Radiomics Analysis Based On Contrast-enhanced MRI For Prediction Of The Early Therapeutic Response To Transarterial Chemoembolization In Hepatocellular Carcinoma

Purpose:1.To explore the value of intratumoral radiomics models based on contrast-enhanced magnetic resonance imaging（CE-MRI）in predicting the early therapeutic response to transarterial chemoembolization（TACE）in hepatocellular carcinoma（HCC）.2.To construct radiomics models based on peritumoral and intratumoral combined peritumoral derived from CE-MRI to predict the early therapeutic response of TACE in HCC.3.To incorporate clinical and conventional radiological information with the radiomics model,and to explore the value of clinical-radiological-radiomics combined model in predicting the early therapeutic response of TACE in patients with HCC.Materials and Methods:From April 2008 to August 2020,we retrospectively collected HCC patients who were pathologically confirmed or met the American Association for the study of liver diseases（AASLD）guidelines in our hospital.All patients have underwent upper abdominal CE-MRI with 1.5T or 3.0T MRI system one month before and three months after TACE treatment.The modified Response Evaluation Criteria in Solid Tumors（m RESICT）criteria was used to evaluate the early therapeutic response of TACE,and the treatment effect was divided into treatment effective（complete response and partial response）and treatment ineffective（disease stable and disease progression）.A total of113 HCC patients were collected in our study,including 55 patients in the effective group and 58 patients in the ineffective group.The patients were randomly divided into the training cohort of 78 cases（effective group（n=38）,ineffective group（n=40））and the validation cohort of 35 cases（effective group（n=17）,ineffective group（n=18））at a radio of 7:3.In the first part of the study about“Intratumoral radiomics model based on CE-MRI to predict the early therapeutic response of TACE in HCC”,the arterial phase（AP）,portal venous phase（PVP）,and delayed phase（DP）images were imported into the A.K.software for pixel resampling and signal intensity standardization.The region of interests（ROIs）were manually delineated around the entire tumor outline on each axial slice using the ITK-SNAP software by two radiologists,and were fused into volume of interest（VOI）.The A.K.software was used to extract 402 radiomics features on each enhanced phase image,including histogram,texture,and shape features.The inter-and intra-observer consistency test,Spearman’s rank correlation test,hypothesis test（Student’s t-test or Mann-Whitney U-test）,and least absolute shrinkage and selection operator（LASSO）algorithm were used for radiomics feature selection in order.The logistic regression was used to construct radiomics model for predicting TACE efficacy,with penalty parameter tuning conducted by 5-fold cross-validation.The radiomics score（Radscore）was calculated using the selected features weighted by their respective coefficients in the regression model.In this study,we built seven intratumoral radiomics models,including AP model,PVP model,DP model,AP-PVP model,AP-DP model,PVP-DP model,and enhanced three phase combined model.The receiver operating characteristic curve（ROC）was used to evaluate the performance of each model.The De Long’s test was used to compare the area under the curve（AUC）between the models and to determine the intratumoral radiomics model with higher performance,which provided the basis for the second part of the study.Calibration curve（Hosmer-Lemeshow test）was used to evaluate the calibration performance of the model.Decision curve analysis（DCA）was conducted to estimate the clinical utility.We performed stratified analysis on MRI scanner subgroups of radiomics models.In the second part of the study about“Predicting the early therapeutic response of TACE in HCC based on peritumoral radiomics model on CE-MRI”,we selected enhanced three phase MR images to perform radiomics analysis on peritumoral region（because our first part suggested that the enhanced three phase combined radiomics model based on intratumoral region obtained better performance（defined as intratumoral radiomics model））.All of the AP,PVP,and DP images and the corresponding tumor segmentations in the first part of the study were imported into the A.K.software.The VOI^peri3,VOI^peri5,and VOI^peri10were automatically generated by equidistant expansion of 3 mm,5 mm and 10 mm from the lesion border,respectively.If the ROI was beyond the parenchyma of the liver after expansion,the portion beyond the parenchyma was removed manually.Radiomics features were extracted using A.K.software for VOI^peri3,VOI^peri5,and VOI^peri10,and a total of 1206 features were extracted for each VOI.The inter-and intra-observer consistency test,Spearman’s rank correlation test,hypothesis test（Student’s t-test or Mann-Whitney U-test）,and LASSO algorithm were used for radiomics feature selection in order.The logistic regression was used to construct radiomics model,with penalty parameter tuning conducted by 5-fold cross-validation,and the radscore was calculated.Our study constructed three peritumoral radiomics models,including 3 mm peritumoral model,5 mm peritumoral model,and 10 mm peritumoral model.Finally,we built intratumoral combined peritumoral radiomics model by combining the radiomics features of peritumoral radiomics model and intratumoral radiomics model.The intratumoral combined peritumoral radiomics models included intratumoral plus 3 mm peritumoral model,intratumoral plus 5 mm peritumoral model,and intratumoral plus 10 mm peritumoral model.The ROC analysis,calibration curve（Hosmer-Lemeshow test）,and DCA were used for model evaluation.The De Long’s test was used to compare the AUC values between the models,and the model with higher performance was selected for the third part of the study.In the third part of the study about“Predicting the early therapeutic response