| Part 1:Predictive factors of lymph node metastasis of intrahepaticcholangiocarcinomaObjective:To explore the predictive factors of lymph node metastasis of intrahepatic cholangiocarcinoma(ICC)and to further explore the relationship between tumor diameter and lymph node metastasisMethod:A total of 645 ICC patients who underwent hepatectomy and lymph node dissection in Easterm Hepatobiliary surgery Hospital from January 2013 to January 2016 were divided into lymph node positive group(n = 157)and lymph node negative group(n＝488).The basic clinicopathological features of lymph node positive group and lymph node negative group were compared.Kaplan-meier method and log-rank test were used to compare the long-term recurrence rate and survival rate between two groups.Cox proportional hazard model was used to analyze the independent risk factors of tumor recurrence and survival.Logistic regression was used to analyze the predictive factors of lymph node metastasis of ICC.Restricted cubic spline(RCS)was used to analyze the relationship between tumor diameter and lymph node metastasisResults:For all cohorts,the 1-,3-and 5-year recurrence rates were 54.2%,71.7%and 77.4%,respectively.The postoperative survival rates were 66.1%,42.0%and 33.0%,respectively.The 1-,3-and 5-year recurrence rates were 71.1%,86.1%,92.3%and 48.8%,67.1%and 72.5%in the lymph node metastasis positive group and lymph node negative group,respectively(p<0.001).The 1-,3-and 5-year postoperative survival rates were 48.2%,29.10%,16.0%and 71.9%,46.2%and 38.9%in the lymph node metastasis positive group and lymph node negative group,respectively(p<0.001).Multivariate Cox regression analysis showed that lymph node metastasis was an independent risk factor for tumor recurrence and survival(Recurrence:P<0.001,HR:1.518,95%CI:1.232-1.869;Survival:P<0.001,HR:1.547;95%CI:1.236-1.937).There were significant differences in serum CA19-9,tumor diameter,vascular invasion,tumor location,tumor differentiation and tumor gross classification between lymph node metastasis positive group and lymph node metastasis negative group(p<0.05).Multivariate Logistic regression showed that the independent risk factors for lymph node metastasis were:periductal infiltrating type(p<0.001,OR:2.583,95%CI:1.482-4.464),tumor located in the left lobe(p<0.001 OR:2.738,95%CI:1.836-4.136),vascular invasion(p<0.001,OR:3.173,95%CI:1.917-5.258),tumor diameter(p=0.021.,OR:1.081.,95%CI:1.011-1.155),lower differentiation(moderate vs.high:p=0.013,OR:2.459,95%CI:1.252-5.262;low vs.high:p=0.032,OR:2.708,95%CI:1.102-6.896).The results of restricted cubic spline analysis showed that when the tumor diameter was less than 7cm,the incidence of lymph node metastasis was linearly correlated with the tumor diameter,while it was not when the tumor diameter was more than 7cm.Conclusion:Lymph node metastasis of ICC is an independent prognostic risk factor.Periductal infiltrating type,tumor located in left lobe,vascular invasion,tumor diameter,tumor differentiation are the predictive factors of lymph node metastasis.The lymph node metastasis rate increased linearly with the increase of diameter when the diameter of tumor was less than 7cm,but did not change with the diameter when the diameter of tumor was below 7cm.Part 2:Development of a radiomics nomogram for prediction oflymph node metastasis of intrahepatic cholangiocarcinomaObjective:To extract radiomics features and develop a radiomics nomogram for preoperative prediction of lymph node metastasis of intrahepatic cholangiocarcinoma.Method:A total of 247 mass-forming ICC patients who underwent hepatectomy,lymph node dissection and preoperative CT scan from January 2010 to December 2016 were included.Of them,70%(172 cases)were randomly selected as the training cohort and the remaining 30%(75 cases)as the validation cohort.All the CT image data are extracted by using PACS system(GE),whose extraction format is DICOM.ITK-SNAP version 3.6.0 was used to segment and outline the region of interest(RQI).The Pyradiomics package of Python was used to extract statistical features,shape features,texture features and wavelet features.The maximum correlation