Improving Optical Body-Composition Estimates with a Standardized Hand Photograph: a DXA-Referenced Study - 26.28
Title:
Improving Optical Body-Composition Estimates with a Standardized Hand Photograph: a DXA-Referenced Study
Authors:
Matthew S. GILMER 1, Chiara BUSSO 2, Marta BOCCARDO 2, Marco A. MINETTO 2, Steven C. HAUSER 1, David BRUNER 1
1 Size Stream LLC, Cary NC, USA;
2 Division of Physical Medicine and Rehabilitation, Department of Surgical Sciences, University of Turin, Turin, Italy
Keywords:
body composition; optical body scanning; hand anthropometry; dual-energy X-ray absorptiometry; tabular foundation model; machine learning
Abstract:
Low-burden, scalable body-composition assessment is of growing interest for clinical and consumer use. Optical methods that estimate composition from body silhouettes are an established low-friction option, but the value of a single localized image, such as a photograph of the hand, has not been characterized. We evaluated whether a standardized hand photograph can improve the body- composition estimates produced by Size Stream's MS1 application, which estimates composition from height, age, weight, and gender together with front and side silhouettes, using dual-energy X-ray absorptiometry (DXA; Hologic Horizon Wi) as the reference. In 50 adults, two MS1 scans and two hand photographs were collected per subject; after MediaPipe landmark detection and segmentation, a homography rectification mapped the photograph to millimeters and 31 hand measurements and shape ratios were extracted and reduced to 17 by an a-priori collinearity screen. MS1 outputs were calibrated onto the DXA scale, and for each metric the residual (DXA minus calibrated MS1 estimate) was modeled from the hand features, the MS1 estimate itself, and gender. TabPFN (a Tabular Prior-data Fitted Network, a pretrained tabular foundation model) and an interpretable ordinary-least-squares (OLS) correction were fit on the same nested-selected features and evaluated by leave-one-out cross- validation (LOOCV). The interpretable OLS correction was the preferred model for most metrics, with TabPFN offering only a marginal, VAT-specific advantage. Lean mass of the legs (LML) improved ~24% in MAE (p=0.02–0.03), visceral adipose tissue (VAT) ~35% in MAE and ~37% in SD (p=0.01–0.05), and bone mineral content (BMC) ~20% in both MAE and SD, the SD gain significant (paired-t p=0.009, Wilcoxon p=0.046) while the MAE gain was not. Lean mass of the arms (LMA) also improved ~43% in MAE (p<0.001), but at this sample size the arm correction used only the recalibrated MS1 estimate and no hand feature. Body-fat percentage (BFP) showed only a small, non-significant improvement; its correction selected two hand features but did not significantly improve on the MS1 estimate. A standardized hand photograph is therefore a useful additive signal for optical body-composition assessment, especially for leg lean mass, visceral fat, and the precision of bone estimates.
Full paper:
Note:
This paper is currently under review for publication in the 3DBDOY.TECH Journal - Vol. 3, Dec. 2026.
It will be updated also in this page once the review process is concluded.
Presentation:
Full video will be available here in Q3.2027; it is already availble in the proceedings (purchase order).
How to Cite (MLA):
Gilmer, Matthew S. et al., "Improving Optical Body-Composition Estimates with a Standardized Hand Photograph: a DXA-Referenced Study", Proceedings of 3DBODY.TECH 2026 - 17th International Conference and Expo on 3D/4D Body Technologies, Lugano, Switzerland, 20-21 Oct. 2026, #28, https://doi.org/10.15221/26.28
Details:
Proceedings: 3DBODY.TECH 2026, 20-21 Oct. 2026, Lugano, Switzerland
Paper/Presentation: #28
DOI: https://doi.org/10.15221/26.28
License/Copyright notice
Proceedings: © Hometrica Consulting - Dr. Nicola D'Apuzzo, Switzerland, hometrica.ch.
Authors retain all rights to individual papers, which are licensed under a Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
The papers appearing in the proceedings reflect the author's opinions. Their inclusion in the proceedings does not necessary constitute endorsement by the editor or by the publisher.
Note: click the + on the top left of the page to open/close the menu.