| Objective:1.To construct seven radiomics models of the arterial phase,portal phase,delayed phase and combination of each phases to determine a reliable and stable radiomics model based on dynamic contrast-enhanced MRI in exploring the early therapeutic response of models for hepatocellular carcinoma after surgical resection.2.To investigate the predictive value of radiomics models of tumor and peritumor parameter characteristics extracted from DCE-MRI for assessing the early recurrence risk after HCC surgical resection.3.To construct a reliable and stable radiomics models to assess the early recurrence risk of HCC after transarterial chemoembolization(TACE)based on dynamic contrast-enhanced MRI.4.To construct a combined model with multimodal MRI features and clinical information,and to explore the comprehensive application value of the prediction model for recurrence and early outcome after surgical resection of HCC.Materials and methods:In this study,285 patients diagnosed with hepatocellular carcinoma in the First Affiliated Hospital of Kunming Medical University from January 2013 to December 2020 were collected retrospectively.According to the Guidelines for Diagnosis and Treatment of Primary Liver Cancer(2022 edition),12 months was used as the threshold to distinguish early recurrence and non-recurrence in patients with hepatectomy in the first and fourth parts of the study.A total of 125 patients were collected,including 51 in the recurrence group and 74 in the non-recurrence group.Patients were randomly divided into 88 cases in the training cohort and 37 cases in the validation cohort according to a 7:3 ratio using a cross-validation method.The second part of the inclusion criteria are the same as the first part.96 patients with HCC were collected,including 39 cases in the recurrence group and 57 cases in the non-recurrence group.patients were randomly divided into a training set of 67 cases and a validation set of 29 cases according to a 7:3 ratio using a cross-validation method.The third part of the study used the AASLD modified response evaluation criteriain solid tumors(mRECIST)as recommended by the TACE clinical practice guidelines.Complete remission CR and partial remission PR were defined as recurrence,and progressive PD and stable SD were defined as non-recurrence.A total of 52 patients with HCC treated with TACE were finally collected,including 30 in the recurrence group and 22 in the non-recurrence group.Patients were randomly divided into 36 cases in the training cohort and 16 cases in the validation cohort according to a 7:3 ratio using a cross-validation method.DCE_MRI sequences were exported in digital imaging and communication in medicine(DICOM)format.The regions of interest(ROI)in all levels of each sequence were manually segmented layer by layer using 3D Slicer(Boston,MA,USA,Version 4.11.20210226)software.Select the segment editor module to add nodes around the tumor region in each phase image to confirm the tumor boundary,and activate the 3D function to obtain a complete 3D volume of interest(VOI)that incorporates the ROI of each layer of interest.Radiomics features were extracted from DCE-MRI arterial phase,portal venous phase,and delayed phase VOI using the open source package Pyradiomics(Version 3.0.1),with custom extraction feature codes enabled all image types and feature types.The Pearson Correlation Analysis,least absolute shrinkage and selection coefficients(LASSO)regression algorithm was used to downscale and filter the radiomics features.Optimized parameter approach to screen for reproducible and stable key radiomics features.Using repeated k-fold cross-validation,PCA principal component analysis was used to optimize the parameters and establish a radiomics-based early efficacy prediction model.The features were extracted with 1549 features in each of the three phases of dynamic contrast-enhanced MRI,and the massive amount of extracted radiomics feature data was preprocessed before data analysis.Firstly,26 non-numeric columns were removed,and the remaining 1523 columns of feature data were converted to decimal format,after mixing all the data,the missing values within the data were filled in,and finally all data were standardized with Z-scores to ensure that all data could be measured and compared uniformly.The measurement data were expressed as mean±standard deviation or median/quartile,and the Enumeration data were expressed as number or proportion.The receiver operation characteristic curve(ROC)and area under curve(AUC)95%confidence interval(CI),sensitivity,and specificity were used to assess the predictive efficacy of the model.Accuracy,precision,recall,and F1 values were calculated by confusion matrix.The degree of model bias was evaluated using calibration curves to assess the fit of the model and the value of the model for clinical application was analyzed