Artificial intelligence can calculate body composition (BC) from a routine magnetic resonance imaging scan in less than three minutes and use the information to spot people at elevated cardiometabolic risk, research indicates.
The research, in the Annals of Internal Medicine, suggests it may be possible to opportunistically assess body composition from routine imaging to identify high-risk groups.
It also provides further evidence that cardiometabolic risk depends on not just on whether fat is present but where it is located.
Only visceral adipose tissue (VAT) and skeletal muscle fat were linked with incident diabetes and major adverse cardiovascular events (MACE) in the open-source deep learning (DL) model.
Subcutaneous adipose tissue (SAT), lying just below the skin, showed no such associations.
“These results corroborate evidence that VAT, but not SAT, is a key driver of adiposity-related cardiometabolic risk,” reported Matthias Jung, MD, from the University of Freiburg, and co-workers.
Currently, definitions of obesity rely on body mass index (BMI), but this conflates excess adiposity with muscle mass and does not account for the location of body fat, which is a critical factor in assessing obesity-related cardiometabolic risk.
Jung and team investigated alternatives using information on whole-body MRIs from 33,432 participants in the UK Biobank who did not have a history of diabetes, myocardial infarction, or ischemic stroke. Their mean age was 65.0 years, and 52.8% of the group was female. Mean body mass index was 25.8 kg/m2.
An open-source AI model was used to estimate 3D body composition volumes. These included subcutaneous adipose tissue, visceral adipose tissue, skeletal muscle volume, and skeletal muscle fat fraction percentage.
During a median follow-up of 4.2 years, 1.1% of women and 2.2% of men were diagnosed with incident diabetes and a corresponding 1.0% and 2.3% were diagnosed with MACE, defined as myocardial infarction or ischemic stroke.
After adjusting for age, smoking, and hypertension, the researchers found that increased adiposity and decreased skeletal muscle proportion were associated with higher incidence of diabetes and MACE in both sexes.
After additionally adjusting for BMI and waist circumference, only elevated visceral adipose tissue and high skeletal muscle fat fraction remained associated with an increased risk of diabetes, with adjusted hazard ratios (aHRs) for men in the top 20% of 1.84 in each case.
For women, only the aHR with visceral adipose tissue did not cross one, with a value of 2.16 for diabetes among those in the top fifth percentile.
Elevated skeletal muscle fat fraction was associated with MACE in women, at an aHR of 1.37 for visceral adipose tissue and 1.72 for skeletal muscle fat fraction, but the confidence intervals crossed one in each case for men.
Among men only, those in the bottom 20% of skeletal muscle levels had an increased risk of both diabetes and MACE, with aHRs of 1.96 and 1.55, respectively.
There was a negative associated between relative subcutaneous adipose tissue and the risk of future diabetes in females, such that women in the bottom fifth had an aHR of 1.46 and those in the top fifth had an aHR of 0.71.
“Growing evidence suggests that BC plays a key role not only in cardiometabolic but also oncologic diseases to personalize risk estimation,” the researchers note.
“In addition, BC could also play a crucial role in estimating treatment tolerability and risk for treatment-related toxicity.
“In this context, beyond defining excess adiposity, our DL model and relative BC profiles could be used as a measure of frailty or overall health to improve treatment decisions for accurate, personalized dosing of systemic drug therapies, including chemotherapy and immunotherapy.”
