Artificial Intelligence for Predicting Treatment Failure in Neurourology: From Automated Urodynamics to Precision Management

Article information

Int Neurourol J. 2025;29(Suppl 2):S55-S64

Publication date (electronic) : 2025 November 30

doi : https://doi.org/10.5213/inj.2550316.158

Seunghyun Youn ¹

, Beom Jin Park^,²

¹GRK Partners Research Center, Seoul, Korea

²Department of Radiology and Advanced Medical Imaging Institute, Korea University Anam Hospital, Korea University College of Medicine, Seoul, Korea

Corresponding author: Beom Jin Park Department of Radiology and Advanced Medical Imaging Institute, Korea University Anam Hospital, Korea University College of Medicine, 73 Goryeodae-ro, Seongbuk-gu, Seoul 02841, Korea Email: radiolbj226@gmail.com

Received 2025 September 2; Accepted 2025 October 29.

Abstract

Artificial intelligence (AI) has emerged as a transformative tool for advancing diagnosis, monitoring, and treatment planning in neurourology. This review synthesizes recent progress in AI-based models for predicting treatment failure in neurogenic lower urinary tract dysfunction. Machine learning and deep learning algorithms applied to urodynamic, clinical, and neuroimaging data have demonstrated strong potential to identify patients at risk of therapeutic nonresponse and improve individualized management. Automated systems now enable precise interpretation of complex bladder signals, multimodal data integration, and real-time prediction of treatment outcomes, marking a shift toward data-driven precision medicine. Nevertheless, most published studies remain limited by small, single-center datasets and a lack of external validation. Broader clinical adoption will require multicenter collaboration, adherence to standardized reporting frameworks such as TRIPOD-ML and PROBAST-AI, and integration of explainable AI to ensure transparency, reproducibility, and clinician trust.

Keywords: Artificial intelligence; Neurourology; Neurogenic bladder; Treatment failure; Predictive modeling; Precision medicine

INTRODUCTION

Neurourology encompasses disorders of the lower urinary tract arising from neurological diseases or injuries, including spinal cord injury (SCI), multiple sclerosis (MS), Parkinson disease, and stroke [1-3]. These conditions, collectively referred to as neurogenic lower urinary tract dysfunction, are characterized by detrusor overactivity, detrusor-sphincter dyssynergia, and impaired bladder compliance, which can lead to urinary retention, recurrent infections, and progressive upper urinary tract damage [4].

Despite various pharmacologic and interventional therapies, treatment failure rates remain high—approximately 40%–60% for oral antimuscarinics and 25%–35% for intradetrusor botulinum toxin injections [5, 6]. Such failures contribute to diminished quality of life, recurrent urinary tract infections, and even irreversible renal impairment [7].

Traditional therapeutic algorithms rely heavily on empirical, symptom-based decision-making that often does not reflect the heterogeneity of underlying neurophysiology [8]. Clinical judgment alone cannot adequately predict which patients will respond to therapy or experience relapse, underscoring the need for objective, data-driven approaches capable of capturing complex physiological patterns.

Machine learning (ML), a core branch of artificial intelligence (AI), provides a framework for identifying nonlinear interactions among diverse clinical and urodynamic variables [9-11]. As illustrated in Fig. 1, peer-reviewed publications addressing AI in neurourology have increased exponentially between 2015 and 2025, reflecting the rapid technological maturation of the field. This surge parallels broader trends in AI-assisted precision medicine and demonstrates the growing interest in leveraging predictive modeling for individualized treatment optimization.

Fig. 1.

Global trend in artificial intelligence-neurourology publications (2015–2025).

Recent advances in AI have also transformed diagnostic and therapeutic paradigms across urology. Techniques ranging from classical ML to deep learning (DL) and multimodal data integration are now used to identify disease patterns, predict treatment responses, and tailor interventions for complex urological disorders [12, 13]. In neurourology, applications have expanded from automated urodynamic interpretation and neuroimaging-based voiding analysis to AI-assisted diagnosis of interstitial cystitis and bladder pain syndromes, collectively signaling a shift toward precision medicine.

