Routine CT Scans Offer New Clues to Cardiovascular Risk

New findings from the Scottish Computed Tomography of the HEART (SCOT-HEART) trial show that machine learning analysis of body composition on routine coronary CT angiography (CCTA) can offer valuable prognostic insights that extend well beyond traditional coronary assessment. 

Michelle C. Williams, MBChB, PhD, professor of cardiovascular imaging at the University of Edinburgh, and her team found that multi-organ metrics, especially skeletal muscle attenuation, were strong, independent predictors of myocardial infarction (MI) and all-cause mortality over 10 years. 

“These findings may help radiologists identify high-risk patients who could benefit from earlier or more intensive intervention,” Professor Williams said. 

The study involved 1,722 participants from the SCOT-HEART trial, with 133 deaths and 106 MIs recorded over a decade of follow-up. Using wide field-of-view CCTA images, Professor Williams and her team applied the TotalSegmentator deep learning model to automatically segment organs and tissues throughout the torso.

For each structure, the team calculated volume and mean attenuation, then used multivariate models to evaluate associations with MI and mortality.

The analysis revealed several organ and tissue measures showed significant relationships with outcomes. For example, lower attenuation of lungs and liver, and higher torso fat volume, were all associated with coronary artery disease. 

When it came to mortality, the researchers found correlations with multiple organ volumes, including skeletal muscle, vertebrae, costal cartilage, aorta, left and right ventricles, and with attenuation of ribs, vertebrae and skeletal muscle. Notably, reduced skeletal muscle attenuation emerged as a key independent predictor of both MI and mortality.

Skeletal Muscle Quality Drives Heart Risk Insights

 “We know that separately other things are important for cardiovascular risk—the rest of the heart, the lungs, the liver, etc. But previous research has always looked at these things separately,” Professor Williams explained. “I wanted a way to look at the interplay between all these different factors and wondered whether it might improve our prediction of cardiovascular risk and mortality.”

Skeletal muscle metrics stood out in particular: patients with attenuation values below the median were 85% more likely to die and 58% more likely to experience MI. After adjusting for coronary calcium, skeletal muscle attenuation remained the only metric significantly associated with MI risk.

“I was surprised and excited about the results because they tell us that lots of other parts of the body are important when considering cardiovascular risk,” Professor Williams said. “In particular, skeletal muscle quality is an important predictor of outcomes beyond other findings. I knew that exercise was important for cardiovascular health, but this research provides tangible evidence for this association.”

The team relied on machine learning to make the study feasible. “Previously my research has focused on individual parts of the body, like the liver and the bones, and used time-consuming manual segmentation,” Professor Williams said. “So, an important part of this research is the use of machine learning tools to automatically segment everything on the CT scan.”

Integrating these tools into routine workflow would require no changes to image acquisition and minimal adjustment during interpretation. “This information could be readily available for clinicians at the time of reporting,” Professor Williams noted. “The models are very quick to run. However, we need to work out how best to combine this information into radiology reports."

Access the presentation, “Multi-Organ Machine Learning Analysis of Computed Tomography Body Composition and Myocardial Infarction and Mortality In The Scot-Heart Trial,” (M6-STCE2-2) on demand at RSNA.org/MeetingCentral.