Lung cancer learned to detect using blood tests and artificial intelligence

Lung cancer learned to detect using blood tests and artificial intelligence
Lung cancer learned to detect using blood tests and artificial intelligence

The new non-invasive method has detected more than 90 percent of lung cancer cases at various stages among nearly 800 people.

Lung cancer

Lung cancer is considered the deadliest in the world, with less than 20 percent chances of surviving for at least five years after diagnosis. This often happens because the disease is detected at a late stage, when treatment is less effective, and the tumor continues to grow. Although large-scale studies have shown that screening for lung cancer with reduced-dose chest computed tomography reduces mortality, the technique remains unpopular due to concerns about radiation and false-positive imaging results.

Therefore, medicine needs non-invasive approaches to help diagnose lung cancer. Studies of proteins, autoantibodies, gene expression profiles, and microRNAs in blood or airway epithelium have shown promising biomarker candidates for early detection of this cancer, but some may be confusing due to patient age, inflammation from smoking or others. accompanying conditions.

At the same time, it is known that mutations or methylation of circulating tumor DNA (fragmented tumor DNA in the bloodstream) may be present in patients with early lung cancer. Based on this, scientists from the Johns Hopkins University School of Medicine (Baltimore, USA) developed a genome-wide approach to the analysis of extracellular DNA fragmentation profiles called DELFI: it allowed to review and estimate the size distribution and frequency of millions of naturally occurring cfDNA fragments. The results of the work were published in the journal Nature Communications.

According to the authors of the study, healthy cells “pack” DNA into the nucleus as if they are putting things in a suitcase correctly: different parts of the genome are neatly placed in their compartments. The nuclei of tumor cells, on the other hand, are scattered randomly, and when they die, they randomly release DNA into the bloodstream. DELFI helps determine the presence of cancer using machine learning - a class of artificial intelligence techniques - that examines millions of extracellular DNA fragments for abnormal patterns, including the size and number of macromolecules in different regions of the genome. The approach only requires low-coverage genome sequencing, which allows the technology to be cost-effective.

Scientists examined blood samples from 365 people who were examined for seven months at the Bispebjerg hospital in Copenhagen (Denmark). Many had a high risk of developing lung cancer (age 50-80 years, smokers with more than 20 years of experience). A total of 323 participants (90%) were included in the cohort with pulmonary, non-pulmonary, or constitutional symptoms, the majority having common signs of illness associated with smoking, such as coughing or shortness of breath. All underwent chest CT or positron emission tomography combined with computed tomography (PET / CT) of the whole body with 18F-FDG.

The scientists then isolated two to four milliliters of blood plasma from each patient and examined the extracted cfDNA using DELFI, sequencing its genome. A few days later, 129 out of 365 people were diagnosed with lung cancer (on average, after 9.5 days), and 87 had histologically confirmed benign neoplasms.

As it turned out, people who were diagnosed with cancer had widespread fragment variations, in contrast to healthy study participants (149).Then the technology was rechecked on another group - 46 patients with cancer and 385 people without it. As a result, it was possible with 94% accuracy to identify a malignant neoplasm, including at the early and late stages of development, as well as with different subtypes.

Popular by topic