AI model uses blood data to detect early liver metastasis in pancreatic cancer: Severance study

A research team at Severance Hospital has developed an artificial intelligence (AI) model that predicts early liver metastasis in pancreatic cancer patients using routine blood test data, offering a potential alternative to imaging-based diagnosis.

The team, led by Professor Lee Hee-seung of the Division of Gastroenterology and first author Ko Yeo-kyung, said Monday that the model demonstrated clinical potential in identifying patients at risk of early metastasis. The findings were publishedin the latest issue of BMC Cancer.

Professor Lee Hee-seung (left) of the Division of Gastroenterology and first author Ko Yeo-kyung at Severance Hospital

Liver metastasis is a key factor in determining treatment strategy and prognosis in pancreatic cancer, but small metastatic lesions are often difficult to detect using conventional imaging methods such as CT and MRI.

To address this limitation, the researchers developed an AI model, named Liver Metastasis in Pancreatic Cancer (LiMPC), which analyzes routine blood test data collected at diagnosis to assess metastasis risk without requiring additional tests or specialized equipment.

The model was trained on blood test data from 2,657 pancreatic cancer patients at Severance Hospital and externally validated in 272 patients across five hospitals, including Gangnam Severance Hospital and Yongin Severance Hospital.

In external validation, the model achieved a sensitivity of 0.81, correctly identifying about 81 percent of patients with liver metastasis as high-risk. It also showed a negative predictive value of 0.87, indicating that most patients classified as low-risk did not have metastasis.

The researchers said the model could serve as a complementary tool to conventional imaging, particularly in cases where small metastases are difficult to detect.

Because the model relies on standard blood tests, it may be applicable in a wide range of clinical settings, including resource-limited environments. The team has also developed an online calculator to support clinical use and plans to release it in the near future.

“Pancreatic cancer is often difficult to assess for metastasis using imaging alone,” Lee said. “Our findings suggest that routine blood tests could help predict liver metastasis and complement standard diagnostic methods.”

Ko added that the model’s accessibility could support broader clinical adoption, noting that further research will focus on improving its performance in real-world settings.