Computed tomography-based deep learning and multi-instance learning for predicting microvascular invasion and
Yong-Yi Cen1,2, Hai-Yang Nong1,2, Xiao-Xiao Huang1,2
1Guangxi Clinical Medical Research Center for Hepatobiliary Diseases, Affiliated Hospital of Youjiang Medical University for Nationalities, Baise 533000, Guangxi Zhuang Autonomous Region, China.
World Journal of Gastroenterology
|September 11, 2025
Summary
A new deep learning model accurately predicts microvascular invasion (MVI) in hepatocellular carcinoma (HCC), outperforming existing methods. This MVI prediction tool aids in personalized treatment strategies for HCC patients.
Area of Science:
- Hepatocellular Carcinoma Research
- Medical Imaging Analysis
- Artificial Intelligence in Oncology
Background:
- Microvascular invasion (MVI) is a critical prognostic indicator in hepatocellular carcinoma (HCC).
- Accurate preoperative prediction of MVI in HCC remains a significant clinical challenge.
Purpose of the Study:
- To develop and validate a 2.5D deep learning-based multi-instance learning (MIL) model for MVI prediction in HCC.
- To compare the MIL model's performance against radiomics and clinical signatures.
- To assess the prognostic value of the MIL signature in surgical resection and TACE cohorts.
Main Methods:
- A retrospective cohort of 192 HCC patients was used, with a 7:3 split into training and validation sets.
- A 2.5D deep learning MIL framework was applied to computed tomography arterial phase images for MVI prediction.
- Performance was evaluated using ROC curves, DCA, and DeLong's test; prognostic value assessed via Kaplan-Meier curves.
Main Results:
- The MIL signature achieved superior AUC (0.877) in validation, significantly outperforming radiomics (0.727) and clinical (0.631) signatures.
- Decision curve analysis showed greater clinical net benefit for the MIL signature.
- The MIL signature effectively stratified patients by recurrence-free survival (surgical cohort) and progression-free survival (TACE cohort).
Conclusions:
- The MIL signature provides more accurate MVI prediction in HCC compared to radiomics and clinical signatures.
- This AI-driven approach offers precise prognostic stratification for HCC.
- The MIL signature supports personalized HCC treatment strategies through enhanced predictive capabilities.
Keywords:
Deep learningHepatocellular carcinomaMicrovascular invasionMulti-instance learningPrognosis
