Accurate Prediction of 3D Humanoid Avatars for Anthropometric Modeling

Objectives: The recent introduction of low-cost 3D optical imaging systems is revolutionizing anthropometric assessment of children and adults and leading to the availability of large digital avatar databases. One application of these humanoid avatars is to serve as the reference for developing manifold regression equations that can be used to predict a person's 3D avatar from their demographic (e.g., sex, age, etc.) and physical (e.g., weight, height, etc.) characteristics without being imaged first on a 3D scanner. Generated avatars using this method can be used for research, clinical, and educational purposes. Accordingly, the aim of the present study was to test the hypothesis that anthropometric dimensions derived from a manifold-regression predicted 3D humanoid avatar are accurate with respect to ground-truth methods. Methods: Adult participants (n, 98) completed dual-energy X-ray absorptiometry and 3D optical reference scans for body composition and ground-truth anthropometry, respectively. 3D multiple humanoid avatars were generated using manifold regression equations with core (sex, age, weight, and height) and supplementary (e.g., %fat, several impedance estimates). Reference avatars and generated avatars were analyzed with the same software for multiple circumferences and regional body surface areas and volumes. Results: Generated (e.g., with %fat in model) and reference avatar measurements were highly correlated and showed no significant mean differences or bias across 6 evaluated circumferences and 7 volumes with corresponding surface areas [e.g., waist circumference, R2 0.96, p< 0.001; X±SD, 97.0±19.0 cm, 98.8±20.2 cm; waist/hip ratio, R2, 0.74, p< 0.001; 0.91±0.09, 0.91±0.10; surface area, R2 0.97, p< 0.001;1.80±0.25 m2, 1.81±0.25 m2]. Conclusions: These results show that highly accurate human avatars can be created using demographic and minimal additional body composition information, thus opening up many new clinical and research possibilities. Funding Sources: Funded by a NIH/NIDDK grant.