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

Area of Science:

• Pediatric Urology
• Medical Informatics
• Machine Learning in Medicine

Background:

• Uroflowmetry (UF) is a key noninvasive test for evaluating pediatric lower urinary tract symptoms (LUTS).
• Expert interpretation of UF voiding patterns shows significant inter-observer variability.
• Machine learning (ML) offers a potential solution for standardizing UF analysis.

Purpose of the Study:

• To assess the accuracy of ML models in interpreting pediatric uroflowmetry voiding patterns.
• To compare the performance of different ML algorithms for UF pattern classification.

Main Methods:

• 500 pediatric uroflowmetry tests from children (4-17 years) with LUTS were analyzed.
• Voiding patterns were initially interpreted by three pediatric urology experts, with consensus reached on discrepancies.
• Five ML models (Decision Tree, Random Forest, CatBoost, XGBoost, LightGBM) were trained on 80% of the data and tested on 20%.

Main Results:

• Initial expert agreement on UF patterns was moderate (Fleiss' κ = 0.608), with 37.8% of tests showing discrepancies.
• The XGBoost model achieved the highest accuracy (85.00% ± 2.90%) in classifying voiding patterns.
• Accuracy varied by pattern, with interrupted patterns showing high accuracy (95%-100%) and tower/plateau patterns showing lower accuracy (61.54%-73.08%).

Conclusions:

• ML models demonstrate acceptable accuracy in interpreting pediatric uroflowmetry patterns.
• Artificial intelligence holds potential for standardizing uroflowmetry voiding pattern analysis in pediatric urology.
• Further research may lead to AI-assisted diagnostic tools for LUTS evaluation.