These developments highlight the transition from descriptive observation to predictive analytics, where multimodal datasets—encompassing clinical, behavioral, imaging, and molecular features —are synthesized to anticipate disease trajectory and treatment failure [12, 13].

This paper provides a comprehensive review of current evidence and methodological trends in AI-based models for predicting treatment failure in neurourology, emphasizing their clinical potential, limitations, and future directions toward precision-guided patient management.

ADVANCES IN AUTOMATED URODYNAMIC INTERPRETATION

Urodynamic testing generates multichannel time-series data, including vesical, abdominal, and detrusor pressures, as well as uroflowmetry and electromyography (EMG) signals. These data are highly dynamic and nonlinear, posing challenges for conventional rule-based or statistical analyses but offering rich physiological information that can be effectively captured by AI-based models. AI and DL algorithms enable automated identification of clinically relevant patterns such as detrusor overactivity, impaired compliance, and bladder outlet obstruction.

Data Sources and Signal Domains

Urodynamic signals represent the integrated function of detrusor contractility, outlet resistance, and neural control mechanisms. Hobbs et al. [14] utilized 805 urodynamic studies from 546 patients with spina bifida to train a support vector machine classifier for detecting detrusor overactivity, achieving an area under the receiver operating characteristic curve (AUC) of 0.919. This study demonstrated that AI could replicate, and in some cases surpass, manual expert interpretation in pediatric neurogenic bladder cohorts. Similarly, Choo et al. [15] developed an automatic interpretation algorithm for uroflowmetry using a convolutional neural network architecture, which achieved 90% accuracy in classifying normal, obstructed, and interrupted flow patterns (Table 1). Together, these studies highlight how standard clinical signals (pressure and flow curves) can serve as robust, high-dimensional input domains for AI-driven diagnostic automation.

Table 1.

Summary of AI diagnostic models in urodynamics

DL for Bladder Signal Recognition

Recent advances in DL have further extended AI’s role from static classification to real-time recognition of dynamic bladder events. Liu et al. [16] implemented a YOLOv5-based real-time recognition system trained on urodynamic traces from neurogenic bladder patients, achieving >95% accuracy in identifying detrusor contraction events and voiding phases. This model represented one of the first practical demonstrations of real-time urodynamic interpretation using DL architectures capable of temporal and spatial feature fusion. Similarly, Cho and Youn [17] developed an intravesical pressure-mapping algorithm utilizing a hybrid DL framework that interprets bladder function and predicts abnormal voiding patterns with an AUC of 0.93 (Table 1). Their approach introduced the novel concept of AI-assisted bladder functional mapping, bridging traditional pressure–volume analysis with data-driven interpretability. As illustrated in Fig. 2, these algorithms process urodynamic signals through a sequential AI pipeline involving data acquisition, preprocessing, feature extraction, model training, and clinical output generation.

Fig. 2.

Machine learning (ML) workflow for treatment failure prediction in neurourology. LASSO, least absolute shrinkage and selection operator.

Clinical Relevance

AI-based automated interpretation systems provide substantial clinical benefits by reducing operator dependency, standardizing evaluation criteria, and improving reproducibility across institutions. They also facilitate the extraction of quantitative urodynamic biomarkers —including pressure variability, compliance slope, contraction amplitude, and voiding time—that can subsequently inform predictive models of treatment response and failure. Ultimately, integrating these automated systems into electronic health record platforms could enable continuous, data-driven urodynamic monitoring, representing a key step toward fully digitalized and personalized neurourology. Collectively, these developments in automated urodynamic interpretation establish the technical foundation upon which AI-based predictive models for treatment failure can be built.

Summary of Recent AI Frameworks

Recent studies have extended automated urodynamic analysis into predictive modeling for treatment response and failure, applying multimodal and transformer-based DL frameworks [18, 19].

These approaches demonstrate progressive integration of urodynamic, EMG, and clinical features, marking a shift from diagnostic automation to prognostic precision. By capturing temporal patterns, multimodal relationships, and neural correlates, such models exemplify how AI is transforming urodynamic interpretation into a foundation for personalized therapy.