minimum redundancy feature selection method and Pearson correlation coefficient were used to further summarize and select the radiomics features.The random forest model was used to optimize the super-parameters of the model and the relative importance of each radiomics feature and the radiomics score were obtained.Radiomics score was further used as a variable in multivariate Logistic regression to analyze the predictive factors of lymph node metastasis..The radiomics nomogram was established based on the Logistic regression.Results:There was no significant difference in basic clinicopathological features between the training cohort and the validation cohort(p>0.05).832 radiomics features were successfully extracted by Pyradiomics,26 of which were further screened by the maximum correlation minimum redundancy method.The Pearson correlation coefficient method was used to select 16 features with correlation greater than 0.9.The relative importance of each radiomics feature and radiomics score are obtained by using random forest classifier model.The radiomics score had a good performance in predicting lymph node metastasis(AUC:0.784).The radiomics score was correlated with vascular invasion and tumor grade(p<0.001,p=0.023).Multivariate Logistic regression analysis showed that the independent predictive factors of lymph node metastasis were:preoperative imaging tumor diameter(p=0.016,OR:2.586,95%CI:1.191-5.616),tumor located in left lobe(p=0.002,OR:5.470,95%CI:1.880-15.919),diagnosis of lymph node metastasis by CT(p=0.001,OR:19.653,95%CI:6.098-63.340)and radiomics score(p=0.001,OR:10.359,95%CI:4.273-25.289).The radiomics nomogram for prediction of lymph node metastasis was developed based on the results of Logistic regression and the RCS transformation of preoperative imaging tumor diameter.The optimal threshold to discriminate high or low risk of lymph node metastasis was 10.5 points.Conclusion:The radiomics score developed by the extraction of radiomics features is an important predictor of lymph node metastasis in ICC.The radiomics score had a good performance in predicting lymph node metastasis.The radiomics score might be a predictor of vascular invasion and tumor grade.The radiomics nomogram for prediction of lymph node metastasis of ICC was an individualized prediction tool for clinical use.Part 3:Validation of the radiomics nomogram for prediction oflymph node metastasis of intrahepatic cholangiocarcinomaObjective:To validate the predictive ability of the radiomics nomogram of lymph node metastasis of intrahepatic cholangiocarcinoma.Method:C-index,ROC curve,decision curve,net classification improvement degree(NRI)and integrated discrimination improvement degree(IDI)were used to compare our radiomics nomogram with CT diagnosis and the nomograms published before.Results:The radiomics nomogram for prediction of lymph node metastasis based on Logistic regression had an validation C-index of 0.926(95%CI:0.887-0.965)in training cohort and 0.862(95%CI:0.778-0.946)in validation cohort.The nomogram published before includes the preoperative nomogram and the pathologic nomogram including the pathological grade variable.The validation C-index is 0.683(95%CI:0.574-0.792),0.743(95%CI:0.628-0.858)and 0.747(95%CI:0.633-0.860)for CT diagnosis and these two models,respectively.The NRI values comparing our radiomics nomogram with CT diagnosis,the preoperative nomogram and the pathologic nomogram were 1.0889(95%CI:0.6985-1.4793,p<0.001),1.1111(95%CI:0.7237-1.4985,p<0.001)and 0.9556(95%CI:0.5474-1.3637,p<0.001),respectively.The corresponding IDI values were 0.3308(95%CI:0.2149-0.4467,p<0.001),0.2221(95%CI:0.109-0.3352,p<0.001)and 0.1709(95%CI:0.0523-0.2895,p=0.005),respectively.The results of decision curve analysis showed that the net benefit of our radiomics nomogram was higher than that of CT diagnosis,preoperative nomogram and pathological nomogram under any threshold probability.Conclusion:The radiomics nomogram for prediction of lymph node metastasis has good prediction ability,and its prediction efficiency is better than CT diagnosis and the previously published nomograms. |
PDF Full Text Request