Artificial Intelligence for Objective Assessment of Pediatric Uroflowmetry Curves

Area of Science:

• Pediatric Urology
• Medical Imaging Analysis
• Artificial Intelligence in Medicine

Background:

• Uroflowmetry curve interpretation is subjective and prone to inter-observer variability.
• Objective classification of uroflowmetry patterns is crucial for accurate diagnosis in pediatric voiding dysfunction.

Purpose of the Study:

• To evaluate the efficacy of AI and ML algorithms in objectively classifying uroflowmetry curves.
• To assess the potential of AI to reduce diagnostic variability and improve accuracy in pediatric uroflowmetry.

Main Methods:

• A dataset of 586 pediatric uroflowmetry curves (ages 5-17) was analyzed.
• Curves were standardized and classified by pediatric urology specialists.
• The YOLOv5×6 algorithm was trained and validated on the dataset, with performance metrics including accuracy, precision, recall, F1-score, and mAP.

Main Results:

• High inter-rater agreement (Fleiss' kappa: 0.948) was observed among specialists.
• The AI model achieved 85.8% overall accuracy, with 96% success in identifying bell-shaped curves.
• High precision (1.00) for plateau curves and a mean Average Precision (mAP@0.5) of ~90% were recorded.

Conclusions:

• AI-driven classification of uroflowmetry curves demonstrates high accuracy and reduces observer variability.
• Further research with multicenter datasets and standardized reporting is recommended for enhanced clinical integration.
• AI holds promise for real-time analysis in uroflowmetry devices.