Prediction Models Of Microvascular Invasion And Postoperative Recurrence Based On Magnetic Resonance Imaging In Stage CNLC Ⅰa Hepatocellular Carcinoma

Part 1Radiomics based on Gd-EOB-DTPA-enhanced magnetic resonance imaging predicts microvascular invasion of stage CNLC Ⅰa hepatocellular carcinoma[Objective]This study aimed to investigate the risk factors for microvascular invasion(MVI)in hepatocellular carcinoma(HCC;stage Ⅰa in the China liver cancer staging system),and to bulid the clinical-radiological model based on clinical and radiographic features,radiomics model based on radiomic features and combined model.[Materials and Methods]This retrospective study focused on HCC patients who underwent radical resection from January 2016 to December 2020.The patients received Gd-EOB-DTPA-enhanced magnetic resonance imaging before surgical treatment.Regions of interest(ROI)were manually delineated on fat-saturated T2WI,arterial,portal venous,transitional,and hepatobiliary phases using ITK-SNAP software.AK software was used to extract radiomic features and IPMs software was used to select the optimized features and then calculated the Radscore.Univariate and multivariate logistic regression analyses were used to determine risk factors for MVI,and different prediction models were established based on independent risk factors.The area under the curve(AUC)and net reclassification index(NRI)were used to evaluate the effectiveness of different models.Based on the optimal model,the nomogram was developed and validated.[Results]AK software was used to extract 6580 radiomic features,and IPMs software was used to screen out 10 optimal radiomic features,and Radscore of all patients was calculated.Univariate and multivariate logistic regression analysis showed the independent risk factors for MVI including total bilirubin>21μmol/L(OR=0.049,95%CI:0.003-0.478,P=0.041),non-smooth margin(OR=0.024,95%CI:0.001-0.392,P=0.020)and increased Radscore(OR=0.360,95%CI:0.133-0.714,P=0.014).In the primary cohort and validation cohort,the AUC of the combined model was 0.955(95%CI:0.879-0.990)and 0.889(95%CI:0.727-0.972)respectively,which was higher than that of the clinical-radiological model(primary cohort:NRI=0.191,P<0.001;validation cohort:NRI=0.128,P=0.011),and higher than that of the radiomics model(primary cohort:NRI=0.150,P=0.037;validation cohort:NRI=0.067,P=0.060).The calibration curve and Hosmer-Lemeshow test showed the predicted results based on combined model was similar to the real situation(primary cohort P=0.941,validation cohort P=0.901).[Conclusions]1.Total bilirubin>21μmol/L,non-smooth margin and increased Radscore were independent risk factors for MVI.2.Compared with clinical-radiological model and radiomics model,the combined model has better predictive efficacy.3.The nomogram based on combined model can quantify the independent risk factors and help clinicians make individualized surgical plan.Part 2Development and validation of a prediction model for postoperative recurrence in stage CNLC Ⅰa hepatocellular carcinoma[Objective]This study aimed to investigate the risk factors for postoperative recurrence in stage CNLC I a hepatocellular carcinoma,and to bulid the prediction model based on clinical,radiographic and radiomic features.[Materials and Methods]This retrospective study focused on HCC patients who underwent radical resection from January 2016 to December 2019.The patients received gadolinium diethylenetriamine pentaacetic acid(Gd-DTPA)-enhanced magnetic resonance imaging before surgical treatment.In 2.1 part,ROI were manually delineated on fat-saturated T2WI,arterial,portal venous and delay phases using ITK-SNAP software.AK software was used to extract radiomic features and IPMs software was used to select the optimized features and then calculated the Radscore.Univariate and multivariate Cox regression analyses were used to determine risk factors for early recurrence,and different prediction models were established based on independent risk factors.In 2.2 part,univariate and multivariate Cox regression analyses were used to determine risk factors for late recurrence and nomogram was bulit by the independent risk factors.Consistency index(C-index)and NRI were used to evaluate the predictive efficacy of models,and calibration curves were used to reflect the consistency.[Results]In 2.1 part,univariate and multivariate Cox regression analysis showed AFP>400 ng/mL(HR=1.941,95%CI:1.068-3.527,P=0.030),peritumoral enhancement(HR=2.514,95%CI:1.376-4.594,P=0.003)and increased Radscore(HR=1.546,95%CI:1.355-1.764,P<0.001)were independent risk factors for early recurrence.In the primary cohort and validation cohort,the C-index of the combined model was 0.839(95%CI:0.787-0.891)and 0.712(95%CI:0.613-0.812)which was higher than that of the clinical-radiological model(primary cohort:NRI=0.356,P<0.001;validation cohort:NRI=0.105,P=0.037),and higher than that of the radiomics model(primary cohort:NRI=0.087,P=0.028;validation cohort:NRI=0.027,P=0.080).In 2.2 part,univariate and multivariate Cox regression analysis showed increased age(HR=1.045,95%CI:1.003-1.089,P=0.036),family history(HR=2.679,95%CI:1.291-5.560,P=0.008),history of postoperative adjuvant therapy(HR=5.090,95%CI:2.046-12.663,P<0.001)and positive MVI(HR=2.176,95%CI:1.014-4.672,P=0.046)were independent risk factors for late recurrence.The C-index of the nomogram model was was 0.766(95%CI:0.660-0.872).The calibration curve showed that the prediction curve was close to the ideal curve.[Conclusions]1.AFP>400 ng/mL,peritumoral enhancement and increased Radscore were independent risk factors for early recurrence.2.Increased age,family history,history of postoperative adjuvant therapy and positive MVI were independent risk factors for late recurrence.3.The predictive model based on independent risk factors has good predictive ability,which helps clinicians to formulate follow-up strategies for early detection and intervention of recurrence.