| PART1 The value of predicting Ki-67 expression based on CT-enhanced findings combined with radiomics of primary lesion of gastric cancerObjective: To investigate the value of predicting Ki-67 expression level of gastric cancer by combining CT-enhanced features and quantitative parameters with radiomics based on preoperative CT-enhanced arterial and venous phase imagings of primary lesion of gastric cancer.Methods: A total of 121 patients(81 males and 40 females)with postoperative pathologically confirmed advanced gastric cancer(AGC)and preoperative enhanced CT examination were retrospectively collected from June 2018 to September 2021.The included cases were divided into the Ki-67 high expression group(Ki-67>50%): 70 cases and Ki-67 low expression group(Ki-67≤50%): 51 cases.The CT-enhanced features of the primary lesion were collected: including location,maximum length diameter,radiology T stage(r T)and radiology N stage(r N).The CT-enhanced quantitative parameters were measured: The maximum cross section of primary lesion was selected on the plain scan,arterial phase and venous phase images respectively to measure the CT values of the third phase,and the arterial enhancement ratio(CERAP)and venous enhancement ratio(CERVP)were calculated.Based on the preoperative images of the primary lesion gastric cancer in arterial and venous phases,the ROI of all layers were manually delineated and fused into volume of interest finally.The 3D radiomic features were extracted respectively based on preoperative CT-enhanced images of the arterial and venous phase of primary lesion of gastric cancer.The GE A.K software was used to screen the optimal radiomic features related to Ki-67 expression level,and the radiomics signature was established.The multivariate logistic regression analysis was used to select the independent predictors related to Ki-67 expression,and the nomogram model was constructed.The predictive ability of the radiomics signature and nomogram model were measured with the receiver operating characteristic(ROC)curve,and the area under the curve(AUC)was calculated.The internal 10-fold cross-validation was used to explain the generalization of the predictive model.Results: There were statistically significant differences between the Ki-67 high and low expression groups of r T,plain CT values,CERAP and CERVP,and the r T and CERAP were independent predictors.A total of 1037 radiomics features were extracted based on CT-enhanced imagings of arterial phase and venous phase.Seven selected radiomics features(including 4 arterial phases and 2 venous phases)associated with the Ki-67 expression level were used to construct the radiomics signature,and the AUC value for predicting Ki-67 high expression level was 0.838(95%CI:0.767~0.909).The combined prediction model constructed by r T(OR=0.276,P=0.014),CERAP(OR=3.913,P=0.019)and radiomic signature(OR=191.766,P=0.000)increased the AUC value to 0.870(95%CI:0.807~0.933).The results of internal 10-fold cross-validation showed that the average C-index was 0.848.Conclusion: The radiomic signature based on CT-enhanced of primary lesion of gastric cancer have a good accuracy in predicting the Ki-67 expression level of gastric cancer before surgery.The nomogram model combined with r T,CERAP and radiomic signature further improve the performance for Ki-67 expression level.PART2 The value of predicting vascular invasion based on CT-enhanced manifestation combined with radiomics of primary lesion of gastric cancerObjective: To investigate the value of predicting vascular invasion status of gastric cancer by combining CT-enhanced features and quantitative parameters with radiomics based on preoperative CT-enhanced arterial and venous phase imagings of primary lesion of gastric cancer.Methods: A total of 121 patients(81 males and 40 females)with postoperative pathologically confirmed advanced gastric cancer(AGC)and preoperative enhanced CT examination were retrospectively collected from June 2018 to September 2021.The included cases were divided into a vascular invasion group:85 cases and a non-vascular invasion group:36 cases,according to postoperative pathological information.The CT-enhanced features and quantitative parameters of the primary lesion were collected.The 3D radiomic features were extracted respectively based on preoperative CT-enhanced images of the arterial and venous phase of primary lesion of gastric cancer.The GE A.K software was used to screen the optimal radiomic features related to vascular invasion status,and the radiomics signature was established.The multivariate logistic regression analysis was used to select the independent predictors related to vascular invasion,and the nomogram model was constructed.The predictive ability of the radiomics signature and nomogram model were measured with the ROC curve.The internal 10-fold cross-validation was used to explain the generalization of the predictive model.Results: There were statistically significant differences between with and without vascular invasion groups of r T,r N and CT values of arterial phase,and the r N was an independent predictor.A total of 1037 radiomics features were extracted based on CT enhanced imagings of arterial phase and venous phase.Four selected radiomics features(including 3 arterial phases and 1 venous phase)that were associated with vascular invasion status were used to construct radiomics signature,and the AUC value for predicting vascular invasion status was 0.794(95%CI:0.713~0.876).The combined prediction model constructed by r N(OR=5.597,P=0.001)and radiomic signature(OR=229.507,P=0.000)increased the AUC value to 0.834(95 %CI:0.756~0.912).The results of internal 10-fold cross-validation showed that the average C-index was 0.828.Conclusion: The radiomic signature based on CT-enhanced of primary lesion of gastric cancer have a good accuracy in predicting the vascular invasion status of gastric cancer before surgery.The nomogram model combined with r N and radiomic signature improved from the single radiomic signature in predicting vascular invasion in gastric cancer. |
PDF Full Text Request