| Part Ⅰ:Clinical and radiomics study of predicting microvascular invasion of hepatocellular carcinoma based on CT imagingObjective:Using conventional enhanced CT images,combined with clinical factors,clinical and radiomics models were constructed to predict microvascular invasion(MVI)and its risk classification in patients with hepatocellular carcinoma(HCC),and to evaluate its clinical application value.Method:A total of 465 patients with HCC who underwent three-phase enhanced CT scan in our hospital in the past 6 years and met the inclusion and exclusion criteria were retrospectively collected.According to the pathological MVI diagnosis,they were divided into negative group(M0),positive low-risk group(M1)and positive high-risk group(M2).The clinical data of 465 patients were collected and randomly assigned to the training cohort(n=327)and the test cohort(n=138)according to the ratio of 7:3.The tumor,peritumor(1cm)and tum or+peri tum or(1cm)regions were used as the region of interest(ROI)of this study.ROIs were delineated on arterial phase(AP),portal venous phase(PVP)and delayed phase(DP)images,respectively.Radiomics features were extracted,feature dimension reduction and screening were performed,and multiple sets of initial radiomics models were constructed and their efficacies were compared.The model with the best predictive efficacy was selected as the radiomics model for predicting MVI and its risk classification of HCC.Multivariate Logistic regression was used to screen the clinical factors related to MVI to establish a clinical model,and the clinical-radiomics model was constructed by combining the radiomics model and the clinical model.The receiver operating characteristic curve(ROC)was used to evaluate the predictive efficacy of the above models.At the same time,the data of HCC patients with MVI diagnosis from two external hospitals were collected to independently verify the constructed models.The calibration curve and clinical decision curve were used to evaluate the goodness of fit and clinical practicability of the above models.Result:Of the 465 patients,222 were MVI-negative,243 were MVI-positive,which 136 were in the low-risk group(M1)and 107 were in the high-risk group(M2)for MVI-positive patients.(1)The construction of radiomics model:The radiomics model constructed by the three-phase images(AP+PVP+DP)of the "tumor+peritumoral" region has the highest efficiency.Based on this,①a radiomics model(training cohort AUC=0.80,test cohort AUC=0.79)for predicting the presence of MVI(MVInegacive/positive)and ②a radiomics model(training cohort AUC=0.80,test cohort AUC=0.74)for predicting MVI risk classification(MVIM1/M2)were constructed.(2)Construction of clinical model:① Seven clinical factors(neutrophils,lymphocytes,tumor margin,peri tumoral enhancement in arterial phase(PEAP),tumor in vein,intratumoral vascularity and AFP)were associated with the prediction of MVI negative or positive,and were included in the construction of MVInegative/positive clinical model.The AUC values in the training cohort and the test cohort were 0.80 and 0.76,respectively.②Four clinical factors(the number of lesions,No.of lobe involved,tumor in vein and hepatitis B)were associated with the prediction of MVI risk classification(M1 and M2),which were included in the construction of MVIM1/M2 clinical model.The AUC values were 0.74 and 0.64 in the training cohort and the test cohort,respectively.(3)Construction of clinical-radiomics model:The corresponding radiomics model and clinical model were combined to construct a ① clinical-radiomics model for predicting MVInegative/positive(training cohort AUC=0.85,test cohort AUC = 0.80)and② clinical-radiomics model for predicting MVIM1/M2(training cohort AUC=0.84,test cohort AUC=0.76).(4)Efficacy evaluation and external validation of the model:①Comparing the radiomics model,clinical model and clinical-radiomics model,the decision curve showed that the clinical-radiomics model of MVInegacive/positive and MVIM1/M2 had the highest net benefit,respectively.The calibration curve showed that the predicted MVI results were in good agreement with the actual MVI(P>0.05).