Development of a machine learning-based predictive model and web application for portal vein thrombosis risk in patients with liver cirrhosis(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2026年第1期
- Page:
- 121-131
- Research Field:
- 医学人工智能
- Publishing date:
Info
- Title:
- Development of a machine learning-based predictive model and web application for portal vein thrombosis risk in patients with liver cirrhosis
- Author(s):
- WANG Ganhong1; XI Meijuan1; HUA Qiuwen1; XIA Kaijian2; 3; NI Xiaochen1; CHEN Jian3; 4
- 1. Department of Gastroenterology, Changshu Traditional Chinese Medicine Hospital (Changshu New District Hospital), Changshu 215500, China 2. Intelligent Medical Technology Research Center, Changshu No.1 Peoples Hospital (Changshu Hospital Affiliated to Soochow University), Changshu 215500, China 3. Changshu Key Laboratory of Medical Artificial Intelligence and Big Data, Changshu 215500, China 4. Department of Gastroenterology, Changshu No.1 Peoples Hospital (Changshu Hospital Affiliated to Soochow University), Changshu 215500, China
- Keywords:
- Keywords: liver cirrhosis portal vein thrombosis machine learning application program Streamlit
- PACS:
- R318；R575.2
- DOI:
- DOI:10.3969/j.issn.1005-202X.2026.01.016
- Abstract:
- Abstract: Objective Portal vein thrombosis (PVT) is a common complication of liver cirrhosis, exacerbating liver damage, portal hypertension, and associated complications. This study aims to establish a predictive model and web application for PVT complicating liver cirrhosis based on various machine learning (ML) methods. Methods A retrospective study was conducted on 662 patients with liver cirrhosis who were divided into either PVT group or non-PVT groups. Twenty-six potential predictive variables were included to construct and validate a LASSO regression model and 5 different ML models. Model performance was evaluated using metrics including the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy, with the optimal-performing model identified. Visual interpretation of the model was achieved using feature importance analysis and SHAP scatter plots. A web application based on the optimal model was ultimately developed using Python-Streamlit framework and externally validated on an independent test set from another hospital. Results In the validation set, the XGBoost model had the optimal performance, superior to the LASSO regression model and the other ML models, achieving an AUC of 0.93 (95% CI: 0.88-0.98). In the test set, the XGBoost model yielded an accuracy of 83.47%, sensitivity of 85.71%, specificity of 82.80%, and AUC of 0.90 (95% CI: 0.81-0.97). Feature importance analysis revealed that prothrombin activity, portal vein diameter, MELD score, D-dimer, platelet count, prothrombin time, and albumin played critical roles in predicting PVT complicating liver cirrhosis. SHAP scatter plots and force plots visualized the impact of key features on PVT prediction. Conclusion The XGBoost-based predictive model and its corresponding web application demonstrates strong predictive capability and excellent portability for assessing PVT risk in patients with liver cirrhosis, thereby providing valuable support for healthcare professionals to identify patients at high risk of PVT.
Last Update: 2026-01-27