Abstract Details

Visceral abdominal fat can predict health outcomes with potential applications on brain health and dementia prevention. Deep learning of efficient whole body MR scans can amplify the success of these objectives.

To apply deep learning to an investigation of body fat on brain atrophy.

A total of 7507 healthy participants from 4 sites were scanned on 1.5T MRI with a rapid whole-body imaging. Sequences included whole body coronal T1 and brain MPRAGE. Deep learning with FastSurfer trained 134 participants aged 27-66 and segmented 96 brain regions. Deep learning also segmented abdominal visceral fat from the scan. Partial correlations of visceral abdominal fat volume and brain volumes were controlled for age, sex, and total intracranial volume. Multiple comparisons were accounted for using the Benjamini-Hochberg False Discovery Rate of 5% Logistic regression models determined risk of brain total gray and white matter atrophy based upon the highest quartile of visceral fat and lowest quartile of total gray and white matter.

Average age was 52.90 ± 13.04 years, spanning 15-97 years with 52.4% men and 47.6% women. Mean visceral abdominal fat was 2981.37 ± 2167.27 ml, ranging from 94.63 to 14005.22 ml. Visceral abdominal fat predicted brain atrophy in: gray and white matter volumes (Partial R= -.24, p= 7.01e-97), hippocampus (Partial R= -.09, p= 3.34e-14), temporal lobes (Partial R= -.23, 4.64e-90), parietal lobes (Partial R= -.11, p=5.84e-25), and right precuneus (Partial R= -.05, p= 2.15e-5). Visceral fat predicted increased risk for lower total gray matter (age 20-39: OR = 3.33; age 40-59, OR = 3; 60-80, OR = 2.85) and white matter atrophy age 20-39: OR = 1.26; age 40-59, OR = 2.05; 60-80, OR = 2.87).

Deep learning determined increased visceral abdominal fat volume predicts brain volume loss, representing a novel modifiable factor in determining brain health.