The CT Radiomics Features To Predict Lymph Node Metastasis In Esophageal Squamous Cell Carcinoma

Objectives:To develop CT radiomics features for prediction lymph nodes(LNs)status in esophageal squamous cell carcinoma(ESCC)patients.Methods:This retrospective study was composed 334 patients with endoscopic biopsy-confirmed ESCC,including 152 patients of without LNM cases,103 regional lymph node metastases(RLNM)cases and 79 non-regional LNM(NRLNM).Their lymphnode status was pathologically confirmed after regional lymphadenectomy,extended regional lymphadenectomy or pathologically confirmed by fine needle aspiration.The thoracic contrast-enhanced CT image of the ESCC patient was manually delineated layer-by-layer with the ESCC region of interest visible at the edge of the lesion,and the tumor image features were extracted.All features were drawn and calculated layer by layer at the two-dimensional level,and the calculation results were reconstructed in three dimensions.In order to ensure the repeatability of the results,z-score normalization was performed as a preprocessing step for all data.After a reproducibility test,a univariate analysis(independent samples t-test or Mann-Whitney U test)and the least absolute shrinkage and selection operator were exploited for dimension reduction and selection of the features.A classical machine learning method named logistic regression was used to produce two predictive radiomics models based on selected CT features previously identified in training cohort to discriminate patients without LNM from those with LNM,and further to distinguish patients with RLNM from those with NRLNM in patients with LNM.The first step is to train the model with the training cohort to obtain the best coefficients of the features with the best model diagnostic performance.The second step is to validate the model with the validation cohort.The area under receiver-operating characteristic curve(AUC),accuracy,sensitivity and specificity assessed the discriminating performance.Results:The radiomics features were developed based on a multivariable logistic regression,which were significantly associated with LNM(P<0.001)in training cohort and validation cohort.AUC,accuracy,sensitivity and specificity of radiomics features could help differentiate between patients of without LNM and patients of with LNM(0.79,95%CI:0.714?0.863;0.75;0.98;0.56 in the training cohort,0.75,95%CI:0.637?0.867;0.71;0.71;0.71 in the validation cohort,respectively).In patients with LNM,AUC,accuracy,sensitivity and specificity of radiomics features could help differentiate between RLNM and NRLNM(0.98,95%CI:0.97?0.99;0.94;0.97;0.91 in the training cohort;0.95,95%CI:0.89?0.90;0.83;0.81;0.86 in the validation cohort,respectively).Conclusion:The present study showed CT radiomics features that could be conveniently used to facilitate the prediction of LNs status in patients with ESCC.