Document resource
Introduction Budd Chiari syndrome (BCS) and non cirrhotic portal vein thrombosis (PVT) are hepatic vascular disorders typically occurring in a prothrombotic context. Early initiation of anticoagulation is recommended to promote vascular recanalization and prevent thrombus extension or recurrence. Imaging-confirmed vascular recanalization is a major prognostic outcome; however, its prediction using conventional statistical approaches remains limited.The aim of this study was to evaluate the performance of an artificial intelligence (AI)-based model in predicting vascular recanalization in patients with BCS and non-cirrhotic PVT.Methods We conducted a retrospective study including patients diagnosed with BCS or non-cirrhotic PVT and managed in a tertiary hepatology center between 2015 and 2025.Vascular recanalization, defined as complete or partial resolution of thrombosis on follow-up imaging, was used as the reference outcome. All patients received long-term anticoagulation tailored to thrombotic risk, using vitamin K antagonists or direct oral anticoagulants, with targeted etiological treatment when appropriate.Baseline clinical, etiological and imaging variables available at diagnosis were used to develop a supervised machine-learning model based on a Random Forest algorithm. Model performance was assessed using receiver operating characteristic (ROC) curve analysis, including the area under the curve (AUC), sensitivity and specificity.Results A total of 96 patients were included (38 with BCS, 58 with non-cirrhotic PVT). Overall, vascular recanalization was observed in 28 patients (29%).In the BCS group, the mean age was 31 years, with a marked female predominance (81%). Recanalization occurred in 11 patients (29%), including complete recanalization in 36% and partial recanalization in 63%. Recanalization was associated with a shorter diagnostic delay and the initiation of specific etiological treatment.In the PVT group, the mean age was 45 years, with a similarly high female predominance (78%). Recanalization was observed in 17 patients (29%), including complete recanalization in 12% and partial recanalization in 29%. Stable thrombosis without extension was observed in 58% of cases.The Random Forest model demonstrated good discriminative performance, with an AUC of 0.81, sensitivity of 80% and specificity of 71% for predicting imaging-confirmed vascular recanalization. The most influential variables in the AI model were diagnostic delay, thrombotic etiology and the extent of vascular involvement on imaging.Conclusions An AI-based model showed promising performance in predicting imaging-confirmed vascular recanalization in patients with BCS and non-cirrhotic PVT. This approach may support early risk stratification and individualized management, pending validation in larger prospective cohorts.