Collectively, these advances not only facilitate diagnostic standardization but also generate structured urodynamic biomarkers that serve as inputs for predictive modeling. As summarized in Table 2, recent frameworks have begun integrating urodynamic, EMG, and clinical variables to forecast therapeutic nonresponse, relapse, or device failure, effectively extending AI’s role from diagnostic automation to prognostic precision.

Table 2.

Representative machine learning models predicting treatment failure in neurourology

PREDICTIVE MODELING FOR TREATMENT FAILURE

The evolution of AI in neurourology has progressed from signal-level automation to outcome-level prediction. As summarized above, advances in automated urodynamic interpretation have enabled precise and reproducible quantification of bladder dynamics [14-19], transforming raw physiologic signals into structured, machine-readable biomarkers. These biomarkers — including detrusor pressure variability, compliance slope, voiding duration, and contraction amplitude —provide a rich substrate for prognostic modeling by capturing subtle dysfunctions that precede clinical deterioration [20, 21].

Building upon this technical foundation, recent studies have expanded AI applications from diagnostic automation to prognostic intelligence, emphasizing early identification of patients at risk for therapeutic nonresponse or relapse [20-27]. The integration of multimodal inputs — including urodynamic, electromyographic, clinical, and demographic data—has enabled ML models to map complex relationships among patient characteristics, treatment modalities, and outcomes [28-30]. Through such integration, AI systems can not only classify existing dysfunctions but also forecast treatment trajectories, bridging the gap between physiological monitoring and personalized therapy [31, 32].

Consequently, predictive modeling in neurourology represents the next phase of digital transformation —shifting focus from post hoc interpretation to anticipatory management. The following section synthesizes representative ML models that exemplify this paradigm, summarizing their algorithms, datasets, performance, and clinical implications (Table 2) [20-23].

Gradient Boosting for OAB Medication Failure

Başaranoğlu et al. analyzed 847 patients with overactive bladder (OAB) treated with anticholinergics or beta-3 agonists using a gradient boosting classifier, achieving an AUC of 0.91 (95% confidence interval [CI], 0.87–0.95) and 87% accuracy, significantly outperforming clinician predictions (AUC, 0.71; P < 0.001) [20]. The model incorporated 23 clinical and urodynamic features, identifying baseline voiding frequency, nocturia, maximum cystometric capacity, and detrusor overactivity as top predictors. However, as the model was validated only through 10-fold internal cross-validation within a single tertiary center, its external generalizability remains uncertain [28, 29].

LASSO Regression for Upper Urinary Tract Damage

Wang et al. [33] developed a least absolute shrinkage and selection operator (LASSO) regression–based nomogram predicting upper urinary tract damage in 301 neurogenic bladder patients, achieving a concordance index (C-index) of 0.80 (95% CI, 0.75–0.84) and 78% accuracy. From 31 candidate variables, 12 were retained as key predictors, including detrusor leak point pressure, bladder compliance, vesicoureteral reflux, and bladder management type. The nomogram is clinically interpretable and easily applicable at the bedside but has not yet undergone external validation.

Graph Neural Networks for Medication Response

Lai et al. implemented a graph neural network model (GNN) in 1,064 patients from the Lower Urinary Tract Dysfunction Research Network cohort, achieving an AUC of 0.76 (95% CI, 0.72–0.80) and 71% accuracy in classifying pharmacotherapy responders [22, 27]. The GNN architecture encoded relationships among lower urinary tract symptom variables, outperforming logistic regression (AUC, 0.72) but requiring advanced computational infrastructure and expertise. While promising, the modest performance improvement raises questions about its cost-effectiveness for clinical deployment.

Neuroimaging-Based Random Forest:

Karmonik et al. [23] employed random forest modeling on functional magnetic resonance imaging connectivity maps from 27 women with MS, achieving 86% accuracy in classifying voiding dysfunction. Connectivity between the pontine micturition center and the prefrontal cortex emerged as the most discriminative feature, suggesting that neuroimaging biomarkers can enhance prediction of neurogenic voiding dysfunction. However, the small sample size raises concerns about overfitting and limited external validation [28].

