Artificial intelligence (AI) and machine learning (ML) offer new tools to advance care in pediatric urology. While there has been interest in developing ML models in the field, there has not been a synthesis of the literature. Here, we aim to highlight the important work being done in bringing these advanced tools into pediatric urology and review their objectives, model performance, and usability.
We performed a comprehensive, non-systematic search on MEDLINE and EMBASE and combined these with hand-searches of publications which utilize ML to predict outcomes in pediatric urology. Each article was extracted for objectives, AI approach, data sources, model inputs and outputs, model performance, and usability. This information was qualitatively synthesized.
A total of 27 unique ML models were found in the literature. Vesicoureteral reflux, hydronephrosis, pyeloplasty, and posterior urethral valves were the primary topics. Most models highlight strong performance within institutional datasets and accurately predicted clinically relevant outcomes. Model validity was often limited without external validation, and usability was hampered by model deployment and interpretability.
Current ML models in pediatric urology are promising and have been applied to many major pediatric urology problems. These models still warrant further validation. However, with thoughtful implementation, they may be able to influence clinical practice in the near future.