Establishment of a non-invasive prediction model for the risk of oesophageal variceal bleeding using radiomics based on CT

Abstract

To establish a non-invasive prediction model for the risk of oesophageal variceal bleeding (OVB) using radiomics based on computed tomography (CT). The study included 317 patients, 69 of whom were OVB-positive and 248 were OVB-negative. The OVB was caused by cirrhosis associated with hepatitis B. All patients underwent both oesophagogastroduodenoscopy (OGD) and triple-phase contrast-enhanced CT with spectral imaging mode within 14 days before OGD. The patients were divided chronologically into training (n=222) and validation (n=95) cohorts at a ratio of 7:3. The clinical and CT features were collected from a picture archiving and communication system, and radiomics features were extracted from the portal venous phase CT. Spearman's correlation, least absolute shrinkage, and selection operator regression analyses were used to select the most correlated features. Models were built using the selected features. The predictive performance of the models was evaluated using the area under the receiver operating characteristic curve (AUC). One clinical feature, five CT features, and three radiomics features were selected, and three non-invasive models were built. Integration of the radiomics, CT, and clinical features model showed a better performance in predicting the risk of OVB, with an AUC of 0.89 (95% confidence interval [CI], 0.84-0.94) in the training dataset and 0.78 (95% CI, 0.68-0.87) in the validation dataset. The combination of radiomics, CT, and clinical features may have added value in the non-invasive prediction of OVB, enabling early prevention and treatment.