Artificial intelligence for urodynamic studies: systematic review of methods, performance, and clinical applications.

BACKGROUND AND

objectiveInterpretation of urodynamic studies (UDS) is essential for the assessment of lower urinary tract dysfunction but remains resource-intensive and highly operator-dependent. Artificial intelligence (AI) has increasingly been applied to support urodynamic interpretation; however, the available evidence is heterogeneous and its clinical relevance remains uncertain. This systematic review aimed to synthesise current evidence on AI applications in urodynamic assessment and to summarise diagnostic performance and methodological quality.

methodsA systematic literature search of PubMed, Embase, Scopus, and IEEE Xplore was conducted in accordance with PRISMA 2020. Studies applying AI to uroflowmetry, cystometry, or pressure-flow studies in human participants were included. Eligible studies were categorised according to data representation as parameter-based, raw signal-based, or image-based models. Diagnostic performance was summarised narratively, and risk of bias was assessed using the QUADAS-AI framework. KEY FINDINGS AND LIMITATIONS: Eighteen studies met the inclusion criteria. Parameter-based models generally demonstrated modest to good diagnostic performance, particularly for binary classification tasks. Raw signal-based models more frequently reported high discriminative performance, with several studies achieving area under the receiver operating characteristic curve values above 0.85 and, in some cases, exceeding 0.90. Image-based models demonstrated feasibility for automated interpretation of urodynamic traces, although reported performance varied substantially. Most studies were retrospective and single-centre, relied on expert-defined reference standards, and lacked external validation, resulting in frequent high or unclear risk of bias. CONCLUSIONS AND CLINICAL IMPLICATIONS: Artificial intelligence has been widely applied to urodynamic assessment and often demonstrates high diagnostic performance in research settings. However, substantial methodological limitations limit confidence in clinical generalisability. At present, AI-assisted urodynamic interpretation should be considered exploratory, and robust prospective studies with independent external validation are required before routine clinical implementation.