Predicting microvascular invasion in patients with hepatocellular carcinoma using interpretable machine learning models

Read full text Summarise with VIRAT
Authors: Zhaobin He, Qisen Yan, Shengbiao Yang, Xiqiang Wang, Huijie Gao, Yongzhe Yu, Cheng Peng
Cited by: 0
References: 49

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Microvascular invasion (MVI) represents a critical pathological hallmark of hepatocellular carcinoma (HCC), strongly associated with early postoperative recurrence and poor outcomes. This study aimed to develop and validate interpretable machine learning (ML) models for the preoperative prediction of MVI in patients with HCC, utilizing routinely accessible clinical data, and to elucidate the key predictive features through Shapley additive explanations (SHAP). Clinicopathologic data were retrospectively collected from 1095 patients who underwent hepatectomy for HCC at Qilu Hospital of Shandong University from December 2020 to December 2024. Key patient demographics, laboratory tests, tumor characteristics, and pathological findings were extracted for analysis. After preprocessing and feature selection, the cohort was randomly divided into a training set (70%) and an internal testing set (30%). Model development and hyperparameter tuning were performed within the training set using tenfold cross-validation. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), as well as accuracy, precision, sensitivity, specificity, and F1 score. The SHAP method was applied to interpret the best-performing models at both global and local levels. Among all algorithms, the random forest, light gradient boosting machine, and extreme gradient boosting models demonstrated the highest predictive performance for MVI (testing cohort AUCs: 0.907, 0.893, and 0.880, respectively). Feature importance analysis with SHAP consistently identified alpha-fetoprotein as the most influential predictor of MVI, followed by des-gamma-carboxy prothrombin, fibrinogen, hepatitis B virus deoxyribonucleic acid, and gamma-glutamyl transferase. SHAP visualizations provided transparent, individualized explanations for predictions, supporting clinical interpretability. Interpretable ML models were successfully developed and internally validated using readily available clinical variables for the preoperative prediction of MVI in patients with HCC. The transparent identification of five key predictors facilitates personalized treatment planning. However, further external validation in multicenter cohorts is warranted to confirm clinical applicability.

Fields of Study & Subjects

Hepatocellular Carcinoma Treatment and Prognosis Radiomics and Machine Learning in Medical Imaging Artificial Intelligence in Healthcare

Publication details

Journal / source
European journal of medical research
Volume
31
Issue
1
Pages
686
Publication date
31 March 2026
Language
EN