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

High accuracy for clinical use: The model achieved prediction errors of approximately 1.7-2.5 cm across genders, with error rates below 1.14%, suggesting sufficient precision for supporting clinical decision-making in pediatric growth assessment.

Transparency through explainability: The researchers used explainable AI methods to identify that height standard deviation scores, height velocity, and soft lean mass velocity were the most important predictors, allowing clinicians to understand which factors drive the model's predictions rather than treating it as a "black box."

Large, real-world dataset: The model was trained on over 588,000 measurements from nearly 100,000 children, providing a robust foundation that likely captures diverse populations and growth patterns rather than relying on limited or homogeneous data.

Personalized rather than one-size-fits-all: The model generates individualized growth curves for each child rather than applying population averages, potentially enabling more tailored clinical interventions for growth disorders and better outcomes for individual patients.

Intended for clinical support, not replacement: The paper frames the tool as supporting clinical decision-making and early identification of growth disorders, suggesting it's designed to augment rather than replace physician judgment in pediatric care.