The Value Of Assessing Lymph Node And Vascular Metastasis In Esophageal Cancer Based On CT Radiomics Model

The first part Thevalueofassessing lymphnodemetastasisinesophageal cancerbasedon CTradiomics modelObjection:Exploring the value of predicting lymph node metastasis in esophageal cancer basedon CTradiomics model.Methods: The clinical and imaging data of 253 patients with surgically pathologically confirmed esophageal squamous carcinoma from January 2015 to May 2022 in Yijishan Hospital of Wanan Medical College were retrospectively analyzed.According to the presence or absence of lymph nodemetastasis in the postoperativepathology,esophageal cancerpatients weredivided into the group with lymph node metastasis(n=111)and the group without lymph node metastasis(n=142),and patients were randomly divided into the training group(n=178)and the validation group(n=75)according to the ratio of 7:3.All patients underwent chest plain and enhanced scans within 2 weeks before surgery.The plain scan,arterial phase and venous phase images were imported into ITK-SNAP software,and the ROI was manually outlined along the lesion layer by layer in each phase image,respectively,and fused into a three-dimensional volumetric region of interest,and texture features were extracted using FAE software,and the texture features were downscaled using minimum redundancy maximum correlation(m RMR)and minimum absolute value convergence and selection operator(LASSO)regression analysis to establish plain scan,arterial phase Theradiomicslabels wereestablished forplain,arterial,venous andcombined phases.Multi-factor logistic regression analysis was used to establish clinical models and personalized diagnostic models combined with radiomics models.The efficacy of each model in predicting esophageal cancer lymph node metastasis was evaluated using subject operating characteristic(ROC)curves,and the clinical application value of the models was assessed using decision curve analysis(DCA).Results: The clinical models had an area under the curve(AUC)of 0.65 and 0.68 in the trainingand validation groups,and CT-reported lymphnodesigns(OR=3.296,P=0.004)and tumor length(OR=1.334,P=0.020)were independent predictors.The AUCs of the combined radiomics ofplain,arterial,venous,andtriphasic phases were 0.78and0.81,0.80 and0.82,0.83 and 0.86,and0.81 and 0.86 in the training and validation groups.The personalized diagnostic model obtained by combining the venous phaseradiomics label with clinicopathological and imaging features with relativelybetterpredictiveefficacy had AUCs of0.85 and0.88 inthe training andvalidation groups,with the rad-score value(OR=2.128,P<0.001)and CT-reported lymph node sign(OR=2.904,P=0.032)as independentpredictors ofthemodel factors.Conclusion: 1.Among the clinicopathologic data and CTfeatures,the presence or absence of preoperativelymphnodemetastasisinesophageal cancerisrelated to CT-reported lymphnodesign and tumor length diameter,and patients with positive CT-reported lymph node sign and larger tumorlengtharemorelikely tohavelymphnodemetastasis.2、Planar,arterial,venous and combined radiomics models have better efficacy in predicting preoperative lymph node metastasis in esophageal cancer,among which venous imaging histologicalmodelsarerelativelybetterthanclinical models.3.The intravenous phase radiomics model and the personalized model combined with clinical data and CT features were the most effective in predicting lymph node metastasis of esophageal cancer,slightly better than the intravenous phase radiomics model and significantly betterthantheclinicalmodel.the second part Evaluationofshort-termefficacy ofbrainmetastasis chemotherapyfornon-smallcell lung cancerbasedon MRI radiomicsmachinelearningmodelObjection:Exploring the value of predicting choroidal carcinoma thrombus invasion in esophagealcancerbasedon CTradiomicsmodelMethods: The clinical and imaging data of 197 patients with surgically pathologically confirmed esophageal squamous carcinoma from January 2015 to December 2022 in Yiji Mountain Hospital of Wanan Medical College were retrospectively analyzed.According to the presence or absence of lymph node metastasis in postoperative pathology,esophageal cancer patients were divided into the group with choroidal carcinoma thrombus invasion(n=75)and the group withoutchoroidal carcinoma thrombus(n=122),and patientswere randomly divided intothe training group(n=139)and the validation group(n=58)according to the ratio of 7:3.All patients underwent chest plain and enhanced scans within 2 weeks before surgery.The plain scan,arterial phase and venous phase images were imported into ITK-SNAP software,and the ROI was manuallyoutlined along thelesionlayerbylayerineachphaseimage,respectively,and fusedinto a three-dimensional volumetric region of interest,and texture features were extracted using FAE software,and the texture features were downscaled using minimum redundancy maximum correlation(m RMR)and minimum absolute value convergence and selection operator(LASSO)regression analysis to establish plain scan,arterial phase The CT radiomics model labels were established for plain,arterial,venous and combined phases.Multi-factor logisticregression analysis was used to establish clinical models and personalized diagnostic models combined with CT radiomics model The efficacy of each model in predicting lymph node metastasis in esophageal cancer was evaluated using subject operating characteristic(ROC)curves,and the clinical application valueofthemodelswasassessed using decisioncurveanalysis(DCA).Results: The area under the curve(AUC)of the clinical model was 0.65 and 0.70 in the training and validation groups,and the degree of tumor differentiation(OR=4.173,P=0.008)and tumor length(OR=1.548,P=0.009)were independent predictors.The AUCs of the combined radiomics model of plain,arterial,venous,and three stages were 0.72 and 0.73,0.74 and 0.75,0.75 and 0.76,and 0.75 and 0.75 in the training and validation groups,respectively.DCA showed that both the personalizeddiagnosticmodel andthe intravenousimaging histologylabelhad a goodnetbenefit.Conclusion: 1.Among the clinicopathological data and CT features,esophageal cancer patients with hypodifferentiated tumor and larger tumor length are more likely to have lymph node metastasismetastasis.2、Planar,arterial,venous and three-stage combined imaging histological models have better efficacy in predicting preoperative esophageal cancer pulsatile metastasis,among which venous imaginghistologicalmodelsarerelatively betterthanclinicalmodels.3.The intravenous phase imaging histological model and the personalized model combined with clinical data and CT features had the best effect in predicting pulsatile thrombus invasion of esophagealcancer,which wasslightly betterthan theintravenous phaseimaging histological model andsignificantlybetterthantheclinical model.