Machine learning detects lung cancer in blood samples with high accuracy

September 2021 Health Innovation Tobias Rawson

Published recently in Nature Communications, a large cohort study was able to detect both early stage (I/II) and later stage (III/IV) lung cancer at rates of 91% and 96%, respectively, using artificial intelligence to detect ctDNA in patient serum samples. Investigated by researchers at John Hopkins University, these results represent a potential breakthrough for lung cancer screening, given that less than 6% of patients in the United States undergo screening for this type of cancer, with even fewer doing so worldwide.

“It is clear that there is an urgent, unmet clinical need for development of alternative, non-invasive approaches to improve cancer screening for high-risk individuals and, ultimately, the general population,” says Dimitrios Mathios, MD, lead author of the study. “We believe that a blood test, or ‘liquid biopsy’, for lung cancer could be a good way to enhance screening efforts, because it would be easy to do, broadly accessible and cost-effective.”

80% specificity to detect 94% of all lung cancer subtypes

The study first investigated this technology in 365 individuals participating in the Danish LUCAS trial. These patients were considered as high risk for developing lung cancer and had smoking symptoms, such as persistent cough and shortness of breath. ctDNA analysis revealed that patients who had confirmed lung cancer had a wide variation in detectable DNA fragments in their blood samples, whilst patients who did not have lung cancer had relatively genetically consistent DNA fragments. In light of these results, this approach was subsequently validated using 385 non-cancer patients from the Danish Endoscopy III study, and 46 patients with lung cancer in the Dutch COCOS trial. Combined with clinical risk factors, CEA levels and CT imaging, this combined ctDNA analysis approach was able to detect lung cancer across all stages and subtypes at a rate of 94% at 80% specificity.

“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,” explains co-author Rob Scharpf, PhD.

 Moving forward, the DELPHI101 trial will evaluate this AI-enhanced ctDNA approach in 1,700 patients in the United States, comprised of a mixed cohort of participants with and without lung cancer, as well as patients with other cancers. Furthermore, the study group hope to expand this technology to other cancers in the future.


Mathios D, Johansen JS, Cristiano S et al., Detectino and characterization of lung cancer using cell-free DNA fragmentomes. Nat Comms. 2021; 12: 5060.