Computed tomography-based radiomics analysis to predict lymphovascular invasion in esophageal squamous cell carcinoma.

Abstract

The present study explored the value of preoperative CT radiomics in predicting lymphovascular invasion (LVI) in esophageal squamous cell carcinoma (ESCC). A retrospective analysis of 294 pathologically confirmed ESCC patients undergoing surgical resection and their preoperative chest-enhanced CT arterial images were used to delineate the target area of the lesion. All patients were randomly divided into a training cohort and a validation cohort at a ratio of 7:3. Radiomics features were extracted from single-slice, three-slice, and full-volume regions of interest (ROIs). The least absolute shrinkage and selection operator (LASSO) regression method was applied to select valuable radiomics features. Radiomics models were constructed using logistic regression method and were validated using leave group out cross-validation (LGOCV) method. The performance of the three models was evaluated using the receiver characteristic curve (ROC) and decision curve analysis (DCA). A total of 1218 radiomics features were separately extracted from single-slice ROIs, three-slice ROIs, and full-volume ROIs, and 16, 13 and 18 features, respectively, were retained after optimization and screening to construct a radiomics prediction model. The results showed that the AUC of the full-volume model was higher than that of the single-slice and three-slice models. According to LGOCV, the full-volume model showed the highest mean AUC for the training cohort and the validation cohort. The full-volume radiomics model has the best predictive performance and thus can be used as an auxiliary method for clinical treatment decision making. LVI is considered to be an important initial step for tumor dissemination. CT radiomics features correlate with LVI in ESCC and can be used as potential biomarkers for predicting LVI in ESCC.