Can machine learning with cfDNA help detect liver cancer at different stages?

Machine learning (ML) combined with cell-free DNA (cfDNA) analysis is a promising approach for detecting liver cancer at different stages. cfDNA are fragments of DNA released into the blood by cells, including cancer cells. ML algorithms can analyze patterns in cfDNA, such as fragmentation or methylation, to distinguish cancer from non-cancer. Studies show that this method can detect liver cancer with high accuracy, especially at later stages, but sensitivity is lower for early-stage disease.

What the research says

A large systematic review and meta-analysis of 109 studies found that ML with cfDNA can detect liver cancer (and other cancers) with high specificity (94%-99%) across all stages ¹. Sensitivity varied by stage: for stage I liver cancer, sensitivity ranged from 44% to 91%; for stage II, 71% to 98%; and for stage III, 83% to 99% ¹. This means the test is better at catching later-stage cancers but may miss some early ones.

Specific studies on liver cancer support these findings. One study using whole-genome cfDNA fragmentome analysis (looking at DNA fragmentation patterns) with a machine learning model detected hepatocellular carcinoma (the most common type of liver cancer) with 88% sensitivity in an average-risk population at 98% specificity, and 85% sensitivity in high-risk individuals at 80% specificity ⁵. Another study used cfDNA methylation patterns (chemical tags on DNA) and reported a sensitivity of 69.1% across multiple cancers (including liver) at 98.9% specificity, with tissue origin accuracy of 83.2% ⁷.

Additionally, a 2024 study using a method called ARTEMIS analyzed repeat elements in cfDNA and found that machine learning could detect early-stage lung or liver cancer in validated cohorts ⁶. This suggests that different cfDNA features (fragmentation, methylation, repeats) can all be leveraged for detection.

While these results are encouraging, most studies are retrospective and need further validation in real-world screening settings. The meta-analysis noted that neural networks and random forest models showed the highest sensitivity ¹, but performance may vary depending on the specific ML approach and population.

What to ask your doctor

• What are the current recommended screening methods for liver cancer, and how does cfDNA testing compare?
• Is cfDNA-based testing available for liver cancer detection, and if so, is it approved for use in my situation?
• How should I interpret a positive or negative cfDNA test result for liver cancer?
• Are there any ongoing clinical trials for cfDNA-based liver cancer screening that I might be eligible for?
• Given my risk factors (e.g., cirrhosis, hepatitis), would cfDNA testing be appropriate for me?