Johns Hopkins’ Artificial Intelligence-Based Blood Test Detects Liver Cancer

November 23, 2022 - Researchers from Johns Hopkins Kimmel Cancer Center have developed an artificial intelligence (AI)-based blood test that can detect liver cancer, based on a new study of 724 people.

The blood testing technology, known as DNA evaluation of fragments for early interception (DELFI), allowed researchers to successfully classify lung cancer in a 2021 study, according to the press release.

DELFI works by detecting changes in fragmentation among cancer cell DNA that is shed into the bloodstream. This type of DNA is known as cell-free DNA (cfDNA). By measuring the amount and size of cfDNA present in different parts of the genome found in a blood sample, researchers can gain insights into the way DNA is packaged inside a cell’s nucleus. These insights can help distinguish healthy cells from cancer cells, as cancer cells release DNA fragments into the bloodstream when they die. By analyzing these cfDNA fragments for abnormal patterns using AI and machine learning (ML), DELFI can identify the presence of cancer.

Following DELFI’s success at detecting lung cancer, the research team sought to apply the technology to liver cancer.

“Increased early detection of liver cancer could save lives, but currently available screening tests are underutilized and miss many cancers,” said Victor Velculescu, MD, PhD, professor of oncology and co-director of the Cancer Genetics and Epigenetics Program at the Johns Hopkins Kimmel Cancer Center, in the press release.

To evaluate DELFI’s utility in liver cancer screening, the researchers applied the technology to 724 blood plasma samples from the US, the European Union (EU), and Hong Kong. They tasked it with detecting hepatocellular cancer (HCC), the most common type of primary liver cancer, according to the Mayo Clinic.

Of the 724 samples, 501 were collected in the US and EU. These included samples from 75 people with HCC, which were used to train the ML model. To validate the model, an additional 223 plasma samples were analyzed from individuals in Hong Kong. These included samples from 90 people with HCC, 66 with hepatitis B virus (HBV), 35 with HBV-related liver cirrhosis, and 32 with no underlying risk factors.

Using these samples, the researchers were able to analyze patterns of cfDNA fragmentation and develop a DELFI score. These scores correlate with cancer presence, with higher scores indicating a higher likelihood of cancer.

Overall, scores were low for cancer-free individuals with viral hepatitis or cirrhosis, with a median DELFI score of 0.078 and 0.080, respectively. However, the scores for the 75 HCC patients in the US and EU samples were, on average, five to 10 times higher, and high scores were observed across all cancer stages, including early-stage disease.

DELFI achieved a sensitivity of 88 percent and a specificity of 98 percent among patients with an average risk of HCC. It showed 85 percent sensitivity and 80 percent specificity among patients at high risk of the condition.

The researchers concluded that DELFI could improve early detection but must be validated further in larger studies before it can be used in a clinical setting.

“Currently, less than 20 percent of the high-risk population get screened for liver cancer due to accessibility and suboptimal test performance. This new blood test can double the number of liver cancer cases detected, compared to the standard blood test available, and increase early cancer detection,” said Amy Kim, MD, assistant professor of medicine at the Johns Hopkins University School of Medicine and co-senior author on the study, in the press release.