Most published ML studies in neurourology rely on internal validation (cross-validation or bootstrapping) rather than independent dataset testing [28, 29]. The 4 core studies reviewed here were all single-center and retrospective, with sample sizes ranging from 27 to 1,064, underscoring the high risk of overfitting and limited demographic diversity. Recent multicenter investigations have begun to address these limitations. As shown in Fig. 3, the relationship between dataset size and model performance indicates that smaller cohorts (e.g., Karmonik et al. [23]) tend to exhibit inflated AUC values, whereas larger datasets (e.g., Başaranoğlu et al. [20]) yield more generalizable results. Quantitatively, smaller cohorts (<100 cases) often report AUC values around 0.86–0.90 despite limited external validation, while larger multicenter datasets ( >500 cases) achieve slightly lower but more reproducible performance (mean AUC ≈ 0.83 ±0.06). Lee et al. [30] externally validated a dual-center ML model predicting multidrug-resistant urinary tract infections in SCI patients (AUC, 0.83), and Werneburg et al. [31] demonstrated external reproducibility of an OAB outcome prediction model (AUC, 0.89). Increasingly, adherence to model reporting standards such as TRIPOD-ML and PROBAST-AI is emphasized to ensure reproducibility and real-world reliability [32, 34-38].

Fig. 3.

Relationship between dataset size and model performance in presentative machine learning (ML) studies. OAB, overactive bladder; MS, multiple sclerosis; fMRI, functional magnetic resonance imaging; UUTD, upper urinary tract dysfunction; LURN, Lower Urinary Tract Dysfunction Research Network; AUC, area under the receiver operating characteristic curve.

ML models in neurourology integrate diverse inputs, including demographics, neurological diagnosis, comorbidities, symptom scores, urodynamic parameters, and neuroimaging features [21-24]. As shown in Fig. 4, key determinants of treatment response consistently include urgency frequency, bladder compliance, detrusor leak pressure, and cortical–pontine connectivity, reflecting multilayer mechanisms of lower urinary tract dysfunction that traditional analyses fail to capture. These representative models —gradient boosting, LASSO regression, graph neural network, and random forest —demonstrate how distinct modalities converge on overlapping predictors that drive both accuracy and interpretability [39].

Fig. 4.

Machine learning versus clinician judgment in overactive bladder treatment failure prediction. ROC, receiver operating characteristic; AUC, area under the ROC curve.

While ensemble models such as gradient boosting and random forest often achieve superior predictive accuracy, their “black box” nature limits transparency and clinician confidence [40]. Interpretable approaches —including LASSO regression and attention-weighted GNNs —offer more intuitive reasoning, while post hoc explainability tools such as SHAP feature maps (Fig. 5) enhance model transparency and clinical trust [21, 22, 41, 42]. Successful clinical adoption requires embedding these models into user-friendly decision-support interfaces codesigned with clinicians and refined through iterative feedback [34, 38, 43]. Ultimately, predictive modeling in neurourology exemplifies AI’s potential to transform empirical treatment decisions into transparent, patient-specific strategies, provided that interpretability, validation, and ethical oversight remain central [37, 38].

Fig. 5.

Visualizes the relative feature importance across machine learning models. OAB, overactive bladder; MS, multiple sclerosis; fMRI, functional magnetic resonance imaging; UUTD, upper urinary tract dysfunction; LURN, Lower Urinary Tract Dysfunction Research Network; PAG, periaqueductal gray; PMC, pontine micturition center; PFC, prefrontal cortex; SMA, supplementary motor area; ACC, anterior cingulate cortex.

CONCLUSION

ML has demonstrated significant promise in predicting treatment failure across neurourological disorders, achieving robust accuracy through diverse algorithmic frameworks such as gradient boosting [20, 25], LASSO regression [21, 26], and graph neural networks [22, 27]. These models outperform traditional clinician-based assessments by identifying complex, nonlinear interactions among clinical, urodynamic, and neuroimaging variables [12, 13, 23]. However, current studies are constrained by small, single-center datasets and limited external validation, which restrict their generalizability [28, 29].

As summarized in Table 3, smaller single-center studies often report higher apparent accuracy despite limited validation, whereas larger multicenter datasets yield slightly lower but more reproducible performance. This inverse relationship underscores the need to expand dataset scale and standardize validation protocols to enhance reliability and applicability.

Table 3.

