Dimitrios Mathios, MD, an assistant professor of neurosurgery at WashU Medicine, has devised a method that uses machine learning to detect brain tumors by analyzing DNA patterns associated with cancer in the blood. In a recent study, Mathios found that this machine-learning technique detected cancer in 73% of cases from a cohort drawn from patients in the U.S. and Korea. The results were validated in a second cohort of samples from brain cancer patients from Poland. In contrast, an earlier study found traditional liquid biopsies that tested blood for standard cancer biomarkers detected less than 10% of brain tumors.
Mathios' study was published in Cancer Discovery April 29.
Detecting brain tumors early, before they have grown and spread, has the potential to improve treatment outcomes. But early detection of brain cancer has proven challenging, in part because the blood-brain barrier - which prevents harmful substances in the bloodstream from entering the brain - also prevents certain biomarkers that would reveal the presence of a brain tumor from entering the circulatory system, where they could be picked up by a blood test.
The presence of brain cancer is hard to detect without expensive scans, which may not be ordered before the cancer has already progressed to the point where it is causing detectable symptoms. However, even early in a tumor's growth the disease will often cause changes in the body's immune response. Tumor cells can also shed DNA pieces into the bloodstream that have different properties than the DNA released from normal tissues. By searching for patterns consistent with these effects in blood samples, the machine-learning tool that Mathios and his collaborators developed was able to reliably detect brain cancer without invasive or expensive procedures.
The researchers estimated that widespread adoption of this tool for screening of individuals presenting with symptoms such as headaches could potentially lead to earlier detection of as many as 1,700 additional brain cancer cases a year in the U.S., which would double the number of patients detected without significantly increasing the use of imaging studies.
