X-ray sowing lung cancer
Credit: Peter Dazeley/Getty Images

To identify the 10% to 20% of lung cancers in people who have never smoked, researchers have developed an AI-deep learning too that flags non-smokers at high-risk for lung cancer from a single chest X-ray.

Known as the Chest-X-ray (CXR) Lung-Risk tool, researchers from Massachusetts General Hospital presented data at the RSNA meeting this week showing results of a 17,407-patient study testing its predictive ability in a clinical setting.

“We do not have a good way of identifying people who are high risk for lung cancer who don’t smoke,” says senior author Michael Lu, MD, MPH, director of artificial intelligence and co-director of the Cardiovascular Imaging Research Center CIRC at MGH. Currently, the tools physicians use clinically now are based on pack-based smoking history. “Increasingly, though, a larger proportion of people who get lung cancer are never smokers and we do not have a way to identify them early. This tool is designed to provide a means of finding high-risk patients from a routine chest X-ray, which may already exist in a patient’s electronic medical record.”  Lu’s team has already published studies of the tool looking at chest X-rays in smokers to predict who is most likely to get lung cancer in the next six years or so.

The patients in the current study were a group of never-smokers having routine outpatient chest X-rays from 2013 to 2014. The primary outcome was a six-year incident lung cancer. Using the CXR tool, risk scores were then converted to low, moderate, and high-risk groups.

Of the total patients (mean age 63 years), 28% were deemed high-risk by the deep learning model, and 2.9% of these patients later had a diagnosis of lung cancer. The high-risk group exceeded the 1.3% six-year risk threshold where lung cancer screening CT is recommended by National Comprehensive Cancer Network guidelines.

“The study found that the tool identifies a group of non-smokers that would be considered a high-risk population,” says Lu. “Since that group had more than double the six-year risk of lung cancer, it’s a strong reason to consider CT-based lung cancer screening in these individuals.”

Lu does not consider the tool a screening test. “Instead, the idea would be that you can identify these never smokers who are high-risk for lung cancer just based on their routine chest X rays they get for a cough or fever, etc.  If you could do that, potentially those patients might benefit from lung cancer screening CT.”

The CXR-Lung-Risk model was developed using 147,497 chest X-rays of 40,643 asymptomatic smokers and never-smokers from the Prostate, Lung, Colorectal, and Ovarian (PLCO) cancer screening trial to predict lung-related mortality risk, based on a single chest X-ray image as input.

“The only input to this AI model is the chest X-ray,” adds Lu, explaining that the team showed an AI tool many tens of thousands of patients’ X-rays, along with what happened to that person in the following 6-12 years. “By doing that, you can teach these tools to make a prediction about how likely it is that a person will get lung cancer.”

An advantage of the AI tool is that it is reproducible and could sit in an electronic record and scan all the chest X-rays and flag people who it identifies as high-risk and might benefit from additional testing.

Also of Interest