TY - JOUR
T1 - A living scoping review and online repository of artificial intelligence models in pediatric urology
T2 - Results from the AI-PEDURO collaborative
AU - Khondker, Adree
AU - Kwong, Jethro CC
AU - Ahmad, Ihtisham
AU - Rajesh, Zwetlana
AU - Dhalla, Rahim
AU - MacNevin, Wyatt
AU - Rickard, Mandy
AU - Erdman, Lauren
AU - Gabrielson, Andrew T.
AU - Nguyen, David Dan
AU - Kim, Jin Kyu
AU - Abbas, Tariq
AU - Fernandez, Nicolas
AU - Fischer, Katherine
AU - t Hoen, Lisette A.
AU - Keefe, Daniel T.
AU - Nelson, Caleb P.
AU - Viteri, Bernarda
AU - Wang, Hsin Hsiao (Scott)
AU - Weaver, John
AU - Yadav, Priyank
AU - Lorenzo, Armando J.
N1 - Publisher Copyright:
© 2025 The Author(s)
PY - 2025/2/5
Y1 - 2025/2/5
N2 - Introduction: Artificial intelligence (AI) is increasingly being applied across pediatric urology. We provide a living scoping review and online repository developed by the AI in PEDiatric UROlogy (AI-PEDURO) collaborative that summarizes the current and emerging evidence on the AI models developed in pediatric urology. Material and methods: The protocol was published a priori, and Preferred Reporting Items for Systematic Review and Meta-analysis Scoping Review (PRISMA-ScR) guidelines were followed. We conducted a comprehensive search of four electronic databases and reviewed relevant data sources from inception until June 2024 to identify studies that have implemented AI for prediction, classification, or risk stratification for pediatric urology conditions. Model quality was assessed by the APPRAISE-AI tool. Results: Overall, 59 studies were included in this review from 1557 unique records. Of the 59 published studies, 44 studies (75 %) were published after 2019, with hydronephrosis and vesicoureteral reflux/urinary tract infection as the most common topics (17 studies, 28 % each). Studies originated from USA (22 studies, 37 %), Canada (10 studies, 17 %), China (8 studies, 14 %), and Turkey (7 studies, 12 %). Neural network (35 studies, 59 %), support-vector-machine (21 studies, 36 %), and tree-based models (19 studies, 32 %) were the most used machine learning algorithms, with 14 studies (24 %) providing useable repositories or applications. APPRAISE-AI assessed 12 studies (20 %) of studies as low quality, 39 studies (66 %) as moderate quality, and 8 studies (14 %) as high quality, with specific improvements noted in model robustness and reporting standards over time (p = 0.03). Findings were synthesized into an online repository (www.aipeduro.com). Discussion: There is an increasing pace of AI model development in pediatric urology. Model topics are broad, algorithm choice is diverse, and the overall quality of models are improving over time. While there is still a lack of clinical translation of the AI models in pediatric urology, the usage of online repositories and reporting frameworks can facilitate sharing, improvement, and clinical implementation of future models.Conclusions: This living scoping review and online repository will highlight the current landscape of AI models in pediatric urology and facilitate their clinical translation and inform future research initiatives. From this work, we provide a summary of recommendations based on the current literature for future studies.[Figure
AB - Introduction: Artificial intelligence (AI) is increasingly being applied across pediatric urology. We provide a living scoping review and online repository developed by the AI in PEDiatric UROlogy (AI-PEDURO) collaborative that summarizes the current and emerging evidence on the AI models developed in pediatric urology. Material and methods: The protocol was published a priori, and Preferred Reporting Items for Systematic Review and Meta-analysis Scoping Review (PRISMA-ScR) guidelines were followed. We conducted a comprehensive search of four electronic databases and reviewed relevant data sources from inception until June 2024 to identify studies that have implemented AI for prediction, classification, or risk stratification for pediatric urology conditions. Model quality was assessed by the APPRAISE-AI tool. Results: Overall, 59 studies were included in this review from 1557 unique records. Of the 59 published studies, 44 studies (75 %) were published after 2019, with hydronephrosis and vesicoureteral reflux/urinary tract infection as the most common topics (17 studies, 28 % each). Studies originated from USA (22 studies, 37 %), Canada (10 studies, 17 %), China (8 studies, 14 %), and Turkey (7 studies, 12 %). Neural network (35 studies, 59 %), support-vector-machine (21 studies, 36 %), and tree-based models (19 studies, 32 %) were the most used machine learning algorithms, with 14 studies (24 %) providing useable repositories or applications. APPRAISE-AI assessed 12 studies (20 %) of studies as low quality, 39 studies (66 %) as moderate quality, and 8 studies (14 %) as high quality, with specific improvements noted in model robustness and reporting standards over time (p = 0.03). Findings were synthesized into an online repository (www.aipeduro.com). Discussion: There is an increasing pace of AI model development in pediatric urology. Model topics are broad, algorithm choice is diverse, and the overall quality of models are improving over time. While there is still a lack of clinical translation of the AI models in pediatric urology, the usage of online repositories and reporting frameworks can facilitate sharing, improvement, and clinical implementation of future models.Conclusions: This living scoping review and online repository will highlight the current landscape of AI models in pediatric urology and facilitate their clinical translation and inform future research initiatives. From this work, we provide a summary of recommendations based on the current literature for future studies.[Figure
UR - http://www.scopus.com/inward/record.url?scp=85217809071&partnerID=8YFLogxK
U2 - 10.1016/j.jpurol.2025.01.035
DO - 10.1016/j.jpurol.2025.01.035
M3 - Review article
C2 - 39956703
AN - SCOPUS:85217809071
SN - 1477-5131
JO - Journal of Pediatric Urology
JF - Journal of Pediatric Urology
ER -