An interpretable model combining brain imaging and clinical indicators for predicting overt hepatic encephalopathy.

Differentiating overt hepatic encephalopathy (OHE) from covert hepatic encephalopathy (CHE) remains challenging due to overlapping symptoms and the limitations of current grading tools. This study aimed to integrate quantitative susceptibility mapping (QSM) features with clinical biomarkers to construct a predictive model for OHE and deploy it via an interpretable web-based tool. Sixty-eight cirrhotic patients were divided into OHE (n = 31) and CHE (n = 37) groups. Over 40 variables were collected. After univariate logistic screening, LASSO regression identified six key predictors: MBP, LRN, RCA, RBC, Hgb, and Fib. Three models were built: a QSM-enhanced logistic model, a conventional logistic model, and a random forest model. Performance was assessed using AUC, PR-AUC, DCA, bootstrap, and cross-validation. A nomogram and SHAP analysis were used for interpretability. The QSM-based logistic model achieved the best performance (AUC = 0.83; PR-AUC = 0.83), exceeding the conventional logistic model (AUC = 0.71) and random forest model (AUC = 0.77). Internal validation confirmed stability (bootstrap AUC = 0.831; cross-validation AUC = 0.823 ± 0.039). SHAP analysis revealed variable importance and interaction effects. An interactive web tool was developed for individualized prediction. Combining QSM imaging markers with routine blood tests enables accurate and explainable prediction of OHE risk. The SHAP-based platform may support early detection and personalized decision-making in hepatic encephalopathy care.