Can AI refine body composition data gleaned from whole-body MRI?

Wednesday, December 2 | 3:00 p.m.- 4:00 p.m. | W7-SSMK08-2 | Room E450A

AI quantification of body composition from routine whole-body MRI helps assess patients' metabolic syndrome risk without additional clinical workup.

The finding could help clinicians better track patients' pathogenic fat distribution and guide preventive measures, according to study results to be presented Wednesday afternoon.

"Body mass index (BMI) fails to distinguish between metabolically distinct fat distribution patterns, inadequately capturing cardiovascular and metabolic disease risk," presenter Sam Hashemi, MSc, of Prenuvo in North Fair Oaks, CA, and colleagues noted. "The visceral-to-subcutaneous adipose tissue (VAT/SAT) ratio provides unique clinical information beyond BMI by directly reflecting pathogenic fat deposition versus benign storage."

Hashemi's group conducted a study that included 20,200 patients who underwent 1.5-tesla whole body-MRI. The team applied a nnU-Net deep-learning (DL) framework to quantify visceral adipose tissue (VAT), subcutaneous adipose tissue (SAT), and lower limb muscle volumes, then validated the model by comparing its segmentations against manual ones on a test set of 40 cases.

The investigators reported the following:

· The model's segmentation performance showed high agreement with the radiologist-vetted segmentations (Dice coefficient: 0.875 for VAT and 0.927 for SAT).

· After matching for BMI and adjusting for lifestyle factors, both VAT volume and the VAT/SAT ratio independently predicted diabetes risk.

· SAT volume alone showed no independent association with diabetes risk.

· Total fat, fat mass index, and total muscle volumes were not independently associated with diabetes.

· VAT measures and the VAT/SAT ratio proved to be independently predictive of hypertension.

"AI-derived VAT and the VAT/SAT ratio from whole body-MRI independently predict diabetes and hypertension…whereas SAT and total fat do not," the team concluded, noting that "these metrics offer a more precise assessment of metabolic risk than traditional anthropometric measures."

Page 1 of 2
Next Page