CT-Based Radiomics For Predicting The Lymphovascular Invasion And Prognosis Of Esophageal Squamous Cell Carcinoma:A Preliminary Study

Part I Computed Tomography-Based Radiomics Analysis to Predict Lymphovascular Invasion in Esophageal Squamous Cell Carcinoma: A Preliminary StudyObjective: To explore the value of CT radiomics in predicting lymphovascular invasion(LVI)before surgery in esophageal squamous cell carcinoma(ESCC).Methods: A retrospective analysis of 294 ESCC patients(66 LVI-present ESCC patients and 228 LVI-absent ESCC patients)with definite LVI status on postoperative pathological results.Their preoperative chest enhanced CT arterial images were used to delineate the region of interest of the lesion.Radiomics features were extracted from single-slice,three-slice and full-volume regions of interest(ROIs).All patients were randomly(stratified sampling)divided into a training cohort and a validation cohort at a ratio of 7:3.In the training cohort,the max-relevance and min-redundancy(m RMR)algorithm and the least absolute shrinkage and selection operator(LASSO)Logistic regression method were applied to select valuable radiomics features for identifying LVI status and calculate the radiomics score(Rad-score).The receiver operating characteristic curve(ROC)was used to evaluate the prediction performance of the three models for LVI status.Delong test was used to compare the difference of prediction performance of the three models.The stability and reliability of the radiomics models were validated using leave group out cross-validation(LGOCV)method.Decision curve analysis(DCA)was used to compare the clinical net benefit by each model.Each model was validated in the validation cohort.Results: A total of 1218 radiomics features were separately extracted from single-slice ROIs,three-slice ROIs and full-volume ROIs,16,13 and 18 features,respectively,were retained after optimization and screening to construct a radiomics model to predict the LVI status of ESCC.For the single-slice,three-slice,and full-volume models,the area under the curve(AUC)values of the training cohort were 0.797(95%CI: 0.727-0.867),0.774(95%CI: 0.701-0.847)and 0.824(95%CI:0.760-0.887),respectively,and the AUC values for the validation cohort were 0.666(95%CI: 0.532-0.800),0.708(95%CI: 0.578-0.838)and 0.738(95%CI: 0.628-0.847),respectively.According to LGOCV,the mean AUC values of single-slice,three-slice,and full-volume radiomics models of the training cohort were 0.698,0.782 and 0.786,respectively,and the mean AUC values of the validation cohort were 0.629,0.639 and 0.701,respectively.The Delong test showed that the full-volume radiomics model had a higher performance than the single-slice and three-slice radiomics models(P<0.05).DCA confirmed the clinical utility of each model.Conclusion: CT-based radiomics models were useful to predicte the preoperative LVI status in ESCC,and the full-volume radiomics model has the best prediction performance.Part II Computed Tomography-Based Radiomics Nomogram Model for Predicting the Prognosis of Esophageal Squamous Cell Carcinoma: A Preliminary StudyObjectives: To explore the predictive value of radiomics nomogram combining with CT radiomics features and clinical features for postoperative recurrence-free survival(RFS)and overall survival(OS)in patients with ESCC.Methods: A total of 294 patients with ESCC confirmed by surgery and pathology who were followed up regularly were retrospectively included.Radiomics features were extracted from pre-treatment chest enhanced CT arterial phase images of each patient,and clinical baseline data were collected for each patient.All patients were randomized stratified sampling at a 7:3 ratio into a training cohort(n=206)and a validation cohort(n=88).In the training cohort,the radiomics features related to 1-,2-,and 3-year RFS and OS of ESCC patients after surgery were screened by m RMR and LASSO regression algorithm to construct the radiomics model.Stepwise regression analysis based on Akaike information criterion(AIC)was used to identify the variables of multivariate Cox proportional hazards regression,which was used to construct a radiomics nomogram combining radiomics features and clinical features.The predictive performance and calibration of different models were evaluated using the Harrell’s concordance index(C-index)and calibration curves.The patients were divided into high recurrence and death risk groups and low recurrence and death risk groups according to radiomics nomogram,and Kaplan-Meier survival analysis was performed.The difference between survival curves was compared using the Log-rank test.The validation cohort was used to validate the performance of the models.Results: After LASSO-Cox regression,15 radiomics features significantly associated with RFS,and 15 radiomics features significantly associated with OS were selected to construct radiomics models.The C-index of the radiomics model for predicting RFS was 0.710(95%CI: 0.667-0.753)and 0.707(95%CI: 0.644-0.770)in the training and validation cohorts,respectively.The C-index of the radiomics model for predicting OS was 0.807(95%CI: 0.744-0.870)and 0.765(95%CI: 0.681-0.849)in the training and validation cohorts,respectively.Finally,the radiomics nomogram predicting postoperative RFS in ESCC was combined with 15 radiomics features and5 clinical features(lymph node ratio,pathological T stage,pathological N stage,LVI and tumor length).The C-index of this model was 0.758(95%CI: 0.717-0.799)in the training cohort and 0.722(95%CI: 0.653-0.791)in the validation cohort.The radiomics nomogram predicting postoperative OS in ESCC was combined with 15 radiomics features and 3 clinical features(recurrence and metastasis pattern,pathological N stage,perineural invasion).The C-index of this model was 0.884(95%CI: 0.841-0.927)in the training cohort and 0.809(95%CI: 0.739-0.880)in the validation cohort.The RFS and OS predicted by the radiomics nomograms showed good agreement with the actual RFS and OS.Kaplan-Meier survival analysis showed that the RFS and OS of the low-risk groups were higher than those of the high-risk groups(P<0.001).Conclusions: CT radiomics models were valuable in predicting postoperative RFS and OS in patients with ESCC,and the radiomics nomograms combined with CT radiomics and clinical features have higher predictive performance.