A NEW study has suggested that deep-learning (DL) radiopathomics models can help predict the presence of vessels encapsulating tumour clusters (VETC) in hepatocellular carcinoma (HCC), a form of liver cancer linked to poor patient outcomes. The research, led by Dr Yixing Yu of the First Affiliated Hospital of Soochow University, China, indicates that these models could also assist in assessing early recurrence risk and progression-free survival.
HCC is the third leading cause of cancer-related deaths globally, and despite advancements in diagnosis and treatment, outcomes for advanced cases remain bleak. One of the challenges in managing HCC is accurately characterising the biological features of tumours. Deep-learning models offer a potential breakthrough by integrating radiomics (quantitative data from medical imaging) and pathomics, which converts digital pathology images into measurable features.
VETC is a distinct vascular pattern in HCC, where tumour clusters are surrounded by a network of sine-wave-like blood vessels. This pattern has been associated with a novel metastasis mechanism and poorer prognosis, but it also predicts a potential benefit from sorafenib, a targeted therapy for liver cancer. The study developed and validated deep-learning radiomics and pathomics models to predict VETC presence and aid in prognosis.
The research team analysed data from 578 HCC patients, divided into training, internal test, and external test groups. Their findings included:
- The deep-learning radiomics model achieved an area under the ROC curve (AUC) of 0.77 in the external test set, while the pathomics model performed slightly better with an AUC of 0.79.
- Patients with the VETC pattern exhibited significantly higher radiomics and pathomics scores across all datasets (p < 0.001).
- A radiopathomics nomogram model successfully differentiated high- and low-risk patients for early recurrence and progression-free survival (p < 0.05).
Visualisation techniques, such as gradient-weighted class activation mapping (Grad-CAM), further demonstrated how deep-learning models identified key tumour characteristics. Heatmaps generated from various deep-learning architectures, including ResNet50 and Vision Transformer, highlighted regions associated with encapsulating blood vessels.
While the findings are promising, the authors stress the need for further research. “In the future, prospective clinical trials are needed to validate the utility of the radiopathomics nomogram models in diverse patient populations,” they concluded.
This study underscores the growing role of artificial intelligence in oncology, offering hope for improved HCC prognosis and treatment strategies.
Reference
Yu Y et al. Deep learning radiopathomics models based on contrast-enhanced MRI and pathologic imaging for predicting vessels encapsulating tumor clusters and prognosis in hepatocellular carcinoma. Radiol Imaging Cancer. 2025;7(2):e240213.