Summary of dataset size and model performance across representative studies

Future research should focus on constructing large, multicenter datasets and incorporating multimodal features —including clinical, imaging, and electrophysiologic data —while adhering to standardized reporting frameworks such as TRIPOD-ML and PROBAST-AI [32, 38-40]. In parallel, close collaboration among clinicians, data scientists, and regulatory bodies will be essential to ensure transparency, interpretability, and ethical implementation in real-world clinical settings [35, 36, 41, 43].

Ultimately, AI-driven predictive modeling represents a paradigm shift in neurourology from empirical management to proactive, precision-guided care, empowering clinicians to anticipate treatment failure, tailor interventions, and optimize long-term outcomes for patients with neurogenic lower urinary tract dysfunction [9-11, 43].

Notes

Grant/Fund Support

This research was supported by a grant from the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (grant number: RS-2022-KH129263).

Conflict of Interest

No potential conflict of interest relevant to this article was reported.

ACKNOWLEDGEMENTS

We would like to thank the Advanced Medical Imaging Institute in the Department of Radiology, the Korea University Anam Hospital in the Republic of Korea, and researchers for providing software, datasets, and various forms of technical support.

AUTHOR CONTRIBUTION STATEMENT

· Conceptualization: SY

· Data curation: SY

· Formal analysis: SY

· Funding acquisition: BJP

· Methodology: SY

· Project administration: BJP

· Visualization: SY

· Writing - original draft: SY

· Writing - review & editing: BJP

References

1. Panicker JN, Fowler CJ, Kessler TM. Lower urinary tract dysfunction in the neurological patient: clinical assessment and management. Lancet Neurol 2015;14:720–32.

2. Groen J, Pannek J, Castro Diaz D, Del Popolo G, Gross T, Hamid R, et al. Summary of European Association of Urology (EAU) guidelines on neuro-urology. Eur Urol 2016;69:324–33.

3. Abrams P, Cardozo L, Khoury S, Wein A. Incontinence: 5th International Consultation on Incontinence Paris: Health Publication Ltd; 2013.

4. Esclarín De Ruz A, García Leoni E, Herruzo Cabrera R. Epidemiology and risk factors for urinary tract infection in patients with spinal cord injury. J Urol 2000;164:1285–9.

5. Chapple CR, Khullar V, Gabriel Z, Muston D, Bitoun CE, Weinstein D. The effects of antimuscarinic treatments in overactive bladder: an update of a systematic review and meta-analysis. Eur Urol 2008;54:543–62.

6. Deffontaines-Rufin S, Weil M, Verollet D, Peyrat L, Amarenco G. Botulinum toxin A for the treatment of neurogenic detrusor overactivity in multiple sclerosis patients. Int Braz J Urol 2011;37:642–8.

7. Cameron AP, Wallner LP, Tate DG, Sarma AV, Rodriguez GM, Clemens JQ. Bladder management after spinal cord injury in the United States 1972 to 2005. J Urol 2010;184:213–7.

8. Singh G, Mittal A, Sinha S, Panwar VK, Bhadoria AS, Mandal AK. Urodynamics in the evaluation of lower urinary tract symptoms in young adult men: a systematic review. Indian J Urol 2023;39:97–106.

9. Rajkomar A, Dean J, Kohane I. Machine learning in medicine. N Engl J Med 2019;380:1347–58.

10. Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med 2019;25:44–56.

11. Deo RC. Machine learning in medicine. Circulation 2015;132:1920–30.

12. Kim ES, Eun SJ, Youn S. The current state of artificial intelligence application in urology. Int Neurourol J 2023;27:227–33.

13. Huang HH, Cheng PY, Tsai CY. Exploring artificial intelligence in functional urology: a comprehensive review. Urol Sci 2025;36:2–10.

14. Hobbs KT, Choe N, Aksenov LI, Reyes L, Aquino W, Routh JC, et al. Machine learning for urodynamic detection of detrusor overactivity. Urology 2022;159:247–54.

15. Choo MS, Ryu HY, Lee S. Development of an automatic interpretation algorithm for uroflowmetry results: application of artificial intelligence. Int Neurourol J 2022;26:69–77.

