AI body composition measurements can predict cardiometabolic risk

Credit: Pixabay/CC0 Public Domain

Adiposity—or the accumulation of excess fat in the body—is a known driver of cardiometabolic diseases such as heart disease, stroke, type 2 diabetes, and kidney disease. But getting the full picture of a person’s risk is harder than it may seem. Traditional measures such as body mass index (BMI) are imperfect, conflating fat and muscle mass and not capturing where in the body fat is located.

A new study by researchers at Mass General Brigham and their colleagues found that an AI tool designed to measure body composition could accurately capture details in just three minutes from a body scan.

Their results, published in Annals of Internal Medicine, show that not all fat is equally harmful and highlight the potential of using AI to repurpose data from routine scans.

“We are hoping that these findings could be used to develop an ‘opportunistic screening’ tool to repurpose existing MRI and CT scans taken at the hospital to find patients with high-risk body composition who may be flying under the radar and could benefit from targeted diabetes and cardiovascular disease prevention,” said co-senior author Vineet K. Raghu, Ph.D., a computational scientist with the Mass General Brigham Heart and Vascular Institute.

Raghu and colleagues conducted a prospective cohort study using data from the U.K. Biobank. The researchers used whole-body MRIs from more than 33,000 adults with no prior history of diabetes or cardiovascular events who were followed for a median of 4.2 years.

The team found that in both men and women, AI-derived visceral adipose tissue volume (fat surrounding the abdominal organs) and fat deposits in muscle were strongly associated with diabetes and cardiovascular disease risk beyond standard measures of obesity like BMI and waist circumference. In men only, lower skeletal muscle volume was strongly associated with risk.

The authors note that future studies are needed to determine if their findings are generalizable and if AI can reliably measure these body composition metrics from routine scans. With further validation, an AI-driven approach could help leverage routine imaging to identify patients at high risk.