Artificial intelligence can boost the value of lung cancer screening, using exams to predict the risk of death from other diseases, according to a study published Tuesday in Radiology [1].

Along with the lungs, low-dose computed tomography scans provide valuable information about other parts of the chest. Scientists from Vanderbilt University previously created an AI algorithm able to analyze LDCT scans to automatically derive body composition measurements—a useful metric for forecasting chronic health conditions.  

For the new study, lead investigator Kaiwen Xu and colleagues tested the technology on CT images from nearly 21,000 individuals involved in the National Lung Screening Trial. They found promising results, with AI measurements improving predictions for patients’ chance of death from lung cancer and cardiovascular disease, along with all-cause mortality.

“Automatic AI body composition potentially extends the value of lung screening with low-dose CT beyond the early detection of lung cancer,” Xu, a PhD candidate in the Department of Computer Science at Vanderbilt, said in a July 25 announcement from the Radiological Society of North America. “It can help us identify high-risk individuals for interventions like physical conditioning or lifestyle modifications, even at a very early stage before the onset of disease.”

The final study sample included 865 participants who were diagnosed with lung cancer by the end of 2009 and 4,180 participants who died by 2015. Of those, 913 deaths were attributed to lung cancer and 972 to cardiovascular disease. Measurements associated with fat located within muscle, in particular, were “strong” predictors of mortality, a finding that aligns with previous studies. Body composition metrics, however, did not improve the prediction of lung cancer incidence.

Experts see great promise for improving population health outcomes with such opportunistic screening via low-dose CT. Current eligibility criteria for lung cancer screening recommend that millions of Americans undergo the exam on an annual basis. However, challenges remain before adding AI to real-world screening programs.

“Obtaining regulatory clearance, building the necessary data pipelines, and implementing quality control of the output generated by AI algorithms requires effort and resources,” Florian J. Fintelmann, MD, a radiologist with Massachusetts General Hospital, wrote in a corresponding editorial [2]. “Currently, generating revenue with automated body composition analysis to offset the necessary investment without a billing code for CT-based analysis of muscle and adipose tissue seems like a stretch, however.”