| Part Ⅰ Prediction of Hepatocellular carcinoma M VI based on peritumoral CT radiomics modelObjective:To evaluate the value of peritumoral 5mm and 10mm CT enhanced Radiomics models in predicting microvascular invasion of isolated hepatocellular carcinoma(HCC)≤5cm.Materials and Methods:A retrospective analysis was performed on 206 patients with isolated hepatocellular carcinoma ≤5cm who were pathologically confirmed and met inclusion criteria from 2017 to 2021,including 177 males and 29 females with an age range of 19-81 years and a median age of 55 years.There were 61 MVI-positive patients and 145 MVI-negative patients.Randomly divided into training group 138 cases,verification group 68 cases.Importing patient information to Siemens Medical Research Platform(Syngo,Research Frontier).Radiomics 2.6 was used to obtain volume area of interest(VOI)from the arterial phase(AP)and portal vein(PVP)images layer by layer for 3D segmentation on preoperative dynamic enhanced CT images of patients.Then the computer automatic dilation algorithm was used to obtain the radiomics characteristics of the 5mm and 10mm volumetric area of interest around the tumor.Each radiomics group contains 1691 features.The first choice is to select the features with good consistency and high stability through observer consistency test,and then remove the features with correlation coefficient greater than 0.90.In the hypothesis test of the remaining features,select the features related to microvascular invasion.Finally,LASSO algorithm is used to select the most predictive features to build the model.Results:1.The AUC values of 5mm and 10mm Radiomics model validation sets for arterial tumors were 0.759 and 0.637,respectively.The AUC values of the 5mm and 10mm Radiomics model validation set in portal vein phase were 0.626 and 0.693,respectively.2.Compared with the conventional imaging feature method,the AUC value of the 5mm Radiomics model as a diagnostic index of arterial tumor was about 0.845.The AUC value of TTPVI as a diagnostic index was 0.662.The AUC value of RVI as a diagnostic index is about 0.603.Delong test P=1.147e-08 for the peritumoral Radiomics model and RVI,and Delong test P=2.776e-04 for the Radiomics model and TTPVI,showing significant statistical differences.Conclusion:1.The 5mm and 10mm peritumor Radiomics models have value in predicting hepatocellular carcinoma MVI.The 5mm peritumor model in arterial stage has higher diagnostic efficacy than the 10mm peritumor model,while the 10mm and 5mm peritumor models in portal vein stage have little difference in diagnostic efficacy.2.Compared with traditional imaging methods,the 5mm peritumoral arterial model has higher accuracy and sensitivity in predicting MVI.Part Ⅱ Prediction of Hepatocellular carcinoma MVI by tumor body combined with peritumoral CT radiomics modelObjective:1.To investigate the value of CT enhanced Radiomics model in predicting microvascular invasion of isolated hepatocellular carcinoma ≤5cm.2.To explore the diagnostic efficacy of combined models of different phase and peritumoral Radiomics features in predicting microvascular invasion of isolated hepatocellular carcinoma ≤5cm,and to establish and verify a model with high diagnostic efficacy.3.The Radiomics model with the highest diagnostic efficiency was fused with clinical labels and image features to establish a model with practical value to predict the preoperative MVI status of patients with isolated hepatocellular carcinoma ≤5cm.Materials and Methods:A retrospective analysis was performed on 206 patients with isolated hepatocellular carcinoma ≤5cm who were pathologically confirmed and met inclusion criteria from 2017 to 2021,including 177 males and 29 females with an age range of 19-81 years and a median age of 55 years.There were 61 MVI-positive patients and 145 MVI-negative patients.Randomly divided into training group 138 cases,verification group 68 cases.Importing patient information into Siemens Medical Research Platform(Syngo,Research Frontier)Radiomics software 2.6 to obtain volume area of Interest(VOI)by layer by layer delineation from arterial phase(AP)and portal vein(PVP)images on preoperative dynamic enhanced CT images of patients.Then,the 5mm volumetric area of interest around the tumor was obtained by using the computer automatic swelling algorithm,so as to obtain the imaging characteristics of four groups for arterial phase tumor,5mm around the tumor in arterial phase,portal vein phase tumor and 5mm around the tumor in portal vein phase.Each group contains 1691 features.The first choice is to select the features with good consistency and high stability through observer consistency test,and then remove those whose correlation coefficient is greater than 0.90.The hypothesis test is carried out on the remaining features to select the ones with significant statistical differences between the negative and positive microvascular invasion groups.Finally,LASSO algorithm is used to select the most predictive features to build the model.First four separate image group model is established,then the four separate model are combined,5 kinds of combination model is established,respectively is:arterial tumors+venous phase of tumors,arterial tumors+tumor arterial peripheral 5 mm,vein tumors+venous tumor peripheral 5 mm,arterial tumors+venous phase tumor peripheral 5mm model,venous phase tumors+arterial tumor peripheral 5 mm model.The best diagnostic model was obtained by comparing different models.Then,we established a fusion model including the image Radiomics features,image features and clinical trial parameters.The diagnostic efficiency of the model was evaluated by receiver operating characteristic curve(ROC),and the clinical application value of the model was analyzed by decision curve analysis(DCA).Results:1.The AUC values of the 5mm Radiomics model validation set for arterial tumor and peritumor were 0.753 and 0.759,respectively.The AUC values of the 5mm Radiomics model validation set in portal vein phase were 0.728 and 0.626,respectively.2.The AUC values of combined Radiomics model validation set of arterial and portal vein tumors were 0.766;The AUC values of combined Radiomics model validation set of arterial tumor and portal peritumor 5mm were 0.594.The AUC values validation set of portal tumor and portal peritumor 5mm were 0.759,respectively.The AUC values of combined Radiomics model validation set of portal vein phase tumor and arterial phase peritumor 5mm were 0.668.The AUC values of combined Radiomics model for arterial tumor and peritumor 5mm were 