Visual Body Composition Assessment Methods: A 4-Compartment Model Comparison of Smartphone-Based Artificial Intelligence For Body Composition Estimation In Heatlhy Adults
Kinesiology and Nutrition
Background & Aims: Visual body composition (VBC) estimates produced from smartphone-based artificial intelligence represent a user-friendly and convenient way to automate body composition remotely and without the inherent geographical and monetary restrictions of other body composition methods. However, there are limited studies that have assessed the reliability and agreement of this method and thus, the aim of this study was to evaluate VBC estimates compared to a 4-compartment (4C) criterion model.
Methods: A variety of body composition assessments were conducted across 184 healthy adult participants (114 F, 70 M) including dual-energy X-ray absorptiometry and bioimpedance spectroscopy for utilization in the 4C model and automated assessments produced from two smartphone applications (Amazon Halo®, HALO; and myBVI®) using either Apple® or Samsung® phones. Body composition components were compared to a 4C model using equivalence testing, root mean square error (RMSE), and Bland–Altman analysis. Separate analyses by sex and racial/ethnic groups were conducted. Precision metrics were conducted for 183 participants using intraclass correlation coefficients (ICC), root mean squared coefficients of variation (RMS-%CV) and precision error (PE).
Results: Only %fat produced from HALO devices demonstrated equivalence with the 4C model although mean differences for HALO were <±1.0 kg for FM and FFM. RMSEs ranged from 3.9% to 6.2% for %fat and 3.1–5.2 kg for FM and FFM. Proportional bias was apparent for %fat across all VBC applications but varied for FM and FFM. Validity metrics by sex and specific racial/ethnic groups varied across applications. All VBC applications were reliable for %fat, fat mass (FM), and fat-free mass (FFM) with ICCs ≥0.99, RMS-%CV between 0.7% and 4.3%, and PEs between 0.3% and 0.6% for %fat and 0.2–0.5 kg for FM and FFM including assessments between smartphone types.
Conclusions: Smartphone-based VBC estimates produce reliable body composition estimates but their equivalence with a 4C model varies by the body composition component being estimated and the VBC being employed. VBC estimates produced by HALO appear to have the lowest error, but proportional bias and estimates by sex and race vary across applications.
Graybeal, A. J.,
Brandner, C. F.,
Tinsley, G. M.
(2022). Visual Body Composition Assessment Methods: A 4-Compartment Model Comparison of Smartphone-Based Artificial Intelligence For Body Composition Estimation In Heatlhy Adults. Clinical Nutrition, 41(11), 2464-2472.
Available at: https://aquila.usm.edu/fac_pubs/20343