Artificial Intelligence-Based Analysis of Uroflowmetry Patterns in Children: A Machine Learning Perspective

Comprehensive Summary

This prospective study focused on evaluating how well machine learning (ML) models can interpret uroflowmetry (UF) patterns in children who present with lower urinary tract symptoms (LUTS). The researchers analyzed UF test results from a large sample of 500 children, ranging in age from 4 to 17 years old. To establish a comparison point, the tests were first interpreted by three pediatric urologists. Their assessments, however, showed only moderate agreement (Fleiss’ κ = 0.608), which highlights that even experienced specialists may vary in how they classify UF patterns. These same test results were then used to train and test five different ML models: Decision Tree, Random Forest, CatBoost, XGBoost, and LightGBM. The findings also showed that the ML models achieved strong performance overall, with classification accuracies ranging from 81.8% to 85.0%. Among the five, XGBoost stood out as the best-performing model. The study also revealed that some patterns were easier for the models to identify than others. For example, interrupted voiding patterns were classified with the highest level of accuracy, while tower-shaped and plateau patterns posed more of a challenge. Together, the results suggest that ML can not only match but in some cases outperform human consistency when it comes to UF interpretation, especially in situations where variability between experts might lead to uncertainty.

Outcomes and Implications

The study is the first to apply ML to UF interpretation in children, demonstrating that AI can reduce interobserver variability and standardize assessment of voiding patterns. With further refinement and validation, AI-assisted tools could enhance diagnostic accuracy, improve efficiency, and support clinical decision-making in pediatric urology, particularly for patients with LUTS.