Artificial intelligence for pediatric height prediction using large-scale longitudinal body composition data

Area of Science:

• Pediatric endocrinology and growth assessment.
• Artificial intelligence in healthcare.
• Biometric data analysis for human development.

Background:

• Accurate prediction of childhood and adolescent height is crucial for monitoring growth and identifying potential disorders.
• Traditional growth assessment methods may lack precision and personalization.
• Advancements in AI offer new possibilities for sophisticated predictive modeling in pediatrics.

Purpose of the Study:

• To develop and validate a precise artificial intelligence (AI) model for predicting future height in children and adolescents.
• To leverage anthropometric and body composition data for accurate growth trajectory estimation.
• To enhance clinical decision support in pediatric growth assessment.

Main Methods:

• Utilized a large-scale Korean longitudinal cohort dataset (96,485 children, 588,546 measurements).
• Developed a prediction model using the light gradient boosting method, incorporating anthropometric metrics, body composition, SDSs, and velocity parameters.
• Assessed model performance using RMSE, MAE, and MAPE; employed SHAP for interpretability.

Main Results:

• The AI model demonstrated high accuracy in predicting future heights for both males and females (RMSE < 2.51 cm).
• Key predictors identified include height SDS, height velocity, and soft lean mass velocity.
• Generated personalized growth curves by estimating individual height trajectories and identifying critical variables.

Conclusions:

• The developed AI model provides accurate, personalized growth curves with explainable AI insights.
• This approach advances pediatric growth assessment and supports clinical decision-making for growth disorders.
• The model shows significant potential for early identification and management of growth abnormalities.