6.2 statistical analysis software was used for analysis,and the least absolute contraction selection operator LASSO-logistic regression model was used to reduce the dimensionality,feature selection,and construction of the imaging omics data using the 10-fold cross-validation method.The penalty parameters are optimized to select the most predictive radiomics features.Univariate and multivariate Logistic regression analysis methods were used to screen valuable clinical feature information,and clinical feature prediction models,radiomics prediction models,radiomics+clinical indicators combined prediction models were established.Then establish the nomogram of the clinical model and the combined model.Compare the performance of radiomic in predicting graft function and patient prognosis after liver transplantation with the results of other related predictive scores.Validate and compare the predictive ability of the new model established in conjunction with imaging omics.Results:1.A total of 136 cases underwent classic orthotopic liver transplantation.There were 106 males and 30 females;67 patients were younger than 50 years old,69 were equal or older than 50 years old;17 patients had BMI less than 18.5,51 patients were between 18.5 and 24,and 68 patients were greater than 24.2.Through univariate analysis and multivariate analysis,it was found that preoperative total protein(OR:1.677;95%CI:1.052-2.678),postoperative TBIL(OR:0.981;95%CI:0.956-1.006),postoperative CR(OR:1.023;95%CI:1005-1.040)and ALT(OR:1.002;95%CI:1.000-1.004)are independent predictors for predicting graft function or poor survival of patients after liver transplantation.The clinical prediction model was established,and the prediction index C-index was 0.884(95%CI:0.793-0.975).By delineating the advanced CT images of arteries,extracting and screening imaging omics features related to prediction.Each patient has 1227 candidate imaging omics characteristics.After LASSO-logistic regression cases are included for variable screening,a total of 7 imaging omics characteristics are screened out,include"wavelet.HLH_glcm_Imcl","wavelet.LLH_glcm_JointEnergy","wavelet.LHL_glcm_InverseVariance","wavelet.LLL_gldm_DependenceVariance","wavelet.HHL_glszm_SizeZoneNonUniformityNormalized","wavelet.HLH_gldm_LargeDependenceLowGr ayLevelEmphasis","log.sigma.5.0.mm.3D_glszm_SizeZoneNonUniformityNormalize d",name them as featurel-7,and establish a RAD-score=3.5044*featurel+3.1687*feature 2-7.4697*feature 3+4.6552*feature 4+0.5194*feature 5+3.4254*feature 6-1.5176*feature 7.3.The combined prediction model of radiomics and clinical was established,and the ROC was compared with the previously established clinical prediction model,radiomics model,EAD model and MELD model.Their AUC in order from largest to smallest is combined model(AUC:0.919;95%CI:0.840-0.998)>clinical model(AUC:0.884;95%CI:0.791-0.977)>imaging omics model(AUC:0.801;95%CI:0.670-0.931)>MELD score model(AUC:0.782;95%CI:0.654--0.909)>EAD model(AUC:0.700;95%CI:0.581-0.819).Through the comparison,it was found that the combined prediction model of radiomics and clinical diagnosis had the best prediction performance.Conclusion:①CT imaging features can be used as an important evaluation and predictive factor for graft function and patient prognosis after liver transplantation;②Radiomics combined with traditional clinical indicators can improve the predictive performance of liver transplantation patients’graft function and patient prognosis;③Compared with the more classic predecessor research models,the combined radiomics model has better prediction performance,which can be used as a reference and supplement;④Radiomics guides personalized management and treatment after predicting the function of the graft and the prognosis of the patient.