| CT examination has been extensively used in preoperative diagnosis,staging,prediction of prognosis and evaluation of curative effect,etc.of gastric cancer,but it is mostly limited to routine morphological analysis.The research on radiomics has gradually become a hot topic in recent years.Studies show that radiomics is significantly superior to traditional clinical factors in the aspects of differentiating benign and malignant tumors,preoperative grading,evaluating treatment response,prognosis,etc..Presently,studies on the prediction of postoperative recurrence/metastasis and disease free survival(DFS)of gastric cancer is relatively less.Therefore,the study is to investigate the CT radiomics method's clinical value in predicting postoperative recurrence/metastasis in patients with gastric cancer and DFS.Part ? The value of radiomics in predicting early recurrence/metastasis of gastric cancer after operationObjective:To investigate the value of radiomics in predicting early recurrence/metastasis(?1 year)of gastric cancer after operation based on CT enhanced radiomics model.Methods:153 patients(113 males and 40 females)performed with gastric cancer resection,pathologically diagnosed as gastric adenocarcinoma and received follow-up visit regularly for at least 1 year were included retrospectively.They were randomly divided into training group(n=106)and validation group(n=47).The clinical risk factors(age,sex,etc.)associated with early recurrence/metastasis of gastric cancer,basic radiomics features(maximum tumor diameter,preoperative T staging,N staging,etc.),tumor indicators(CEA,CA-199,etc.)and postoperative pathological risk factors(VEGF,EGFR,etc.)were analyzed to establish a clinical model in predicting early recurrence/metastasis.3D-slicer software was used to draw the region of interest(ROI)of gastric cancer lesions in all the layers in 1.25mm layer-thickness images in venous phase,so as to extract radiomics features.LASSO analysis was utilized to select effective texture features to establish the radiomics signature,and to establish a comprehensive prediction model in combination with clinical prediction model.So that the receiver operating characteristic curve(ROC)and area under the curve(AUC)of the clinical prediction model,radiomics signature and comprehensive prediction model in predicting the early recurrence/metastasis in the subjects with gastric cancer(?1 year)after operation were calculated;Delong test was adopted to compare the AUC values of postoperative clinical prediction model,radiomics signature and comprehensive prediction model of gastric cancer.Furthermore,radiomics signature and other independent predictors were used to establish a nomogram model to predict the risk of early recurrence/metastasis.Finally,the decision curve analysis(DCA)was calculated and its benefits of the clinical model and radiomics signature were compared.Results:Out of a total of 1409 texture features which were extracted from CT 1.25mm layer-thickness images in venous phase,25 effective texture characteristic values were selected to establish the radiomics signature.The multivariate analysis on overall data showed that radiomics signature before and after surgery were independent risk factors for predicting early recurrence/metastasis of gastric cancer(P<0.001);among clinical factors,age(P=0.046),VEGF(P=0.032)and EGFR(P=0.019)were independent risk factors for predicting early recurrence/metastasis after operation.The AUC values under the ROC curves of radiomics signature,clinical prediction model and comprehensive prediction model were 0.732,0.648 and 0.774 respectively.The ability of comprehensive prediction model and clinical prediction model in predicting early recurrence/metastasis of gastric cancer was significantly different(P=0.003).The AUC of Nomogram prediction model and radiomics signature were 0.802 and 0.809 respectively,which were significantly different from other indexes(P<0.001).The decision curve analysis chart(DCA)showed that the prediction probability of Nomogram prediction model was about 10%and 85%;the threshold probability of radiomics signature was about 15%and 75%.Conclusion:Radiomics signature is an independent predictor in predicting early recurrence/metastasis of gastric cancer after operation.It can significantly improve the ability of clinical model in predicting early recurrence/metastasis of gastric cancer after operation.It is better for Nomogram and radiomics signature to predict the risk of early recurrence/metastasis of gastric cancer.Part ? Comparison of radiomics between early recurrence/metastasis and late recurrence/metastasis in Patients with Postoperative Gastric CancerObjective:To investigate the correlation of radiomics signature between early recurrence/metastasis(?1 year)and late recurrence/metastasis(>1 year)in patients with postoperative gastric cancer.Methods:84 patients with postoperative recurrence/metastasis of gastric cancer were retrospectively included and randomly divided into training group(n=58)and validation group(n=26).The clinical risk factors associated with recurrence/metastasis(age,sex,etc.),basic radiomics features(maximum tumor diameter,preoperative T staging,N staging,etc.),tumor indicators(CEA,CA199,etc.),and postoperative pathological risk factors(VEGF,EGFR,etc.)were analyzed to establish a clinical model in predicting early recurrence/metastasis and late recurrence/metastasis.3D-slicer software was used to draw the region of interest(ROI)of gastric cancer lesions in all the layers,so as to extract radiomics features.LASSO analysis was utilized to select effective texture features to establish the radiomics signature.The differences in Rad-score(rs)values between early recurrence/metastasis and late recurrence/metastasis in the training group and validation group were analyzed.The AUC,optimal cut-off values,sensitivity and specificity in the training group and validation group in predicting early recurrence/metastasis were analyzed.Univariate and multivariate analysis were made on various risk factors of training group to select independent predictor in distinguishing the early recurrence/metastasis and late recurrence/metastasis of gastric cancer after operation.Results:Out