Artificial Intelligence for Pediatric Height Prediction Using Large-Scale Longitudinal Body Composition Data

Height g rowth serves as a key health indicator, reflecting the interplay of genetic, environmental, and socioeconomic factors (Norris et al., 2022; Baxter - Jones et al., 2011; Hargreaves et al., 2022). Monitoring height growth enables early detection of disorders, facilitating timely interventions (Saari et al., 2015; Craig et al., 2011; Grote et al., 2008; Zhang et al., 2016). Accurate future height prediction is essential for diagnosing growth disorders, initiating hormone therapy, and evaluating treatment efficacy (Collett - Solberg et al., 2019; Ostojic, 2013; Cuttler & Silvers, 2004). Traditional height prediction methods rely on skeletal maturity assessment using hand - wrist radiographs. These include the Bayley - Pinneau (Bayley and Pinneau, 1952), Tanner - Whitehouse (Tanner et al., 1975), and Roche - Wainer - Thissen (Roche et al., 1975) met hods. However, these approaches have limitations including radia tion exposure, the need for specialized expertise, and high interobserver variability (Bull et al., 1999; Chávez - Vázquez et al., 2024; Prokop - Piotrkowska et al., 2021).