Test 5 Visceral and Subcutaneous Abdominal Fat Predict Brain Volume Loss at Midlife in 10,001 Individuals

Cyrus A Raji^1, *, Somayeh Meysami^2, 3, Sam Hashemi^4, 5, Saurabh Garg^4, 5, Nasrin Akbari^4, 5, Ahmed Gouda^4, 5, Yosef Gavriel Chodakiewitz⁴, Thanh Duc Nguyen^4, 5, Kellyann Niotis^6, 7, David A Merrill^2, 8, 9, 
Rajpaul Attariwala^4, 5, 10

Abstract

Abdominal fat is increasingly linked to brain health. A total of 10,001 healthy participants were scanned on 1.5T MRI with a short whole-body MR imaging protocol. Deep learning with FastSurfer segmented 96 brain regions. Separate models segmented visceral and subcutaneous abdominal fat. Regression analyses of abdominal fat types and normalized brain volumes were evaluated, controlling for age and sex. Logistic regression models determined the risk of brain total gray and white matter volume loss from the highest quartile of visceral fat and lowest quartile of these brain volumes. This cohort had an average age of 52.9 ± 13.1 years with 52.8% men and 47.2% women. Segmented visceral abdominal fat predicted lower volumes in multiple regions including: total gray matter volume (r = -.44, p<.001), total white matter volume (r =-.41, p<.001), hippocampus (r = -.39, p< .001), frontal cortex (r = -.42, p<.001), temporal lobes (r = -.44, p<.001), parietal lobes (r = -.39, p<.001), occipital lobes (r =-.37, p<.001). Women showed lower brain volumes than men related to increased visceral fat. Visceral fat predicted increased risk for lower total gray matter (age 20-39: OR = 5.9; age 40-59, OR = 5.4; 60-80, OR = 5.1) and low white matter volume: (age 20-39: OR = 3.78; age 40-59, OR = 4.4; 60-80, OR = 5.1). Higher subcutaneous fat is related to brain volume loss. Elevated visceral and subcutaneous fat predicted lower brain volumes and may represent novel modifiable factors in determining brain health.

Key words

visceral fat / subcutaneous fat / brain volume loss / deep learning

INTRODUCTION

North American populations suffer a high prevalence of obesity with approximately 36% in the U.S. and 25% in Canada estimated as obese [1, 2]. Overweight and obese persons combined reach rates of over 60% in each country [1, 3]. High body adiposity across overweight and obese status has numerous cardiovascular risks. These include hypertension, hyperlipidemia, type 2 diabetes mellitus and atherosclerotic heart disease [3, 4]. Such complications, in turn, increase both morbidity and mortality particularly with aging populations. Thus, the elevated cardiovascular risk profile associated with obesity can also influence public health as mitigation and prevention strategies may reduce the rates of subsequent vascular disease.

Multiple prior investigations have suggested a connection between body fat accumulation and increased Alzheimer’s dementia (AD) risk. A study by Whitmer et al. (2008) demonstrated that midlife obesity translated to an increased risk of late life AD, suggesting a long-term influence of obesity on brain health [5]. Another study [6] reported that midlife overweight or obese status raised dementia risk with odds ratios of 1.71 and 3.88 respectively. A meta-analysis by Pedditizi et al. (2016), further noted that while midlife obesity does increase dementia risk, the relationship reverses in late life [7]. These results suggest that characterizing obesity earlier in life is important for understanding increased AD risk. Kivimäki et al. confirmed these results in a larger follow up study of 1.3 million persons that showed the risk of dementia in relation to Body Mass Index (BMI) increases when that metric is evaluated greater than 20 years prior to dementia diagnosis versus 10-20 or 10 years [8]. Thus, the public health implications of higher body fat tissue extend beyond cardiovascular considerations and into brain health.

One underlying observation that is thought to explain the high risk of AD with obesity is the increased burden of brain atrophy in this population. An early study demonstrated lower volume as a function of higher BMI in 94 cognitively normal persons who remain so 5 years after their scan [9]. These observations were replicated both in larger community and referral clinic cohort samples [10, 11]. Population based cross-sectional work done with the U.K. Biobank of over 12,000 individuals also showed lower brain volumes related to higher BMI [12].

However, when trying to understand the relationship between body fat tissue, brain health, and potential downstream effects such as higher AD risk, BMI has several limitations. Computed in part by the ratio of weight in pounds to height in inches, BMI is a surrogate of human body fat, not a direct measure, as it includes bone and muscle mass [13]. As such, this measure does not singularly track nor characterize the anatomical distribution of body fat. This distribution is characterized in two main types: visceral fat (vfat) or visceral adipose tissue that deposits around organs and subcutaneous fat (sfat) or subcutaneous adipose tissue, the latter type accounting for 80-90% of variance in BMI while visceral fat only accounts for 10-20% [14]. This distinction has potential importance to brain health as visceral fat has been related to both higher dementia risk [15] and brain atrophy [16]. However, studies examining the relationship between both of these fat types and underlying brain structure in larger populations are lacking in the literature.

We therefore aimed to investigate the associational relationships between visceral, subcutaneous fat and brain structure on MR imaging in a large cross-sectional sample of individuals across the lifespan. Our hypothesis is that visceral fat will be related to lower brain volumes, a biomarker for neurodegeneration [17], from the macrostructural tissue class perspective to lobar brain volumes and Alzheimer disease specific regions affected early in the pathological process: the hippocampus, posterior cingulate, and precuneus. We also evaluated these questions with subcutaneous fat.

MATERIALS AND METHODS

Participant Whole Body MR Imaging

All analyses were done with IRB approval (Advarra, WPBP-001). Participants were scanned on 1.5T Philips Ingenia Ambition, Siemens Espree and Aera scanners at the following locations: Vancouver, BC, Canada; Redwood City, CA; Los Angeles, CA; Minneapolis, MN; Boca Raton, FL; Dallas, TX. Each participant received a non-contrast whole body MRI scan that has been previously detailed [18]. Briefly, each scan included whole body coronal T1-and axial T2-weighted with whole body coronal STIR, axial T1 in and out of phase images with Dixon technique [19, 20] allowing for visual identification and quantification of vfat and sfat. Additional whole-body sequences included diffusion weighted imaging (DWI), STIR, and axial T2 BLADE of the chest, abdomen and pelvis. Brain sequences included sagittal 3D T1 MPRAGE, axial 2D FLAIR, and time-of-flight MRA but only 3D T1 MPRAGE brain images were the focus of neuroimaging analysis in this work.

Deep Learning Analyses and Quantification of Body Fat

We used a T1 weighted MRI scan of the whole-body to segment the visceral fat among 28 anatomical structures. The dataset consists of 102 Siemens MRI scans, and it was divided into 72 scans for training, 18 for validation, and 12 for testing with an age range of 27-66 years. All anatomical structures were manually annotated by radiologists using ITK-SNAP[21]. QC was performed by another trained radiologist for generating the ground-truth masks. We used nnU-Net model[22], as a fully supervised segmentation architecture for training and inferring the different segmentation classes. The average testing dice score for the visceral fat is (0.8402 ± 0.07). Finally, the visceral fat and subcutaneous fat volumes for each patient were computed in milliliters (ml) by multiplying the number of its predicted voxels by the 3D MRI voxel space.