Machine Learning Analysis of Routine CT Scans Reveals Hidden Heart Risks

New research from the SCOT-HEART trial suggests that coronary CT angiography (CCTA) holds untapped potential for predicting long-term cardiovascular outcomes—by looking beyond the heart.

Michelle C. Williams, MBChB, PhD, professor of cardiovascular imaging at the University of Edinburgh, led a team that applied deep learning to routine CCTA images, finding that body composition markers—especially skeletal muscle quality—can serve as powerful indicators of future myocardial infarction (MI) and all-cause mortality.

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

The study analyzed data from 1,722 participants in the SCOT-HEART trial, a large-scale effort to evaluate the role of CCTA in diagnosing and managing coronary artery disease. Over a 10-year follow-up period, 133 deaths and 106 MIs were recorded. Using wide field-of-view images, researchers applied the TotalSegmentator deep learning model to automatically segment major organs and tissue structures throughout the torso.

By quantifying both volume and attenuation (a measure of tissue density) across multiple structures, the team found that certain features—such as higher fat volume and lower attenuation of lungs and liver—were associated with coronary artery disease. But when it came to predicting major outcomes like MI and death, skeletal muscle attenuation proved to be the most telling.

Patients with skeletal muscle attenuation below the median were 85% more likely to die and 58% more likely to experience MI. Even after adjusting for traditional coronary calcium scores, skeletal muscle quality remained a strong, independent predictor of heart attack risk.

“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,” said Professor Williams. “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.”

The answer, it seems, is yes. The analysis revealed that muscle quality may offer unique insight into a patient’s overall cardiovascular resilience. “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,” Williams added. “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 use of machine learning was key to making the study possible. Previous efforts to analyze extra-cardiac features required painstaking manual segmentation. Now, with tools like TotalSegmentator, fully automated analysis of dozens of structures is possible in minutes.

“This information could be readily available for clinicians at the time of reporting,” said Professor Williams. “The models are very quick to run. However, we need to work out how best to combine this information into radiology reports.”

The findings point to a future where CCTA may serve as more than a coronary imaging test—it could become a comprehensive, whole-body risk assessment tool.