AI tool assessed body composition, diabetes risk using whole-body MRI
Patients whose MRIs showed high levels of visceral fat or fat as a fraction of skeletal muscle were more likely to develop diabetes, a study using artificial intelligence (AI) found. The results highlight the potential for opportunistic screening by AI, according to the authors.
An artificial intelligence (AI) tool effectively used whole-body MRIs to identify body composition factors that put patients at risk for diabetes or major cardiovascular events (MACE).
The prospective cohort study was based on whole-body MRI scans from 33,432 British patients with no history of diabetes, myocardial infarction, or ischemic stroke (mean age, 65 years; mean body mass index [BMI], 25.8 kg/m2). The AI tool used the MRI results to derive three-dimensional measures of body composition, including subcutaneous adipose tissue, visceral adipose tissue, skeletal muscle, and skeletal muscle fat fraction and compare those measures with development of diabetes or MACE during follow-up (median, 4.2 years). Results were published by Annals of Internal Medicine on Sept. 30.
After adjustment for age, smoking status, and hypertension, having greater adiposity or less muscle was associated with higher incidence of diabetes and MACE. After additional adjustment for BMI and waist circumference, only patients in the top fifth percentile for visceral fat or skeletal muscle fat fraction had a significantly elevated risk for diabetes. For men, being in the bottom category of skeletal muscle was also associated with diabetes risk.
The study's findings highlight the potential of this model as “an opportunistic screening strategy, where [body composition] data are automatically extracted from routine clinical MRI or CT scans, regardless of their initial indication,” said the study authors, noting that more research is needed to determine whether scans of sections of the body, such as the liver or kidneys, could provide the same predictions. “If successful, this model could be implemented in the electronic medical record without disrupting established workflows,” they added.
Limitations of the study include that all participants were White, older than age 45 years, and living in the United Kingdom.