A Computed Tomography-Based Clinical-Radiomics Model For Prediction Of Lymph Node Metastasis And Infiltration Depth In Esophageal Carcinoma

Part 1.A computed tomography-based clinical-radiomics model for prediction of lymph node metastasis in esophageal carcinomaAims:Esophageal carcinoma(EC)is one of the most common malignancies in the digestive tract.Presently,surgical resection remains the mainstay of treatment.Accurate evaluation of the status of lymph node metastasis(LNM)is crucial for the determination of treatment plans and prognosis.Currently,assessment of LNM in EC is largely dependent on histopathologic evaluations of surgical specimens,which are only available after surgery.Computed tomography(CT)is commonly performed before surgery,but its accuracy in identifying LNM preoperatively is limited.The prediction of lymph node status in EC is critical for clinical treatment.In clinical practice,CT has been frequently used to determine the location,anatomical relationships,and significant metastasis of EC.CT radiomics features can further provide information to reflect potential biological heterogeneity.Single-phase CT radiomics features for the preoperative diagnosis of LNM in patients with EC have been previously identified.Less studies have yet investigated the use of multiscan CT radiomics features in conjunction with clinical factors to predict LNM in EC.Therefore,we aimed to develop the best CT radiomics model based on multiscan CT and clinical factors to predict LNM in EC.Subjects and Methods:This is a retrospective study of 195 patients with biopsy-proven EC.All patients underwent three-phases CT scan,and the original images were uploaded to the multimodal scientific research platform.The regions of interest(ROI)were manually outlined and the relevant radiomics features were selected by Intraand interclass correlation coefficients(ICC)tests,the analysis of variance and the least absolute shrinkage and selection operator(Lasso).Then,all cases were randomly divided into training and testing cohort at a ratio of 7:3.A univariate logistic regression was used to establish the radiomics model,and the testing cohort was used for verification.The receiver operating characteristic curve(ROC)was drawn for the training and testing groups,and the area under the ROC curve(AUC)was calculated to select the best model.In addition,two clinical features,i.e.,CT reports of LNM and tumor locations,were evaluated in the training cohort by univariate logistic regression analysis,and the variables with a P-value<0.05 were further included in the development of radiomics model.Eventually,the decision curve analysis was conducted to quantify the net potential benefits of including the variables.Results:Seven radiomics models were established based on features identified on single-phase images(plain,P;arterial phase,A;venous phase,V)and multi-phase images(P+A,P+V,A+V,P+A+V).The model that included 26 features derived from P+A+ V had the best predictive value in the training cohort(AUC=0.783)and testing cohort(AUC=0.741).Following the univariate analysis,CT report of LNM was significantly associated with the status of LNM(P<0.001),whereas tumor location failed to show any significant association(P=0.429).Therefore,the CT report of LNM was added to the radiomics model to establish a clinical-radiomics model.The addition of CT report of LNM resulted in remarkable improvements in the AUC values from 0.783 to 0.814,respectively in the training cohort(p=0.024).Conclusions:A clinical-radiomics model based on radiomics features of multi-phase contrast-enhanced CT study and CT report of LNM may be useful for predicting LNM in patients with EC.Further studies are warranted to confirm these findings.Part 2.A computed tomography-based clinical-radiomics model for prediction of infiltration depth in esophageal carcinomaAims:EC is one of the common malignancies that endanger human health.Accurate evaluation of the stages in EC is crucial for the determination of treatment plans and prognosis.The most commonly used staging guidelines for EC is the TNM staging system of the American Joint Committee on Cancer(AJCC)/Union for International Cancer Control(UICC).The TNM stage includes the depth of invasion of the primary tumor(T),the number of regional lymph node metastases(N),and the presence or absence of distant metastases(M).Predicting the depth of tumor infiltration in EC is important for clinical treatment.For patients with EC,surgical resection remains the primary treatment.Accurate assessment of the depth of tumor invasion in esophageal cancer is key to determining treatment options and prognosis.In clinical practice,determining the depth of infiltration in EC often requires invasive histopathological evaluation of biopsy samples.CT is usually examined preoperatively to determine the location,anatomical relationship,and significant metastasis of the EC,but its accuracy in predicting the preoperative depth of tumor invasion is very limited.In recent years,radiomics techniques established using CT images have been gradually used for the clinical evaluation of tumor patients,and radiomics features can further provide information reflecting the underlying biological heterogeneity.Therefore,we aimed to establish the optimal CT radiomics model based on multiscan CT to predict the depth of invasion in EC.Subjects and Methods:This is a retrospective study of 226 patients with clinical-proven EC.We defined T1-T2 stage as non-deep infiltration(n=92)and T3-T4 stage as deep infiltration(n=134).All patients underwent three-phases CT scan,and the original images were uploaded to the multimodal scientific research platform.The ROI were manually outlined and the relevant radiomics features were selected by ICC tests and the analysis of variance.Then,all cases were randomly divided into training and testing cohort at a ratio of 7:3.A univariate logistic regression was used to establish the radiomics model,and the testing cohort was used for verification.The model with the best AUC value was selected among all the radiomics models established.In addition,three clinical features,i.e.,CT reports of LNM and tumor locations and peritumoral adipose space,were evaluated in the training cohort by univariate and multivariate logistic regression analysis,and the variables with a P-value<0.05 were further included in the development of radiomics model.Eventually,the decision curve analysis was conducted to quantify the net potential benefits of including the variables.40 patients from another hospital with esophageal cancer were selected in this experiment,and external verification was also conducted to test the stability of the model.Results:Seven radiomics models were established based on features identified on single-phase images(plain,P;arterial phase,A;venous phase,V)and multi-phase images(P+A,P+V,A+ V,P+ A+V).The model that included 23 features derived from arterial phase had the best predictive value in the training cohort(AUC=0.905)and testing cohort(AUC=0.863).Following the univariate and multivariate logistic regression analysis,the peritumoral adipose space was selected as a clinical feature for building a clinical-imaging radiomics model in arterial phase.The AUC of the training cohort was 0.915.There was no significant correlation with the improvement of predictive performance in arterial phase radiomics model(P=0.273).External validation of the arterial phase model was performed,and the AUC value in testing cohort was 0.761.Conclusions:A radiomics model based on radiomics features of arterial phase CT study may be useful for predicting infiltration depth in patients with EC.Further studies are warranted to confirm these findings.