Urodynamic studies (UDS) are an integral diagnostic test in urology. The use of artificial intelligence (Al) in healthcare has shown great potential with a growing body of evidence of superiority in the fields of radiology and pathology. This study aims to compare UDS traces interpretation using ChatGPT to expert human interpretation. In addition, treatment plan formulation by ChatGPT will be compared to experts.

We conducted a comparison between ChatGPT analysis and expert urologist assessments of UDS traces. Inclusion criteria were patients aged 18-85 who underwent UDS between March and September 2024. UDS traces were selected based on the clarity and completeness rather than specific patient demographics, age, gender, or pre-existing diagnoses. Each case included patient basic demographic details (age and gender) and chief complaint to provide clinical context. The selected UDS trace components as well as nomograms were analyzed independently by both ChatGPT and expert urologists. A diagnosis and treatment plan were then formulated by ChatGPT systematically and compared to those of the experts. The concordance of each parameter was evaluated and analyzed. Institutional board review approval was obtained.

We analyzed 100 traces majority of which were females (75%), with a mean age of 49. Mixed urinary incontinence was the most common presenting complaint. ChatGPT interpretations of urodynamic studies were compared to expert urologists, assessments for each parameter, diagnosis, and management decision using a binary concordance system (0 = no match, 1 = match). During the filling phase, ChatGPT demonstrated high concordance rates with expert assessments: bladder capacity (96%), compliance (90%), leakage (88%), sensation (87%), EMG readings (85%), and uninhibited detrusor contractions (80%). In the voiding phase, high concordance was calculated for voided volume (93%), Qmax (80%), while moderate concordance was observed for Pdet at Qmax (66%), after-contractions (62%), and bladder contractility index (60%). ChatGPT's diagnostic conclusions matched expert opinions in 77% of cases, and agreement on clinical management plans was achieved in 76%. The overall concordance rate across all evaluated parameters, diagnosis, and management recommendations was 80%.

ChatGPT effectively interpreted key urodynamic parameters, performing well in assessing bladder capacity, compliance, leakage, and voided volume, closely matching expert urologists. However, it showed limitations in interpreting more advanced metrics, including detrusor pressures, bladder contractility index, and after-contractions, leading to lower reliability in complex assessments. Similarly, ChatGPT accuracy in formulating diagnoses and clinical management plans was generally good but less reliable in nuanced or complicated scenarios, highlighting clear areas requiring further improvement before broader clinical integration.

This study demonstrates that ChatGPT showed high concordance rates with expert interpretation of UDS. Similarly, the suggested treatment plan was highly concordant with human expert plans. This exploratory study paves the way for further studies to unlock the potential applications of Al-powered software in the interpretation of UDS.

Zhou, Q., Li, G., Cui, K., Mao, W., Lin, D., Yang, Z., Chen, Z., Hu, Y., & Zhang, X. (2024). Using machine learning to construct the diagnosis model of female bladder outlet obstruction based on urodynamic study data. Investigative and clinical urology, 65(6), 559–566. https://doi.org/10.4111/icu.20240111
Zhou, Q., Chen, Z., Wu, B., Lin, D., Hu, Y., Zhang, X., & Liu, J. (2023). A Pilot Study: Detrusor Overactivity Diagnosis Method Based on Deep Learning. Urology, 179, 188–195. https://doi.org/10.1016/j.urology.2023.04.030
Bentellis, I., Guérin, S., Khene, Z. E., Khavari, R., & Peyronnet, B. (2021). Artificial intelligence in functional urology: how it may shape the future. Current opinion in urology, 31(4), 385–390. https://doi.org/10.1097/MOU.0000000000000888