Predicting the Grade of Esophageal Varices in Cirrhosis Based on Spectral Computed Tomography Radiomics

Abstract To investigate whether the spectral computed tomography (CT) radiomics may predict the grade of esophageal varices (EV). We retrospectively collected 301 cases with cirrhosis and EV and randomly divided them into training and test sets. Patients' clinical data, conventional enhanced CT characteristics, spectral CT 60 keV single-energy images, and iodine-based images were retrieved from the Picture Archiving and Communication Systems. Through comprehensive statistical analyses, including univariate analysis, correlation assessment, Least Absolute Shrinkage and Selection Operator regression, and multivariate regression analyses, the factors most related to EV grade were selected to construct the models. Predictive performance was evaluated by the receiver operating characteristic curve. A calibration curve was used to show the degree of fit between the nomogram and the actual results. The clinical utility of the model was evaluated using decision curve analysis (DCA). The diameters of the left gastric vein and EV were independent predictors of EV grade. Five out of 1,896 CT radiomics characteristics were correlated with EV grade. Six models were constructed. The model integrated with conventional enhanced CT and spectral CT radiomics characteristics performed best. The area under the curve in the training and test sets were 0.812 and 0.821, respectively. The calibration curve showed that this model had the highest agreement between observation and prognosis. The DCA found that this model provided the most clinical net benefit. The comprehensive model based on enhanced CT and spectral CT radiomics performed best in predicting the grade of EV and may be used as a reference for clinical decision-making.