Height estimation in children and adolescents using body composition big data: Machine-learning and explainable...

Area of Science:

• Pediatric endocrinology
• Biostatistics
• Artificial Intelligence in Medicine

Background:

• Accurate height estimation is crucial for monitoring child and adolescent growth and identifying potential health issues.
• Traditional methods may not fully capture the complex interplay of factors influencing growth.
• Leveraging advanced AI and body composition data can enhance predictive accuracy and interpretability.

Purpose of the Study:

• To develop a precise and interpretable height estimation model for pediatric populations.
• To utilize body composition variables and explainable AI (XAI) techniques for height prediction.
• To identify key body composition indicators that influence height in children and adolescents.

Main Methods:

• A light gradient boosting machine learning model was trained on a large dataset (n=54,374) of children and adolescents (ages 6-18).
• The model incorporated anthropometric and detailed body composition measures.
• Explainable AI methods, including SHAP and PDP, were used to interpret model predictions.

Main Results:

• The height estimation model demonstrated high accuracy, with mean absolute percentage errors of 1.64% for boys and 1.63% for girls.
• Soft lean mass (SLM) and body fat mass percentage (BFMP) were identified as significant predictors of height.
• A positive association was observed between SLM and estimated height, whereas BFMP showed an inverse relationship.

Conclusions:

• Body composition variables are reliable predictors of height in pediatric populations.
• The developed AI model provides accurate and interpretable insights into height estimation.
• This approach holds promise for advancing pediatric growth assessment and monitoring tools.