CT-Based Radiomics Nomogram For Differentiating Gastric Hepatoid Adenocarcinoma And Predicting Outcome

Part Ⅰ CT-based radiomics-clinical nomogram for differentiation gastric hepatoid adenocarcinoma from gastric adenocarcinomaObjectives:This study aimed to develop a CT-based radiomics nomogram for the highprecision preoperative differentiation of gastric hepatoid adenocarcinoma(GHAC)patients from gastric adenocarcinoma(GAC)patients.Methods:108 consecutive patients with GHAC from 6 healthcare centers and 108 GAC patients matched by age,sex and T stage undergoing pathological examination were retrospectively reviewed.Patients from centers 1-5 were randomly divided into two cohorts(derivation and internal validation)at a 7:3 ratio,the remaining patients were external validation cohort.Sex,age,preoperative laboratory tests(CEA,CA199,CA72-4,AFP)and preoperative CT imaging features(tumor location,tumor characteristics,maximum tumor diameter)were analyzed to develop clinical score(Cli-score).Radiomics features were extracted from contrast enhanced CT images.A radiomics score(Rad-score)was developed based on reproducible features by means of the least absolute shrinkage and selection operator(LASSO)method.Integrating Rad-score and Cli-score,a Combined-Nomogram was developed.The diagnostic performance of each model was evaluated by using the Area under the curve(AUC)of Receiver operating characteristics(ROC),and the calibration degree and clinical application value of each model were evaluated by using calibration and Decision curve analysis(DCA).Results:Multivariate logistic regression showed that preoperative AFP(OR=99.31;95%CI:24.9-396.09;P<0.001)and CA72-4(OR=2.86;95%CI:1.02-8.03;P=0.046)were independent predictors of GHAC and GAC,and the AUC in the training set,internal validation set and external test set were 0.882(95%Cl,0.882-0.928)、0.938(95%Cl,0.914-0.998)、0.831(95%Cl,0.725-0.938).Ten features were used to build Rad-score,which yielded an AUC of 0.998,0.942,and 0.731 in the derivation,internal validation,and external validation cohort,respectively.The Combined-Nomogram showed better discrimination,with an AUC of 0.998,0.954,and 0.909 in the derivation,internal validation,and external validation cohort,respectively.The Combined-Nomogram showed good calibration.Decision curve analysis(DCA)demonstrated the clinical usefulness of the Cli-score and Combined-Nomogram.Conclusion:The noninvasive CT-based nomogram,including radiomics score,AFP,and CA72-4,showed favorable predictive efficacy for differentiating GHAC from GAC and might be useful for clinical decision-making.Part Ⅱ CT-based radiomics-clinical nomogram for prediction of overall survival in patients with gastric hepatoid adenocarcinomaObjectives:Gastric hepatoid adenocarcinoma(GHAC)is a highly aggressive solid tumor with a high degree of malignancy,predisposition to extensive lymph node metastasis and liver metastasis,and a poor prognosis.Therefore,it is of great significance to accurately predict the overall survival(OS)before surgery,stratify the risk,and realize GHAC personalized treatment,so as to prolong the survival time of patients.Therefore,the aim of this study was to use radiomics methods combined with clinical risk factors to develop and establish a model for predicting overall survival in GHAC patients.Methods:This retrospective study involved 107 patients with GHAC from 6 medical centers from January 2007 to June 2021 who were pathologically confirmed and had enhanced CT images available before treatment.The patients were randomly assigned to a derivation cohort(n=74)and a validation cohort(n=33).We used Cox regression for univariable and multivariable analysis to develop a clinical model,and independent predictors related to OS were screened out to develop clinical factors models(Cli-score).Radiomics feature extraction is the same as the first part,and feature screening uses LASSO Cox regression to construct an radiomics model(Rad-score).A Combined-Nomogram was constructed by combining the radiomics and the clinical factors model.The Harrell consistency index(C-index),the timedependent ROC and DC A was used to quantify the diagnostic performance of each model;The calibration curve is used to evaluate the consistency between the model prediction and the actual results;The proportional hazard hypothesis of the model was verified by examining the scaled Schoenfeld residual plots;The optimal cut-off values of the three models were used to divide patients into low-risk group and high-risk group;The Kaplan-Meier method was drawn,and the log-rank test was used to compare the survival curve differences between high-risk group and low-risk group.Results:OS was similar among the derivation cohort and the validation cohort(P=1.000,log-rank test).The multiple cox regression analysis showed that AFP(P=0.006),CA72-4(P=0.002)and liver metastases(P=0.025)remained as independent risk predictors associated with OS.These three predictors were used to build the Cli-score,and the C-index in the derivation cohort and validation cohort were 0.784(95%CI:0.686-0.883)and 0.764(95%CI:0.659-0.869),respectively.With use of 2.733 as cutoff score of the training cohort,the Cliscore identified two risk categories of recurrence.There were significant differences between the low-and high-risk subgroup(P<0.001 for two cohorts,log-rank test).Eight radiomics features were used to build Rad-score,which yielded C-index of 0.756,0.765,and 0.741 in the derivation and validation cohort,respectively.With use of 0.132 as Rad-score cutoff score,there were significant differences between the low-and high-risk subgroup(P<0.001 for two cohorts,log-rank test).The Combined-Nomogram showed better predictive performance,with C-index of 0.841,0.821 in the derivation validation,and validation cohort,respectively.With use of 1.886 as Combined-Nomogram cutoff score,there were significant differences between the low-and high-risk subgroup(P<0.001 for two cohorts,log-rank test).The calibration curve showed good agreement between the OS predicted by the Combined-Nomogram and observed outcomes.DCA demonstrated the clinical usefulness of the Combined-Nomogram.Conclusion:Combined-Nomogram,based on CT radiomics combined with clinical model,showed good predictive efficacy in predicting OS in GHAC patients.It is of great clinical significance for the realization of individualized treatment of patients and thus prolonging the survival time of patients.