②Two external hospital data(36 and 77 cases,respectively)were used to verify the clinical-radiomics model of MVInegative/positive,with AUC values of 0.76 and 0.72,respectively.The clinical-radiomics model of MVIM1/M2 was verified,with AUC values of 0.72 and 0.72,resnectivelv.Conclusion:In this study,based on the three-phase enhanced CT images of tumor+peri tumor,the radiomics model,clinical model and clinical-radiomics model for predicting MVI and its risk classification of HCC were constructed and independently verified.All three models can predict MVI and its risk classification of HCC,among which the clinical-radiomics model has the best efficacy and the best clinical benefit.The establishment of the model provides a new solution for individualized risk assessment of liver cancer,which is helpful for clinical decision-making.Part Ⅱ:Clinical and radiomics study of predicting pathological grade of hepatocellular carcinoma based on CT imagingObjective:To evaluate the clinical application value of three-phase enhanced CT images in constructing the clinical and radiomics prediction model of HCC pathology(ES)grading,and to provide new ideas for non-invasive prediction of HCC pathology grading before surgery.Method:A retrospective collection of 426 patients with HCC who underwent three-phase enhanced CT scans in our hospital in the past 6 years(May 2017-September 2022),met the inclusion and exclusion criteria,and had surgical pathology ES classification,were divided into low-level group(ES Ⅰ+Ⅱ)and high-level group(ES Ⅲ+Ⅳ).The relevant clinical data of 426 patients were collected and randomly assigned to the training cohort(n=299)and test cohort(n=127)according to the ratio of 7:3.Tumor and tumor+peritumor(10 mm)were used as the ROI.ROIs were delineated on AP,PVP and DP images respectively,and radiomics features were extracted for feature dimension reduction and screening.Multiple sets of initial radiomics models were constructed and their performance was compared.The model with the best predictive performance was selected as the radiomics model for predicting ES grade of HCC.Multivariate Logistic regression was used to screen clinical factors related to pathological grading to establish a clinical model.Combined with radiomics model and clinical model,a clinical-radiomics prediction model for ES grading of HCC was constructed.The area under the ROC curve(AUC)was used to evaluate the predictive efficacy of the above model.At the same time,the data of HCC patients confirmed by pathology from two other hospitals were collected to independently verify the constructed model.The calibration curve and clinical decision curve were used to evaluate the goodness of fit and clinical practicability of the above model.Results:Of the 426 patients with HCC,168 were in the low-grade group(ES-Ⅰ+ES-Ⅱ)and 258 were in the high-grade group(ES-Ⅲ+ES-Ⅳ).(1)Construction of the best radiomics model:By comparison,we found that the radiomics model constructed by the three-phase image of the tumor area had the best performance,which was used as the radiomics prediction model of HCC ES grading.The training cohort AUC=0.81,the test cohort AUC=0.75.(2)Construction of clinical model:Eight clinical factors(largest tumor size,tumor growth pattern,child-pugh classification,tumor margin,peritumoral enhancement in arterial phase(PEAP),nonrim arterial phase hyperenhancement(NAPH),bile duct invasion and AFP)were related to the prediction of ES grade of HCC,and were included in the construction of clinical model.The training set AUC=0.78,the test set AUC=0.69.(3)Construction of clinical-radiomics model:The above radiomics model and clinical model were combined to construct a clinical-radiomics model for predicting ES grading of HCC.The training cohort AUC=0.85 and the test cohort AUC=0.75.(4)Effectiveness evaluation and external validation of the model:Comparing the radiomics model,clinical model and clinical-radiomics model,the decision curve showed that the latter had the best net benefit,and the calibration curve showed that there was no significant difference between the predicted ES grade and the actual result(P>0.05);two external hospital cases(33 and 24,respectively)independently validated the clinical-radiomics model,with AUC values of 0.75 and 0.79,respectively.Conclusion:In this study,based on the three-phase enhanced CT images of the tumor,the radiomics model,clinical model and clinical-radiomics model for preoperative prediction of ES grade of HCC were constructed and independently verified.All three models can predict the histological grade of HCC,and the predictive efficacy of the clinical-radiomics model is better than the former two,with the best clinical benefit.The establishment of ES grading model can accurately predict