A device-agnostic shape model for automated body composition estimates from 3D optical scans

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Anthropometry

Background:

• Metabolic disease morbidity is linked to body shape, with 3D optical (3DO) scanning offering a safe, accessible method for body composition assessment.
• 3DO scanning accurately predicts body composition variables linked to mortality risk, surpassing traditional methods like DXA in cost and accessibility.
• Standardization issues across 3DO scanner manufacturers (pose, mesh resolution, post-processing) hinder consistent data interpretation.

Purpose of the Study:

• Introduce a scanner-agnostic algorithm to standardize 3DO scans by fitting consistent human meshes to point clouds.
• Develop a standardized body shape model to predict clinically relevant body metrics from any 3DO scanner.
• Enable consistent body composition analysis across different 3DO devices through automated mesh generation and standardized outputs.

Main Methods:

• Automated registration of a fixed-topology body mesh template to 848 training scans from three 3DO systems using PCA-based domain deformation.
• Optimization of the template mesh to fit target scans via smooth, per-vertex surface-to-surface deformation.
• Training linear regression models to predict body composition from mesh PCA coefficients, validated against DXA measurements on 562 withheld scans.

Main Results:

• Achieved R-squared values above 0.8 for nine fat and lean predictions, with a maximum of 0.94 for total fat and trunk fat.
• Root-mean-squared errors for all predictions were below 3.0 kg, with most body composition variables showing no significant difference from DXA.
• Repeatability precision (%CV) was 2-3x lower than DXA, with visceral fat and female total fat mass showing %CVs below 2x and 5x DXA, respectively.

Conclusions:

• The developed method provides an accurate, automated, and scanner-agnostic framework for standardizing 3DO scans.
• This approach offers a low-cost, radiation-free alternative to traditional radiology for body composition analysis.
• A web application is available for demonstrating the mesh templating and body composition prediction capabilities.