Construction Of A Model For Predicting Lymph Node Metastasis In Gastric Cancer Based On Dual-energy CT Radiomics

Part ?Construction of a lymph node metastasis prediction model based on dual-energy CT radiomics of gastric cancer lesionsObjective To explore the value of tumor markers of gastric cancer(GC),quantitative iodine concentration(IC)in arterial phase and venous phase of dual energy CT(DECT)scan,and radiomic in predicting lymph node metastasis(LNM).Methods Analyzed the data of 177 patients who received DECT scan before operation and confirmed GC by postoperative pathology in the Third Affiliated Hospital of Kunming Medical University.The tumor markers(CA125?CA199?CEA),IC and standardized iodine concentration(nIC)of GC focus in arterial phase(AP)and venous phase(VP)were collected.According to the pathological results,the patients were divided into lymph node metastasis group and non-metastasis group.Wilcoxon rank sum test was used to compare the tumor markers,IC,NIC,etc.in GC lesions of the two groups,and ROC curve was drawn to evaluate the effectiveness of predicting LNM.The feature extraction of image group by SIEMENS syngo.via Frontier Radiomics software,and Python Scikit-learn software was used to construct a random forest(RF)model.All patients were randomly divided into a training set and a test set to predict LNM at a ratio of 7:3.Using ROC curve to evaluate the diagnostic efficiency of RF model training set and test set;Combined with the independent predictive factors screened out above,a Nomogram is constructed using R language 3.5.2,randomly divided into a training set and a test set according to a 7:3 ratio,and its diagnostic efficacy is evaluated using ROC curve;The calibration curve was used to verify the accuracy of the Nomogram,and the clinical value of the RF model and Nomogram was evaluated by clinical decision curve analysis.Results Among the 177 GC patients,83 were in the lymph node metastasis group and 94 in the lymph node non-metastasis group.The parameters of CA125,CA199 and CEA serum tumor markers were statistically significant for the prediction of LNM(P<0.05).The ICs of AP and VP in the metastatic group were 2.63(2.3,3.00)mg/ml and 3.60(3.23,4.03)mg/ml,and the areas under the AUC curve were 0.83 and 0.91;nIC were 0.18(0.15,0.21)mg/ml.ml,0.78(0.65,0.86)mg/ml,the area under the AUC curve was 0.79,0.87,respectively.Both the IC and nIC in the metastatic group were higher than those in the non-metastatic group(P<0.05).It is of high diagnostic value to predict random metastasis of GC lymph nodes by extracting the radiomics of GC lesions to establish a random forest(RF)model.The AUC values of the training set and test set are 0.959 and 0.977 respectively;the Nomogram train set and test set predict LNM The AUC values were 0.996 and 0.976,respectively.Conclusion(s)Models based on preoperative serum tumor markers(CA125,CA199,and CEA)in patients with GC,quantitative DECT parameter values of the lesions(arterial and venous IC,nIC),and radiomics have a higher diagnosis of predicting LNM,The value of Nomogram in combination with multi-parameter construction is higher,and it can provide a reliable basis for preoperative evaluation of LNM.Part ?Construction of lymph node metastasis evaluation model based on dual-energy CT radiomics of lymph nodes in gastric cancerObjective To investigate the value of gastric cancer(GC)quantitative iodine concentration(IC)and radiomics of DECT scan of arterial and venous phases to identify the properties of gastric cancer lymph nodes.Methods Analyzed 265 lymph nodes of the Third Affiliated Hospital of Kunming Medical University who received DECT scans before surgery and whose properties were confirmed by postoperative pathology.The arterial(AP)and venous(VP)lymph nodes were measured for IC and standardized iodine concentration(nIC)and routine imaging parameters were collected.They were divided into metastatic group and non-metastatic group according to pathological results.The Wilcoxon rank sum test was used to compare the IC,nIC,and conventional imaging parameters(long diameter,short diameter,and short-to-length ratio)of the lymph nodes in the two groups,and the ROC curve was drawn to evaluate its effectiveness in identifying the properties of GC lymph nodes.The feature extraction of image group by SIEMENS syngo.via Frontier Radiomics software,and Python Scikit-learn software was used to construct a random forest(RF)model.All patients were randomly divided into a training set and a test set according to the ratio of 7:3 to identify the nature of GC lymph nodes.Using ROC curve to evaluate the diagnostic efficiency of RF model training set and test set;Combined with the independent predictive factors screened out above,a Nomogram is constructed using R language 3.5.2,randomly divided into a training set and a test set according to a 7:3 ratio,and its diagnostic efficacy is evaluated using ROC curve;The calibration curve was used to verify the accuracy of the Nomogram,and the clinical value of the RF model and Nomogram was evaluated by clinical decision curve analysis.Results Among the 265 lymph nodes,128 cases were in the metastatic group and 137 cases were in the non-metastatic group.Routine imaging parameters(long diameter,short diameter,and short-to-length ratio)were statistically significant for identifying the properties of GC lymph nodes(P<0.05).The arterial and venous ICs in the metastatic group were 2.48(2.00,3.06)mg/ml and 3.20(2.74,3.57)mg/ml,and the areas under the AUC curve were 0.619 and 0.751,the nIC was 0.18(0.14,2.22)mg,/ml,0.64(0.54,0.74)mg/ml,the area under the AUC curve is 0.576,0.677,respectively.The IC and nIC of lymph nodes in the metastasis group were higher than those in the non-metastasis group(P<0.05).Lymph node radiomics have high diagnostic value to identify its properties.Random forest(RF)model train set and test set predict the gastric cancer lymph node metastasis AUC values of 0.918 and 0.977,respectively;the Nomogram train set and test set to predict AUC of gastric cancer lymph node metastasis The values are0.991and 0.936.Conclusion(s)GC lymph nodes conventional imaging parameters(long diameter,short diameter,short length ratio),quantitative DECT parameter values(spectrum curve slope,arterial phase and venous phase IC,nIC),and radiomics have certain diagnosis to identify the properties of lymph nodes The value is not high.Based on the radiomics of GC lymph nodes and Nomogram constructed by multi-parameters,the diagnostic efficiency can be significantly improved,which can provide a basis for judging the properties of lymph nodes before surgery.