the histological grade of preoperative HCC,which provides a new solution for individualized risk assessment before treatment of liver cancer and is helpful for clinical decision-making.Part Ⅲ:Clinical and radiomics study of predicting overall survival of HCC based on CT imagingObjective:The purpose of this study is to combine the radiomics model of MVI and its risk classification constructed in part Ⅰ with the clinical factors related to prognosis to construct a clinical-radiomics model for preoperative prediction of overall survival(OS)of HCC,so as to provide help for individualized evaluation and treatment of HCC patients.Method:A retrospective collection of 427 patients with HCC who underwent three-phase enhanced CT scans in our hospital in the past 6 years(May 2017-September 2022),met the inclusion and exclusion criteria,were pathologically confirmed and followed up for 5 years.According to the ratio of 7:3,they were randomly divided into training cohort(n=299)and test cohort(n=128).Univariate and multivariate Cox regression analysis was used to screen the clinical factors related to OS to construct a clinical model.The clinical model was combined with the part I of the MVInegative/positive and MVIM1/M2 radiomics models to construct the OSMVInegative/positive clinical-radiomics model and the OSMVI M1/M2 clinical-radiomics model to predict the survival of patients with HCC within 5 years after surgery.The risk score of patients with HCC was calculated according to the model formula and a nomogram was drawn.Kaplan-Meier analysis was used to describe and compare the survival curves of the high-risk group and the low-risk group.The area under the time-dependent curve(integrated AUC,iAUC)and calibration curve were used to evaluate the efficacy of the OS prediction model.Results:A total of 427 HCC patients(mean age:55.5 ± 10.6 years,374 males and 53 females)were included in the survival analysis,including 211 MVI negative(M0)and 216 MVI positive(M1 124,M2 92).There were 45 deaths(21.3%)in M0,38 deaths(30.6%)in M1,and 47 deaths(51.1%)in M2.(1)Construction and evaluation of clinical model:① Three clinical indicators(peritumoral star node,AFP,and albumin)were associated with the prediction of MVI negative or positive OS,which were combined into a clinical model of OSMVInegative/positive;its iAUC of training cohort=0.69,test cohort iAUC=0.68.②Four clinical factors(enhancing capsule,peritumoral star node,tumor in vein,and AFP)were associated with the prediction of OS of M1 and M2,and they were combined to construct a clinical model of OSMVI M1/M2.Its training cohort iAUC=0.72,test cohort iAUC=0.71.(2)Construction and evaluation of clinical-radiomics model:①Combined with the MVInegative/positive radiomics prediction model and the OSMVInegative/positive clinical model in the part I,the OSMVI negative/positive clinical-radiomics prediction model of HCC was constructed.The iAUC of the model was 0.72 in the training cohort and 0.71 in the test cohort.Kaplan-Meier survival analysis showed that the OS of the high-risk group and the low-risk group was significantly different between the training set and the test set(training cohort P<0.0001,test cohort P<0.0001).The calibration curve showed no significant difference between the predicted OS and the actual overall survival(P>0.05),indicating good consistency.② Combined with the MVIM1/M2 radiomics model and the OSM1/M2 clinical model in the part I,the OSMVIM1/M2 clinical-radiomics prediction model of HCC was constructed.The iAUC of the model was 0.74 in the training cohort and 0.74 in the test cohort.Kaplan-Meier survival analysis showed that the OS of the high-risk group and the low-risk group was significantly different between the training set and the test set(training cohort P<0.0001,test cohort P=0.016).The calibration curve showed no significant difference between the predicted OS and the actual OS(P>0.05),indicating good consistency.Conclusion:In this study,based on the radiomics prediction model of MVI and its risk classification,combined with survival-related clinical factors,a clinical-radiomics prediction model of OSMVInegative/positive and OSMVIM1/M2 was constructed,which can better predict the survival probability of patients with HCC 1-4 years after surgery,and its predictive efficacy is higher than that of the clinical prediction model alone,providing an effective aid for the selection of individualized treatment plans. |
PDF Full Text Request