Application Of Radiomics In Predicting The Degree Of Pathological Differentiation And Lymph Node Metastasis Of Esophageal Squamous Cell Carcinoma

Part Ⅰ: Application of different radiomics diagnostic models based on enhanced CT in the preoperative prediction of esophageal squamous cell carcinoma differentiationObjective: This study is to establish radiomics models of preoperative pathological differentiation degree of esophageal squamous carcinoma by using venous phase enhanced CT images and different machine learning models,exploring the feasibility and value of different radiomics diagnostic models in predicting the differentiation degree of esophageal squamous carcinoma.Methods: 172 cases of esophageal squamous carcinoma confirmed by operation and pathology in our hospital were retrospectively analyzed.They were divided into two groups:well differentiated group and poorly differentiated group.There were 114 cases and 58 cases respectively.The venous phase enhanced CT images were imported into the yizhun-ai system in DICOM format and randomly divided into the training set and the test set according to the ratio of 3:1.The radiologic features of CT images in venous phase were extracted,and the dimensionality was reduced by the minimum and maximum normalization method,optimal feature screening and the least absolute shrinkage and selection operator method.Then support vector machine model,random forest model and logistic regression model were constructed respectively.The three models were trained by 5-fold cross-validation,and the ROC curves of the training set and the test set were drawn to evaluate the diagnostic efficiency.Results: The AUC of the three radiomics diagnostic models were greater than 0.7,and the AUC of the support vector machine model was the highest,the AUC was 0.88.There was no significant difference in clinical inflammatory indexes among different differentiation groups（P>0.05）.There was no significant difference in clinical characteristics between the training set and the test set（P>0.05）.Conclusion: The three radiomics diagnostic models based on enhanced CT can be used to predict the degree of pathological differentiation of esophageal squamous carcinoma before operation.Part Ⅱ: Application of nomogram based on enhanced CT imaging in predicting lymph node metastasis of esophageal squamous carcinomaObjective: To investigate the feasibility and value of radiomics nomogram based enhanced CT for predicting lymph node metastasis in esophageal squamous carcinoma.Methods: We retrospectively reviewed the clinical data of 172 patients with postoperative confirmed esophageal squamous cancer（138 in the training set and 34 in the test set）.The radiologic features features were extracted from the venous CT images of each patient by the yizhun-ai system,and then the minimum maximum normalization method,optimal feature screening and minimum absolute contraction and selection operator regression were used to gradually screen the features,and the radiomics score were established.Clinical risk factors were identified by univariate and multivariate logistic regression analysis.Three models were constructed: a clinical model based on clinical risk factors,a radiomics model based on radiomics score,a clinical-radiomics nomogram based on combined risk factors and radiomics score.Using the receiver operating characteristic and the area under curve to evaluate the models diagnostic efficacy.Analysis of calibration curve and decision curve to determine the prediction accuracy and clinical benefits of nomogram.Results: The nomogram for predicting lymph nodes in esophageal squamous cell carcinoma include radiomics score,NLR ratio and CT reported lymph node status.The AUC of clinical-radiomics model is 0.777 in the training set and 0.751 in the test set,which has the best prediction performance.The calibration curve of nomogram shows that the predicted probability is in good agreement with the observed actual reflection probability.The decision curve shows good clinical practicability.Conclusion: The clinical-radiomics nomogram established by Rad-score,NLR ratio and CT reporting of lymph node status has good predictive efficiency.It shows that the nomogram is helpful to guide clinical practice and has potential clinical application value.