of a total of 1409 texture features which were extracted from CT 1.25mm layer-thickness images in venous phase,9 effective texture characteristic values were selected finally to establish the radiomics signature.There was significant statistical significance for radiomics signature between early recurrence/metastasis and late recurrence/metastasis in the training group(P<0.001).There was significant significance for radiomics signature between early recurrence/metastasis and late recurrence/metastasis in the validation group(P=0.009).The optimal cut-off value of radiomics signature in distinguishing early recurrence/metastasis from late recurrence/metastasis in the training group was 0.724,with the AUC value of 0.907(95%CI:0.833-0.982),the specificity of 0.905,and the sensitivity of 0.784.The optimal cut-off value of radiomics signature in distinguishing early recurrence/metastasis from late recurrence/metastasis was-1.437,with the specificity of 0.500,the sensitivity of 0.812,and the AUC value of 0.619(95%CI:0.388-0.850).Univariate analysis showed that Rad-score(P<0.001),CA-199(P=0.038),preoperative T staging(P=0.021)and EGFR(P=0.038)were of statistical significance in differentiating early recurrence/metastasis from late recurrence/metastasis in the training group.Multivariate analysis showed that radiomics signature was of statistical significance in differentiating early recurrence/metastasis from late recurrence/metastasis of gastric cancer(P<0.001).Conclusion:Conclusion:Radiomics signature is an independent risk factor for differentiating early recurrence/metastasis from late recurrence/metastasis of gastric cancer.RS values of early recurrence/metastasis and late recurrence/metastasis were of significant difference.The training group had good efficacy in distinguishing early recurrence/metastasis from late recurrence/metastasis,and the validation group had poor efficacy in distinguishing early recurrence/metastasis from late recurrence/metastasis.Part ? Exploration of the value of radiomics signature in predicting disease free survival(DFS)of patients with gastric cancer after operationObjective:To explore the value of radiomics methods in predicting disease-free survival(DFS)of patients with gastric cancer after operation.Methods:179 patients(129 males and 50 females)with gastric adenocarcinoma pathologically diagnosed were included retrospectively(the period of regular follow-up visit>6 months)and randomly divided into training group(n=124)and validation group(n=55).The clinical factors(age,sex,etc.),radiomics(maximum tumor diameter,preoperative T staging,N staging,etc.)and postoperative pathological risk factors(VEGF,EGFR,etc.)associated with DFS were studied.The effective texture features were screened by LASSO regression and the radiomics signature were established.X-tile software was used to process the radiomics signature values(Rad-score),and patients were grouped according to different recurrence/metastasis risk.The follow-up deadline was decided according to recurrence/metastasis.Fisk factors affecting DFS were analyzed and subgroup analysis was conducted.The clinical model of predicting DFS of patients with gastric cancer was established by Cox regression and the comprehensive model of radiomics signature was added.The AUC values of clinical model and comprehensive model were compared by using Delong test.Nomogram chart was drawn,and the risk of recurrence/metastasis of gastric cancer was calculated and predicted.Consistency index(C-index)was used to evaluate the predictive ability of the model,and its benefits of clinical models and radiomics signature were compared by drawing DCA chart.Results:At the end of follow-up visit,there were 84 patients in total who were attacked by recurrence/metastasis of gastric cancer,and 95 patients whose data were missing.A total of 1409 texture features were obtained from 1.25mm venous-phase images,and 7 effective features out of 1409 texture features were selected by LASSO regression to form the radiomics signature.Rad-score values were of statistical significance for preoperative patients with different prognosis(P<0.001).X-tile was used to divide the included patients into high,moderate and low-risk groups.The recurrence/metastasis ratio(78.38%)in the high-risk group was significantly higher than that in the moderate-risk group and the low-risk group.Univariate and multivariate analysis showed that radiomics signature(HR=0.770;95%CI:0.635-0.934;P=0.008),CA-199(HR=0.523;95%CI:0.293-0.934;P=0.029)and lymphatic metastasis(HR=2.026;95%CI:1.028-3.992;P=0.041),CD34(HR=2.065;95%CI:1.070-3.985;P=0.031)predicting DFS was of statistical significance;AUC values of radiomics signature,clinical prediction model and comprehensive prediction model were 0.770,0.738 and 0.837 respectively.The ability comparison of the comprehensive model and clinical model in predicting DFS was of statistical significance(P=0.002).Nomogram model showed that the probability of DFS at 1 year,2 years and 3 years after operation was 0.5-0.85,0.2-0.75 and 0.1-0.6 respectively.The calibration curve showed that the prediction ranges of 2-year DFS of training group and validation group were larger than that of 1-year and 3-year DFS.C-index of the training group and validation group was 0.824 and 0.705 respectively.The decision curve analysis(DCA)showed that the threshold probability of the Nomogram model was about 15%-90%,the threshold probability of the radiomics signature was about 20%-75%,and the threshold probability of the traditional clinical model was about 30%-85%..Conclusions:(1)Radiomics signature is an independent risk factor in predicting DFS,and it can improve the ability of traditional clinical model in predicting disease free survival(DFS)after gastric cancer resection.(2)The efficacy of comprehensive model based on radiomics signature and clinical risk factors in predicting DFS was remarkably higher than that of clinical model.(3)The Nomogram chart could visualize the scores of various risk factors.The calibration curve showed that the prediction efficacy of the 2-year DFS was better.DC A showed that the DFS of gastric cancer patients could be predicted by Nomogram in a wider range. |
PDF Full Text Request