using decision curves.In the first part of the study,"Prediction of early recurrence risk after surgical resection of hepatocellular carcinoma based on multimodal MR radiomics models",seven radiomics models were constructed:①A arterial phase model;②V portal venous phase model;③D delayed phase model;④ A+V model;⑤V+D model;⑥A+D model;⑦A+V+D model.In the second part of the study,"Prediction of early recurrence risk after surgical resection of hepatocellular carcinoma based on multimodal MR imaging tumor+perimeter model",two radiomics models were constructed:①A+V+D tumor+perimeter 3mm model;②A+V+D tumor+perimeter 6mm model.In the third part of the study,"Prediction of early recurrence risk after TACE for hepatocellular carcinoma based on multimodal MR radiomics model",one A+V+D TACE radiomics model was constructed.In the fourth part of the study,"Prediction of early ecurrence risk after surgical resection of hepatocellular carcinoma based on multimodal MR radiomics+clinical model",a multimodal radiomics+clinical information combined model was established by combining clinical data with radiomics features.Results:In the first part of the study,the AUC values of training cohort for the arterial,portal venous,and delayed phase models ranged from 0.69 to 0.81,and the AUC values of validation cohort ranged from 0.63 to 0.70;the AUC values of training cohort for the arterial+portal venous,arterial+delayed,and portal venous+delayed phase models ranged from 0.74 to 0.85,and the AUC values of validation cohort ranged from 0.63 to 0.78;the arterial+portal venous+delayed phase radiomics model had AUC values of 0.91 for the training cohort and 0.66 for the validation cohort.All seven models had good clinical application in predicting the early recurrence risk of surgical resection in HCC patients,and the combined dynamic contrastenhanced three-phase model had better efficacy than the other six models.In the second part of the study,the A+V+D G6mm model combined the radiomics features of three arterial+portal venous+delayed phase and tumor+perineural 6 mm.According to the statistical results,the AUC values of the A+V+D G3mm and G6mm models ranged from 0.82 to 0.93 for the training cohort and 0.61 to 0.78 for the validation cohort,and the overall efficacy of the A+V+D G6mm model was slightly better than that of the A+V+D G3mm model.The radiomics model of tumor+peri-tumor has good clinical application in predicting the early therapeutic response of surgical resection in HCC patients.In the third part of the study,the TACE model combined dynamic contrast-enhanced threephase arterial+portal venous+delayed phase radiomics features.the TACE radiomics model obtained an AUC value of 0.82 for the training cohort and 0.61 for the validation cohort by ROC analysis.According to the statistical results,the confusion matrix of the TACE radiomics model,the decision curve showed that the model has good clinical application in predicting the therapeutic response of HCC patients after TACE.In the fourth part of the study,the AUC values of the clinical,radiomics,and fusion models of the multimodal radiomics+clinical model were 0.76,0.67,and 0.79,respectively,with an accuracy of 0.74,sensitivity of 0.63,specificity of 0.82,and Yordon index of 0.45.The ROC analysis indicated that the fusion model was more effective than the clinical model and radiomics models.The results showed that the predicted probability deviated to a lesser extent from the actual probability,and the calibration curve of the fusion model fitted well.The decision curve analysis showed that the model could achieve better net clinical benefit in most of the threshold probability ranges.Conclusion:1.Radiomics models constructed based on MR dynamic contrast-enhanced sequences A/V/D phase and A+V/A+D/V+D phase and A+V+D phase do have predictive value for the assessment of early recurrence after surgical resection in HCC patients.The A+V+D phase radiomics model was more effective than the A/V/D phase and A+V/A+D/V+D phase model.2.The overall efficacy of the A+V+D G6mm model was slightly better than that of the A+V+D G3mm model.The radiomics model of tumor+peri-tumor has good clinical application in predicting the early therapeutic response of surgical resection in HCC patients.3.MR dynamic contrast-enhanced sequences arterial+portal venous+delayed phase radiomics model has good clinical application in predicting the early therapeutic response after TACE in HCC patients.4.Clinical model is valuable in assessing the early outcome of HCC patients after surgical resection.The predictive efficacy of the multimodal radiomics features+clinical fusion model was higher than that of the single model.The multimodal fusion model provides an important basis for preoperative clinical assessment and is an important guide for the selection of clinical treatment plans. |
PDF Full Text Request