Author(s): STRASSOLDO DI VILLANOVA, F., MORGANTI, G., VITARELLI, M., QUARANTELLI, M., ANDRIEU, B., RUSCELLO, B., PADUA, E., Institution: SAN RAFFAELE OPEN UNIVERSITY, ROME, Country: ITALY, Abstract-ID: 1965

INTRODUCTION:
Body composition (BC) assessment is critical in sports science and health promotion for monitoring nutritional status and adapting diet or training [1;2;3]. Gold standards like dual-energy X-ray absorptiometry (DXA), air displacement plethysmography (ADP), and bioimpedance analysis (BIA) are often limited by costs and device variability. Recently, AI-driven body scans offer accessible alternatives, with 2D mobile scans being more practical than 3D ones [4;5;6]. While DXA is the benchmark for tissue differentiation, ADP serves as a suitable reference for validating volumetric algorithms. Accordingly, this study aims to investigate accuracy and scientific validity, as well as precision (intra-operator/device), and reproducibility (inter-operator/device) of a mobile AI-2D scan for measuring multi-compartment BC data in comparison with ADP and BIA, as no prior research exists on this topic.
METHODS:
In this cross-sectional comparative study, 100 adults (age 18-70 y.o., BMI 18.5-29.9 kg/m²) were recruited. Assessment included height, weight, Body Volume (BV) via ADP (BODPOD), Total Body Water (TBW) via BIA (Tanita BC-418), and AI-2D scan (SizeYou®) on two devices (iPad, iPhone) by two operators (three consecutive measurements each). The app uses AI pose detection to extract body contours from two photos (posterior, lateral). Statistical analyses evaluated: accuracy using Bland-Altman and Pearson’s r; precision via ICC, SEM, CV, and MDC; reproducibility with paired t-tests, Cohen’s d, Pearson’s r, and MAPE.
RESULTS:
Preliminary analyses included 43 participants (age 26.2±7.0 y.o; 42% males, BMI 21.0±2.24 kg/m²; 58% females, BMI 18.0±2.62 kg/m²). Accuracy analyses showed BV was slightly underestimated (Bias≈ -1.1 L), while TBW was marginally overestimated (Bias=0.03–0.45 L). BV had wider Limits of Agreement(-18.3–16.1 L) than TBW (-3.4–3.9 L), but both showed strong correlations with gold standards (r≥0.78). Precision was nearly perfect for BV (ICC=1, SEM=0, CV≤0.1%, MDC=0) and TBW (ICC≥0.999, SEM≤0.26, CV≤0.6%, MDC≤0.72 L). Despite significant t-test differences between devices-operators (p<0.001, d>1), high correlations (r≥0.999), low mean differences (0.1L for BV; -0.4L for TBW), and MAPE (0,22% for BV; ≤0.4% for BW) support clinical interchangeability.
CONCLUSION:
SizeYou® provides validity and precision for evaluating BV and TBW. The app performs superiorly in BV as it relies on direct geometric reconstruction. While TBW correlates strongly with BIA (r≥0.97), both methods rely on indirect predictive modelling. Thus, discrepancies may result from BIA variability rather than app error. Repeatability enables device and operator interchangeability, ideal for longitudinal monitoring in sports and health promotion settings. In conclusion, this AI-driven 2D technology seems to offer a cost-effective alternative to BC assessment methods.

References:
1. Castro et al.,2020
2. Dehghani et al.,2025
3. Duren et al.,2008
4. Ferreira et al.,2025
5. Yamamoto et al.,2025
6. Graybeal et al.,2022