Artificial Intelligence Helps in Early Detection of Liver Cancer Risk, Study Shows

A machine learning-based tool that analyzes electronic health records, test results, and demographic data could help physicians identify individuals at high risk for hepatocellular carcinoma (HCC)—the most common form of liver cancer. This was reported by Pharmaphorum.

This conclusion was reached by a study published in the scientific journal Cancer Discovery, which utilized data from the UK Biobank involving over 500,000 participants. The database included 538 cases of HCC, with more than two-thirds (69%) occurring in patients without classic risk factors such as cirrhosis, viral hepatitis, or other chronic liver diseases.

Researchers trained their models using 80% of the data, with initial validation performed on the remaining 20%. Further verification was conducted using the US-based All of Us research program, which includes approximately 400,000 individuals and 445 cases of HCC.

The aim of the study is to improve the current approach to identifying individuals at risk for liver cancer, which currently focuses primarily on a limited group of high-risk patients through imaging and blood tests—an approach that may miss a significant number of at-risk patients.

“Screening is typically recommended for patients with proven liver cirrhosis or severe liver disease, as many cases of HCC occur specifically in them. However, there are many people with undiagnosed cirrhosis or other risk factors who would also benefit,” states study co-leader Caroline Schneider from RWTH Aachen University in Germany.

One version of the algorithm—PRE-Screen-HCC (Model C)—analyzes a wide range of data, including demographic characteristics, lifestyle, medical history, and blood markers, and succeeds in stratifying individual risk of developing HCC “with high accuracy,” according to the authors.

Interestingly, the addition of genomic and metabolomic data—which are more difficult to collect at scale—did not significantly improve the model’s performance.

“This shows that we can predict HCC risk using simple and widely available data, without the need for complex and expensive genetic testing,” Schneider emphasizes, adding that this increases the potential for widespread application, especially in resource-limited countries.

Hepatocellular carcinoma is the fifth most common malignant tumor and the third leading cause of cancer death worldwide. Its incidence is rising, primarily due to increasing cases of liver disease, making it a serious public health issue.

Although the model was trained primarily on data from white participants in the UK Biobank, it maintained its effectiveness when analyzing the more diverse population in the US All of Us registry, the researchers noted.

The study’s first author, Jan Klussmann from the Technical University of Dresden, emphasizes the importance of such technologies:
“With so many factors influencing risk, there is an urgent need for effective tools to help physicians identify high-risk patients. Machine learning tools that can work simultaneously with different types of clinical data can be particularly useful for this serious challenge.”