0.820,respectively.3.After logistic univariate and multivariate analysis of clinical baseline data,the indicators with statistical significance were:capsular invasion in postoperative pathological results;Preoperative image features peritumoral enhancement and pseudocapsule integrity,and differences in the density of the interface between tumor and liver.And TTPVI.The AUC of traditional clinical model was 0.781.The AUC value of the optimal image combination model was 0.891,and that of the clinical and omics fusion model was 0.910.4.The decision curve showed that the combined tumor and peritumoral Radiomics model had a slightly higher benefit than the traditional model,while the fusion model could achieve a higher benefit,which showed that the fusion model had a good application value in clinical work.Conclusion:1.The efficacy of combined tumor and peritumoral Radiomics model is higher than that of single tumor or peritumoral Radiomics model.The best combination was arterial phase tumor and peritumor 5mm combination.2.The fusion model based on tumor and peritumoral Radiomics and clinical data can help the preoperative diagnosis of microvascular invasion of isolated hepatocellular carcinoma ≤5cm,and this fusion model has good application value.Part Ⅲ Prediction of early postoperative recurrence of hepatocellular carcinoma by tumor and peritumoral CT radiomics modelObjective:1.To investigate the value of multiphase CT enhanced Radiomics model with peritumoral(5mm)microenvironment in predicting early postoperative recurrence(within 2 years)of isolated ≤5cm hepatocellular carcinoma.2.The diagnostic efficacy of combined models of arterial stage,venous stage and arterial phase+venous phase in predicting early recurrence of isolated hepatocellular carcinoma ≤5cm was compared,and a model with high diagnostic efficacy was established and verified.3.The Radiomics model with the highest diagnostic efficiency was fused with clinical labels,image features and postoperative pathology to establish a model with high practical value to predict the early postoperative recurrence of isolated hepatocellular carcinoma patients with ≤5cm.Materials and Methods:A total of 175 patients with isolated hepatocellular carcinoma(≤5cm)in our hospital from 2017 to 2021 were retrospectively analyzed,including 152 males and 23 females,with an age range of 31-72 years and a median age of 55 years.None of the patients had undergone any preoperative treatment.The diagnosis of all the patients was confirmed by the pathology of surgical specimens.SPSS software was used to randomly divide the population into the training group and the verification group.117 cases were randomly divided into the training group,with an age range of 31-71 years and a median age of 54 years.Fifty-eight patients were included in the validation group,ranging in age from 34 to 72 years,with a median age of 56.5 years.Patient by patient information into the Siemens Medical Research Platform(Syngo,Radiomics software(Radiomics 2.6)in Research Frontier was used to delineate layer by layer from arterial phase(AP)and portal vein(PVP)images on preoperative dynamic enhanced CT images to obtain volume area of interest(VOI).Then,the 5mm volumetric area of interest(ROI)around the tumor was obtained by automatic computer expansion algorithm,so as to obtain the imaging characteristics of arterial+5mm and venous+5mm volumetric area of interest.Each imaging group contains 1691 features.The first choice is to select the features with good consistency and high stability through observer consistency test,and then remove those whose correlation coefficient is greater than 0.90.The hypothesis test is carried out on the remaining omics features to select the omics features with significant statistical differences between the negative and positive microvascular invasion groups.Finally,LASSO algorithm is used to select the most predictive features to build the model.Based on the results of the above study in this paper,in this part,the "unit range"after tumor expansion of 5mm was taken as a whole study,the Radiomics model of arterial phase including 5mm peritumor,portal vein phase including 5mm peritumor,and combined portal vein phase including 5mm peritumor was established.Different models were compared to obtain the model with the best diagnostic efficiency,and the model was compared with postoperative clinicopathological model.Subsequently,we established a fusion model including peritumoral Radiomics,clinical information and postoperative pathological information.The receiver operating Characteristic Curve(ROC),decision curve analysis(DCA)were used to analyze the diagnostic efficiency of the model,the clinical application value of the model and the calibration degree of the model.Results:1.The AUC values of validation set of arterial tumor+5mm Radiomics model were 0.642;The AUC values of validation set of the venous phase tumor +5mm Radiomics model were 0.690.The AUC values of validation set of the arterial and venous tumor+5mm Radiomics model were 0.706.2.After logistic univariate and multivariate analysis of clinical baseline data,the indicators with statistical significance included postoperative pathological grade,MVI status,and peritumor enhancement in preoperative images.The AUC value of the ROC curve predicted by the clinicopathological model for early postoperative recurrence was 0.753(95%CI:0.7205,0.8466).About the Preoperative peritumoral Radiomics model,the AUC value of ROC curve was 0.786,95%CI:(0.6836,0.8161).Delong test was performed on the two models(P=0.5856),and there was no statistical difference between the two models in predicting early postoperative recurrence of isolated ≤5cmHCC.3.The decision curve shows that the fusion model can achieve higher returns by fusing preoperative Radiomics and postoperative pathological parameters.The fusion model has good application value in clinical work,and the correction curve of the fusion model has high correction degree.Conclusion:1.The tumor+5mm imaging model has the value of predicting early postoperative recurrence,and the combined model of arterial phase+portal phase is better than any single phase model.2.There was no statistical difference between the combined radiomics model the postoperative clinicopathological model in the predicting early recurrence.3.The fusion model composed of radiomics and clinical and pathological information has high clinical application value in predicting early postoperative recurrence of isolated ≤5cm hepatocellular carcinoma... |
PDF Full Text Request