AI-Based System Can Detect More than 90% of Lung Cancers from a Blood Test

AI-Based System Can Detect More than 90% of Lung Cancers from a Blood Test

A new testing approach that incorporates an artificial intelligence system has been shown to be highly effective at detecting different types of lung cancer from a blood test.

The AI system was trained using blood testing technology developed at Johns Hopkins Kimmel Cancer Center, using samples obtained from 800 cancer patients and healthy individuals. The AI system is used to detect patterns in the fragmentation of DNA that is shed from cancer cells. The system, named DELFI (DNA evaluation of fragments for early interception), identifies patterns in the fragmentation of DNA shed from cancer cells that circulate in the bloodstream, known as cell free DNA (cfDNA).

Worldwide, lung cancer claims around 2 million lives each year. It is the leading cause of cancer-related death and the number of individuals diagnoses with lung cancer is increasing. Lung cancer has a poor survival rate, with fewer than 20% of patients surviving more than 5 years after being diagnosed. One of the main reasons for this is not that lung cancer cannot be treated, but that patients are often only diagnosed when the cancer is in the late stages when treatment is less effective. If a diagnosis is made earlier in the progression of the disease, tens of thousands of lives could be saved.

Lung cancer can be diagnosed using low-dose computed tomography (LDCT), and while patients at risk of developing lung cancer are screened, the number of individuals at risk who undergo LDCT screening is only around 6% in the United States. There are various reasons for the low uptake, such as the cost of investigation of false positives, the risk from exposure to radiation, or the risk from invasive procedures. There is a clear need for a less invasive and less risky approach to screening for lung cancer and a blood test – a liquid biopsy – would be an ideal option for individuals at risk of developing lung cancer and others in the general population.

DELFI could be the answer. This method indirectly measures how DNA is packaged inside cells by studying the amount and size of the cfDNA in the blood. The DNA of cancer cells is packaged in a much more disorganized way than the DNA of normal cells. When cancer cells die, their DNA is released into the blood stream and it is possible, using DELFI, to detect the cfDNA and differentiate it from the cfDNA released by normal cells that have died.

The artificial intelligence element is able to analyze millions of cfDNA fragments and identify abnormal patterns, which include the size and amount of DNA from different parts of the genome. The view of the cfDNA is known as the fragmentome.

“To increase the sensitivity of detection of early-stage cancers we have developed a genome-wide approach for analysis of cfDNA fragmentation profiles called DELFI,” explained the authors. “This approach provides a view of cfDNA ‘fragmentomes,’ permitting evaluation in any individual of the size distribution and frequency of millions of naturally occurring cfDNA fragments across the genome.” This approach “has the potential to identify a large number of tumor-derived changes in the circulation.” Further, this approach would be cost-effective to perform as part of a screening program.

The new approach was tested using blood samples taken from 365 individuals who had participated in a 7-year study in Denmark named LUCAS. Most of the patients in the study were at a high risk of developing lung cancer, and generally had smoking-related symptoms such as difficulty breathing. The researchers found that patients who developed lung cancer had much greater variation in their fragmentome than patients who did not develop cancer, whose fragmentome profiles were very consistent. The differences in the fragmentome profiles in cancer patients were seen in multiple regions, genome-wide, across different cancer stages.

The technique was validated using data from a different sample of 385 individuals without cancer and 46 individuals with cancer. The researchers were able to identify more than 90% of patients with lung cancer, 91% of patients with invasive stage I/II cancers, and 96% of patients with stage III/IV cancers.

“DNA fragmentation patterns provide a remarkable fingerprint for early detection of cancer that we believe could be the basis of a widely available liquid biopsy test for patients with lung cancer,” said study co-author Rob Scharpf, PhD, associate professor of oncology at the Johns Hopkins Kimmel Cancer Center.

You can read more about the study in the paper – Detection and characterization of lung cancer using cell-free DNA fragmentomes – which was recently published in Nature Communications. DOI: 10.1038/s41467-021-24994-w