AI-produced CT biomarkers predict heart attack, death—topping traditional risk scoring

Researchers from the National Institutes of Health have created an artificial intelligence tool capable of extracting information from screening exams to assess a patient’s risk of experiencing a major cardiovascular event, according to a new study.

The algorithm analyzes abdominal CT scans performed during colorectal cancer screening to find important heart-related risk information. Writing March 2 in the Lancet Digital Health, the researchers found that their CT biomarkers are more accurate than current standards—Framingham risk score and body mass index.

It’s a reminder of the untapped potential hidden within the 80 million body CT scans performed each year in the U.S., said senior author Ronald M. Summers, MD, PhD, with the NIH Clinical Center.

“We found that automated measures provided more accurate risk assessments than established clinical biomarkers,” Summers said in a statement. “This demonstrates the potential of an approach that uses AI to tap into the biometric data embedded in all such scans performed for a wide range of other indications, and derive information that can help people better understand their overall health and risks of serious adverse events.”

For their study, Summers and colleagues at the University of Wisconsin had five AI computer programs gather a number of clinical measures from CT scans. Those included liver volume and fatty change, visceral fat volume, skeletal muscle volume, spine bone mineral density, and artery narrowing. In total, the researchers involved 9,223 people in their analysis.

Summers et al. determined that the CT biomarkers could more accurately predict downstream heart attack, cerebrovascular accident, congestive heart failure or death, compared to the Framingham score and BMI. Additionally, the AI-based aortic calcification measurements—calcium deposits in the aortic valve—were far more accurate at assessing major heart events and overall survival.

Artificial intelligence also handily beat out body mass index at predicting adverse cardio events, the team noted. 

Going forward the team plans to test their technique in other studies, including a more racially diverse population, said Perry J. Pickhardt, MD, with the University of Wisconsin School of Medicine & Public Health.

“This opportunistic use of additional CT-based biomarkers provides objective value to what doctors are already doing,” Pickhardt added. “This automated process requires no additional time, effort or radiation exposure to patients, yet these prognostic measures could one day impact patient health through presymptomatic detection of elevated cardiovascular or other health risks.”