- A machine learning approach based on convolutional neural networks enabled highly accurate estimation of visceral (VAT), abdominal subcutaneous (ASAT), and gluteofemoral (GFAT) adipose tissue volumes from body MRI in UK Biobank participants
- To quantify an individual's fat distribution out of proportion to their overall size, the authors derived body mass index-adjusted metrics for each fat depot (e.g., VAT adjusted for BMI, VATadjBMI)
- BMI-adjusted adipose tissue volumes exhibited depot-specific and divergent associations with type 2 diabetes and coronary artery disease. VATadjBMI was associated with increased risk, while GFATadjBMI was associated with decreased risk
- Individuals with particularly high VATadjBMI or particularly low GFATadjBMI, even if they were of normal BMI, were found to have significant elevated risk of cardiometabolic disease
- These BMI-adjusted adipose tissue volumes may prove to be useful endpoints in future trials examining the cardiometabolic implications of a given obesity intervention
It's well known that two individuals with the same body mass index (BMI) can have different cardiometabolic disease risks. Variation in fat distribution has been proposed as one potential explanation. Several prior studies have sought to use medical imaging-derived body composition measurements to understand this gap. Most of these studies have been small, cross-sectional, and excluded gluteofemoral adipose tissue (deposited at the hip, thighs, and buttocks), which may confer protection from cardiometabolic disease.
Subscribe to the latest updates from Cardiovascular Advances in Motion
Massachusetts General Hospital researchers used body MRI data from the UK Biobank, a large longitudinal observational study, to investigate this issue. In Nature Communications, they present more definite evidence that the location of specific "depots" of fat does affect the risk of cardiometabolic disease.
The authors are Saaket Agrawal, MD, and Marcus D.R. Klarqvist, PhD, from the Cardiovascular Disease Initiative of the Broad Institute of MIT and Harvard, Amit V. Khera, MD, MSc, a researcher at the Center for Genomic Medicine at Mass General, and colleagues.
The UK Biobank enrolled more than 500,000 individuals ages 40 to 69 between 2006 and 2010. As part of an imaging substudy, 43,531 underwent full-body MRI imaging, and—following exclusions for technical problems or artifacts in the collected images—40,032 participants were available for analysis (51% female, median age 65, 97% white).
Subsets of these participants had adipose tissue volumes previously measured and available for download from the UK Biobank:
- Visceral adipose tissue (VAT, n=9,040)
- Abdominal subcutaneous adipose tissue (ASAT, n=9,041)
- Gluteofemoral adipose tissue (GFAT, n=7,754)
For each fat depot, participants with available adipose tissue volume data were randomly split into 80% for model training and 20% for model testing. In each training set, the team trained a convolutional neural network (CNN) using corresponding body MRI data to predict the volume of each fat depot. A CNN is a form of artificial intelligence that can be trained to make predictions from images—in this case, body MRIs.
Estimating Fat Depot Volumes
Among the 20% of participants who were designated for model testing, CNNs demonstrated near-perfect ability to estimate each fat depot volume (r2 = 0.99, 0.99, and 0.98 for VAT, ASAT, and GFAT, respectively). Accuracy was similar across subgroups by age, sex, BMI, and self-reported race/ethnicity.
BMI-adjusted Fat Depots and Disease Prevalence
To examine the association of local adiposity with cardiometabolic diseases, the researchers adjusted VAT, ASAT, and GFAT volumes for BMI (e.g., VAT adjusted for BMI, VATadjBMI). These new variables reflect how an individual's fat distribution differed from those expected based on their BMI.
Significant heterogeneity was noted in how BMI-adjusted fat volumes affected the risk of prevalent cardiometabolic disease (i.e., already diagnosed on the day of imaging):
Prevalent type 2 diabetes
- VATadjBMI—Increased risk (OR, 1.49; P = 9.9 × 10−76)
- ASATadjBMI—Largely neutral effect (OR, 1.08; P=0.002)
- GFATadjBMI—Decreased risk (OR, 0.75; P = 6.4 × 10−28)
Prevalent coronary artery disease
- VATadjBMI—Increased risk (OR, 1.17; P = 3.0 × 10−11)
- ASATadjBMI—No effect (OR, 1.00; P=0.92)
- GFATadjBMI—Decreased risk (OR, 0.89; P= 3.5 × 10−5)
BMI-adjusted Fat Depots and Disease Incidence
Incident type 2 diabetes
- VATadjBMI—Increased risk (HR, 1.45; P = 1.3 × 10−11)
- ASATadjBMI—Largely neutral effect (OR, 0.96; P=0.49)
- GFATadjBMI—Decreased risk (OR, 0.84; P=0.005)
Incident coronary artery disease
- VATadjBMI—Increased risk (HR, 1.17; P = 8.1 × 10−5)
- ASATadjBMI—Largely neutral effect (OR, 1.04; P=0.41)
- GFATadjBMI—Marginally decreased risk (OR, 0.91; P=0.05)
These results add to the increasing evidence that different fat depots have distinct metabolic profiles. Changes in measures of local adiposity, independent of weight and BMI, may be reliable proxies of the cardiometabolic benefits of a given obesity intervention, and they warrant consideration as additional endpoints in clinical trials.
Learn more about the Center for Genomic Medicine
Refer a patient to the Corrigan Minehan Heart Center