RSNA 2019: Deep Learning Tool and Cardiovascular Health on Lung Cancer Screenings
Posted: Wednesday, December 11, 2019
Coronary artery calcium scores, derived from chest CT scans, are a well-established method that helps physicians decide whether patients with lung cancer require preventative statins. However, a sophisticated type of artificial intelligence known as deep learning may help detect coronary artery calcium on chest CT images more efficiently than manual methods. Study findings, presented at the 2019 Radiological Society of North America (RSNA) Scientific Assembly and Annual Meeting in Chicago (Abstract RC303-02), showed that the tool appears to be capable of accurately evaluating large numbers of patients more quickly than human readers and may help to assign patients into high- and low-risk categories.
“For select patients at intermediate risk of heart disease, if the calcium score is 0, [a] statin can be deferred,” Michael T. Lu, MD, MPH, of Massachusetts General Hospital, Boston, explained in an RSNA press release. “If the calcium score is high, then those patients should be on a statin.”
The research team developed the deep learning algorithm using 1,600 cardiac CT scans from human-read measurements for reference. The deep learning calcium score was stratified into high, moderate, low, and very-low levels. The prognostic value of the tool was tested on 14,959 participants from the National Lung Screening Trial (NLST).
The researchers reported a significant association between the deep learning calcium score and all-cause mortality (P < .001). The automated coronary artery calcium score also corresponded strongly with cardiovascular mortality (P < .001). The intraclass correlation coefficient between manual and automated calcium classes was 0.858.
“There’s information about cardiovascular health on these CT scans,” Dr. Lu stated. “This is an automated way to extract that information, which can help patients and physicians make decisions about preventative therapy.”
Disclosure: For disclosures of the study authors, visit meeting.rsna.org.