Study Of CT Radiomics Features Of Metastatic Lymph Node Secondary To Esophageal Squamous Cell Carcinoma

Part one:Development of CT Radiomics Models of Primary Tumor and Metastatic Lymph Node of Esophageal Squamous Cell Carcinoma and Its Study on Preoperative Diagnosis of Metastatic Lymph NodeObjective:To develop and validate CT radiomics models and Radiomic-clinical model of primary tumor and metastatic lymph node(MLN)of esophageal squamous cell carcinoma(ESCC)for diagnosis of MLN secondary to ESCC.Methods:The CT images and clinical data of 249 patients with ESCC confirmed by biopsy were collected,among them,there were 120 patients with MLN and 129 patients without MLN.Image segmentation and extraction of radiomics features were performed by MATLAB.Univariate analysis and the least absolute shrinkage and selection operator method were operated by R-studio for dimension reduction and optimal features selection,and logistic regression analysis was conducted to develop primary tumor and nodal radiomics models.For clinical data,the quantitative data were compared by independent sample t test or Mann Whitney U test,and the qualitative data were compared by Chi-square test or Fisher test.Developing theradiomic(LN+ESCC)-clinical model by combining the radiomics and clinical features.The diagnosis performance of models was estimated by area under the curve(AUC),accuracy and other indicators of receiver operator characteristic curve.Results:Eight and nine optimal radiomics features were extracted from primary tumor of ESCCand lymph nodes,respectively,to construct radiomics models.T stage and LN distribution region combined with the above two types of radiomics features to develop aradiomic(LN+ESCC)-clinical model.The AUC ofradiomic(LN+ESCC)-clinical model,radiomics model of lymph node and primary tumor to preoperatively diagnose MLN secondary to ESCC in the training cohort were 0.948,0.918 and 0.721,respectively,and in the validation cohort were 0.885,0.877 and 0.656,respectively.Conclusion:The diagnostic efficacy of the radiomics model of ESCC primary tumor is limited for the preoperativediagnoses of MLN,while the LNradiomics model outperforms the model of primary tumor.The model of combined radiomics and clinical features is superior to a single radiomics model,which shows that developing model with the combination of imaging and clinical information of ESCC have preferable performance in diagnosing MLNpreoperatively.Part two:Study of CT Radiomics Model of Metastatic Lymph Node with Maximum Short Diameter less than 1 cm Secondary to Esophageal Squamous Cell CarcinomaObjective:To develop and test the CT radiomics model ’of metastatic lymph node(MLN)with maximum short axis diameter(MSAD)<1 cm secondary to esophageal squamous cell carcinoma,and investigate the diagnostic value of the model for diagnosing MLN.Methods:Enhanced CT images and related clinical information of 146 consecutive patients with ESCC confirmed by biopsy were collected continuously.Among the patients,the LN with MS AD<1 cm of 30 patients were MLN,and LN with MSAD<1 cm of 111 patients were non-MLN.Using MATLAB to draw the region of interestof LN with MSAD<1cm and extract the radiomics features.Univariate and LASSO analysis were conducted on R-studio for dimension reduction and features selection.Univariate analyses were performed on SPSS to compare the clinical features.And the selected radiomics and clinic features were analyzed by logistic regression to build the radiomics and radiomic-clinical models of LN to diagnose MLN with MSAD<1cm secondary to ESCC.Ultimately,the diagnostic performance of the two models in training and testing cohorts were evaluated with receiver operator characteristic curve.Results:9 optimal radiomics features were extracted from LN with MSAD<1cmwere selected to build a radiomics model,3 clinical features(including primary tumor location,LN distribution region and MSAD)combined with LN radiomics features to build a radiomic-clinical model to diagnose the MLN with MS AD<1cm secondary to ESCC.The AUCof radiomic-clinical model and LN radiomics model in the training cohort were 0.893 and 0.860,and the accuracy were 0.862 and 0.853,respectively;in the testing cohort,the AUC were 0.886 and 0.858,and the accuracy were 0.867 and 0.867,respectively.Conclusion:The LNradiomics model has good performance in diagnosingMLN with MSAD<1cm secondary to ESCC,which provides a new idea for clinical diagnosis.The radiomics characteristics of ESCC primary tumor can not effectively predict the metastasis of small lymph nodes.