Large language model and Gd-EOB-DTPA-enhanced MRI-based risk stratification system for postoperative hepatocellular
Can Yu1, Qi Zhang2, Jia-Xuan Ding1
1Department of Radiology, Harbin Medical University Cancer Hospital, Harbin, China.
European Radiology
|February 23, 2026
Summary
A new Fully Automated Stratification System (FASS) integrates serum biomarkers, radiomic features, and large language model (LLM) insights for hepatocellular carcinoma (HCC) risk prediction after surgery.
Area of Science:
- Oncology
- Artificial Intelligence
- Medical Imaging
Background:
- Hepatocellular carcinoma (HCC) is a major global health concern.
- Accurate prognostic prediction is crucial for guiding treatment decisions after surgical resection.
- Current risk stratification methods for HCC often lack integration of advanced imaging and semantic analysis.
Purpose of the Study:
- To develop and validate a Fully Automated Stratification System (FASS).
- To integrate serum biomarkers, automated radiomic features, and large language model (LLM)-derived semantic features.
- To predict prognosis in patients with solitary HCC post-hepatic resection.
Main Methods:
- Retrospective enrollment of 448 solitary HCC patients from three centers.
- Automated tumor segmentation using a modified MedNeXt-loss framework on Gd-EOB-DTPA-enhanced MRI.
- Comparison of five LLMs for feature extraction, with ChatGPT-4o selected for FASS integration.
- Prognostic model evaluation using concordance index, time-dependent ROC, and decision curve analyses.
- Exploration of biological relevance via RNA sequencing and pathway enrichment.
Main Results:
- Robust tumor segmentation achieved (Dice = 0.77) using the MedNeXt-loss framework.
- ChatGPT-4o demonstrated optimal predictive accuracy and completeness for semantic feature extraction.
- Alpha-fetoprotein (AFP), AST, and LLM-derived irregular margin identified as independent predictors of overall survival.
- Integrated FASS achieved high prognostic performance (C-index 0.78/0.76) and effective patient risk stratification (log-rank p < 0.05).
- Transcriptomic analysis indicated inflammatory and cytokine signaling activation in the high-risk group.
Conclusions:
- FASS provides fully automated, interpretable, and biologically informed prognostic assessment for solitary HCC.
- The system supports precision decision-making in hepatobiliary oncology.
- This automated platform enables reliable postoperative risk stratification, aiding early identification of high-risk patients and potentially improving outcomes.
