Researchers at the Johns Hopkins Kimmel Cancer Center have validated an artificial intelligence–enabled blood test that can identify liver cancer in two populations with very different genetic, geographic, and disease-risk profiles. The study also clarified some of the biological mechanisms that allow the test to recognize signs of cancer in circulating blood DNA.
Published on July 31 in Cell Press Blue, the research expands on the team’s earlier work with the DELFI platform, short for DNA Evaluation of Fragments for Early Interception. This liquid-biopsy technology examines millions of cell-free DNA fragments found in the bloodstream. Rather than focusing only on individual mutations, DELFI evaluates genome-wide patterns in the size, distribution, and characteristics of DNA fragments.
The latest findings confirm that a previously developed liver cancer classifier can perform effectively in independent groups of people at elevated risk of the disease. The research follows work reported in March 2026, when the same research team showed that fragmentome-based analysis could also help detect liver fibrosis and cirrhosis—two chronic liver conditions that can increase the likelihood of liver cancer.
Why Earlier Detection of Liver Cancer Matters
Liver cancer remains one of the most common causes of cancer-related mortality worldwide. Its growing prevalence is linked in part to metabolic liver disease, obesity, diabetes, viral hepatitis, alcohol-related liver damage, and other risk factors.
Finding liver cancer in its earlier stages can significantly expand treatment options. However, standard screening methods—mainly ultrasound imaging combined with measurement of the blood protein alpha-fetoprotein (AFP)—may fail to identify some early-stage tumors.
The Johns Hopkins team aimed to determine whether a blood-based fragmentome test could detect liver cancer reliably even when the underlying causes of disease differ between populations.
Testing the DELFI Platform in Guatemala and Romania
The investigators examined blood samples from 377 participants in Guatemala and Romania, including individuals with and without hepatocellular carcinoma, the most common type of primary liver cancer.
The two study groups represented substantially different pathways to liver disease and cancer. In Romania, liver cancer risk was more commonly associated with viral hepatitis and alcohol use. In Guatemala, many participants had metabolic liver disease connected to obesity and diabetes. Some individuals in the Guatemalan population had also been exposed to aflatoxin, a naturally occurring toxin that is known to contribute to liver cancer development.
Despite these important differences, the AI-supported blood test detected liver cancer consistently in both groups. When researchers combined the fragmentome analysis with AFP results and basic clinical information, including age and sex, the approach showed greater sensitivity for identifying both early- and late-stage disease than blood testing alone.
Looking Beyond Tumor DNA
A central finding of the study is that the test does not rely exclusively on DNA released by tumor cells. The team used a new DNA-source mapping method called MethID to investigate where the circulating DNA fragments originated.
Their analysis indicated that the DELFI platform captures biological signals from several sources, including cancer cells, liver cells, blood vessels, and immune cells reacting to the tumor environment. This broader set of signals may explain why genome-wide fragmentome analysis can provide useful information about both the presence of cancer and the biological changes occurring as the disease develops.
The researchers also found molecular differences between the two populations. For instance, participants from Guatemala showed a distinctive genome-wide mutation pattern associated with aflatoxin exposure. Even so, the overall liver cancer classifier remained effective across the different causes and risk profiles represented in the study.
Toward More Adaptable Liquid Biopsy Tests
The results suggest that genome-wide analysis of cell-free DNA may identify both shared features of liver cancer and region-specific molecular signatures. This could make the technology suitable for use across diverse populations rather than limiting it to patients with a single type of liver disease or environmental risk factor.
The work also supports a larger goal for fragmentome technology: creating noninvasive blood tests that can screen for several diseases using a common analytical platform. The Johns Hopkins researchers and DELFI Diagnostics have already reported clinical validation of FirstLook Lung, a blood test designed to support lung cancer screening that is available in selected health systems.
Future research will focus on prospective clinical studies and on improving multi-layered liquid-biopsy models. These models may combine fragmentome data with protein biomarkers and patient risk factors to make liver cancer screening more accurate, accessible, and effective at detecting disease earlier.