Abstract
BACKGROUND: Extrahepatic portal vein obstruction (EHPVO) is a leading cause of pediatric portal hypertension. While invasive portography remains the diagnostic gold standard, its risks highlight the need for non-invasive alternatives. This study aims to integrate ultrasound imaging features and serological markers to establish a machine learning model for noninvasive, simplified preoperative assessment of the portal system in pediatric patients with EHPVO. The model will serve as a reference for selecting optimal surgical strategies. METHODS: A total of 103 pediatric EHPVO patients who underwent surgery were enrolled, including 81 Rex shunt and 22 Warren shunt cases. In the training set, the least absolute shrinkage and selection operator (LASSO) algorithm identified potential predictors. Five machine learning algorithms were employed for modeling. Model performance was evaluated through internal validation and external validation. RESULTS: Baseline characteristics showed no significant differences between training and validation sets. LASSO-selected features were used to construct five prediction models. The extreme gradient boosting (XGBoost) model outperformed the others. It achieved an area under the receiver operating characteristic curve (AUC) of 0.90 [95% confidence interval (CI): 0.79-0.99] on the training set and 0.75 (95% CI: 0.54-0.97) on the validation set. An online platform (https://rexshunt.shinyapps.io/rexorwarren/) was subsequently developed based on this optimal model. CONCLUSIONS: This study established a predictive model combining serological markers and ultrasound parameters to preoperatively assess portal venous anatomy in pediatric EHPVO. The online tool provides a noninvasive, user-friendly solution to guide surgical strategy selection for children with EHPVO.