of TACE in HCC based on the clinical-radiological-radiomics combined model on CE-MRI”,we collected clinical data（age,gender,history of hepatitis B,serum alpha fetoprotein（AFP）,alanine aminotransferase（ALT）,aspartate aminotransferase（AST）,gamma glutamyltransferase（GGT）,alkaline phosphatase（ALP）,total bilirubin（TBIL）,albumin（ALB）,platelet（PLT）,prothrombin time（PT）,Child-Pugh class,ECOG score,and BCLC stage）and conventional radiological characteristics（tumor location,number,shape,size,margin,intratumor necrosis,intratumor hemorrhage,intratumor fat,tumor capsule,peritumoral enhancement,satellite nodule,internal arteries,radiologic cirrhosis,and venous tumor thrombus）.Univariate and multivariate logistic regression analyses of clinical and conventional radiological characteristics were used to identify the independent factors for predicting the early treatment efficacy of TACE.The logistic regression was used to construct clinical-radiological model.Because the second part suggested that intratumoral plus 3 mm peritumoral radiomics model obtained better performance,so we selected it as the final radiomics model.The established radscore and independent clinical-radiological factors were combined to construct the clinical-radiological-radiomics combined model（combined model）.The combined model was presented as the nomogram.The ROC analysis（De Long’s test）,calibration curve（Hosmer-Lemeshow test）,and DCA were used for model evaluation.We performed stratified analysis on the subgroups of age,gender,AFP,Child-Pugh class,ECOG score,BCLC stage,tumor size,and vein tumor thrombus.Results:In the first part of the study,we compared the performance of intratumoral radiomics models on each enhanced phase and its different combinations.（1）Single enhanced phase（AP,PVP,DP）radiomics models:the AUCs were 0.763-0.779 and0.775-0.781 in the training and validation cohorts,respectively.The DP model showed higher performance（AUC=0.779 and 0.781）,but there were no significant differences in the AUCs between the models（P>0.05）.（2）Combined enhanced phase（AP-PVP,AP-DP,PVP-DP,enhanced three phase combined）radiomics models:the AUCs were0.774-0.810 and 0.768-0.830 in the training and validation cohorts,respectively.Among all radiomics models,the enhanced three phase combined model showed higher performance（AUC=0.810 and 0.830）.However,there were no significant differences in the AUCs between the models（P>0.05）.All radiomics models showed good calibration performance and clinical utility.The stratified analysis showed that the performance of each radiomics model was not affected by MRI scanners with different magnetic field（P>0.05）.In the second part of the study,we compared the performance of peritumoral and intratumoral combined peritumoral radiomics models.（1）Peritumoral（3 mm,5 mm,and 10 mm）radiomics models:the AUCs were 0.760-0.812 and 0.742-0.837 in the training and validation cohorts,respectively.The 3 mm peritumoral model showed higher performance（AUC=0.812 and 0.837）,but there were no significant differences in the AUCs between the models（P>0.05）.（2）Intratumoral combined peritumoral（3mm,5 mm,and 10 mm）radiomics models:the AUCs were 0.822-0.864 and 0.833-0.873 in the training and validation cohorts,respectively.The intratumoral plus 3 mm peritumoral model showed higher performance（AUC=0.864 and 0.873）,but there were no significant differences（P>0.05）.The 3 mm peritumoral model showed comparative performance compared with the intratumoral model（training cohort:AUC,0.812 vs.0.810,P=0.977;validation cohort:AUC,0.837 vs.0.830,P=0.950）.When the intratumoral model was combined with 3 mm peritumoral model,the intratumoral plus 3 mm peritumoral model showed better performance compared with intratumoral model,but there were no significant differences（P>0.05）.All peritumoral and intratumoral combined peritumoral radiomics models showed good calibration performance and clinical utility.In the third part of the study,ALP and tumor size were independent factors for predicting the early therapeutic efficacy of TACE.In the training cohort,the performance of the clinical-radiological-radiomics combined model（combined model）（AUC:0.900）was significantly higher than that of the clinical-radiological model（AUC:0.774）,and there was significantly different of the AUC between the models（P=0.042）.In the validation cohort,the performance of the combined model（AUC:0.889）was slightly higher than that of the clinical-radiological model（AUC:0.771）,but there was no significant difference（P>0.05）.In both cohorts,the combined model showed slightly higher performance compared with the radiomics model（AUC:0.864 and0.873）,but there were no significant differences（P>0.05）.The clinical-radiological model,radiomics model,and combined model showed good calibration performance and clinical utility.The stratified analysis showed that there were no significant differences in the AUCs between different subgroups of combined nomogram（P>0.05）.Conclusions:1.The intratumoral radiomics model based on standardized CE-MR images has a certain value in predicting the early therapeutic response after TACE in patients with HCC.The enhanced three phase combined model is the optimal intratumoral radiomics model in this study,and has certain generality in 1.5T and 3.0T MRI scanners.2.The radiomics method can be used to analyze the peritumoral tissue in a non-invasive way before treatment.The 3 mm peritumoral radiomics model based on CE-MRI showed comparative predictive power compared with the intratumoral radiomics model for TACE response.The combination of intratumoral radiomics model and 3 mm peritumoral radiomics model obtained higher performance.3.The clinical-radiological-radiomics combined model is the optimal model for therapeutic response prediction in this study,which can be presented as a visual nomogram.It potentially provides a simple,individualized,and accurate prediction tool for clinicians,which has certain generality in clinic.