Macrotrabecular-massive subtype in hepatocellular carcinoma based on contrast-enhanced CT: deep learning outperforms
Lulu Jia1, Zeyan Li2, Gang Huang3
1The First Clinical Medical College of Lanzhou University, Lanzhou, China.
Insights Into Imaging
|August 28, 2025
Summary
A novel deep learning model accurately predicts the macrotrabecular-massive (MTM) subtype of hepatocellular carcinoma (HCC) using CT scans. This advanced tool surpasses traditional machine learning, aiding in clinical decisions and personalized therapy for HCC patients.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Medicine
- Oncology
Background:
- Hepatocellular carcinoma (HCC) is a primary liver cancer with distinct subtypes impacting prognosis and treatment.
- Accurate non-invasive subtyping of HCC, such as the macrotrabecular-massive (MTM) subtype, is crucial for effective clinical management.
- Current diagnostic methods may require invasive procedures, highlighting the need for advanced imaging-based predictive tools.
Purpose of the Study:
- To develop and validate a deep learning (DL) model for predicting the MTM subtype of HCC using contrast-enhanced computed tomography (CT) images.
- To compare the diagnostic performance of the developed DL model against conventional machine learning (ML) models and baseline DL approaches.
- To explore the potential of integrating clinical parameters with DL models for enhanced MTM subtype prediction.
Main Methods:
- Retrospective collection of contrast-enhanced CT data from 368 HCC patients across two medical centers.
- Development of a novel DL network, ResNet-ViT Contrastive Learning (RVCL), based on ResNet-50 architecture.
- Comparative analysis of RVCL against baseline DL and ML models using Area Under the Receiver Operating Characteristic Curve (AUC) for performance evaluation.
Main Results:
- The RVCL model achieved a superior AUC of 0.93 for MTM subtype prediction on the external test set, significantly outperforming baseline DL (AUCs 0.46-0.72) and ML models (AUCs 0.49-0.60).
- Integration of Alpha-Fetoprotein (AFP) with the RVCL model did not yield a statistically significant improvement in diagnostic performance.
- The RVCL model demonstrates robust accuracy in predicting the MTM subtype non-invasively.
Conclusions:
- Contrast-enhanced CT combined with the RVCL deep learning model provides an accurate method for predicting the MTM subtype of HCC.
- The RVCL model represents a significant advancement over traditional ML techniques for HCC subtyping.
- This AI-driven approach offers a valuable tool for clinical decision-making and can guide personalized therapeutic strategies in HCC management.