16. Liu X, Zhong P, Chen D, Liao L. Real-time typical urodynamic signal recognition system using deep learning. Int Neurourol J 2025;29:40–7.

17. Cho Y, Youn S. Intravesical bladder treatment and deep learning applications to improve irritative voiding symptoms caused by interstitial cystitis: a literature review. Int Neurourol J 2023;27(Suppl 1):S13–20.

18. Janicki JJ, Zwaans BMM, Bartolone SN, Ward EP, Chancellor MB. Advancing interstitial cystitis/bladder pain syndrome (IC/BPS) diagnosis: a comparative analysis of machine learning methodologies. Diagnostics (Basel) 2024;14:2734.

19. Fan B, Zhang L, Cui H, Bai S, Gao H, Xiang S, et al. Development of a machine learning-based model to predict urethral recurrence following radical cystectomy: a multicentre retrospective study and updated meta-analysis. Sci Rep 2025;15:19573.

20. Başaranoğlu M, Taşdemir İK, Akbay E, Doruk HE. Artificial intelligence-based prediction of treatment failure and medication non-adherence in overactive bladder management. BMC Urol 2025;25:209.

21. Li Q, Cai M, Pu Q, Wu S, Liu X, Lin T, et al. A nomogram for predicting upper urinary tract damage risk in children with neurogenic bladder. Front Pediatr 2022;10:1050013.

22. Lai HH, Zhang H, Li F. PD14-03 Prediction of response to OAB medications using deep learning and artificial intelligence – a LURN research study. J Urol 2023;209(Supplement 4):e413.

23. Karmonik C, Boone T, Khavari R. Data-driven machine-learning quantifies differences in the voiding initiation network in neurogenic voiding dysfunction in women with multiple sclerosis. Int Neurourol J 2019;23:195–204.

24. Beam AL, Kohane IS. Big data and machine learning in health care. JAMA 2018;319:1317–8.

25. Friedman JH. Greedy function approximation: a gradient boosting machine. Ann Stat 2001;29:1189–232.

26. Tibshirani R. Regression shrinkage and selection via the Lasso. J R Stat Soc Series B Stat Methodol 1996;58:267–88.

27. Zhou J, Cui G, Hu S, Zhang Z, Yang C, Liu Z, et al. Graph neural networks: a review of methods and applications. AI Open 2020;1:57–81.

28. Steyerberg EW, Harrell FE. Prediction models need appropriate internal, internal-external, and external validation. J Clin Epidemiol 2016;69:245–7.

29. Arlot S, Celisse A. A survey of cross-validation procedures for model selection. Stat Surv 2010;4:40–79.

30. Lee SJ, Yoon H, Shin JC, Cho SR. Machine learning prediction of multidrug-resistant urinary tract infections in brain and spinal cord injury patients: a dual-center validation study. Int J Med Inform 2025;206:106143.

31. Werneburg GT, Werneburg EA, Goldman HB, Slopnick E, Roberts LH, Vasavada SP. External validation demonstrates machine learning models outperform human experts in prediction of objective and patient-reported overactive bladder treatment outcomes. Urology 2024;194:56–63.

32. Altman DG, Vergouwe Y, Royston P, Moons KG. Prognosis and prognostic research: validating a prognostic model. BMJ 2009;338:b605.

33. Wang W, Fang H, Xie P, Cao Q, He L, Cai W. Create a predictive model for neurogenic bladder patients: upper urinary tract damage predictive nomogram. Int J Neurosci 2019;129:1240–6.

34. Riley RD, Snell KI, Ensor J, Burke DL, Harrell FE, Moons KG, et al. Minimum sample size for developing a multivariable prediction model: Part II - binary and time-to-event outcomes. Stat Med 2019;38:1276–96.

35. Ramspek CL, Jager KJ, Dekker FW, Zoccali C, van Diepen M. External validation of prognostic models: what, why, how, when and where? Clin Kidney J 2021;14:49–58.

36. Kappen TH, van Klei WA, van Wolfswinkel L, Kalkman CJ, Vergouwe Y, Moons KGM. Evaluating the impact of prediction models: lessons learned, challenges, and recommendations. Diagn Progn Res 2018;2:11.

37. Collins GS, de Groot JA, Dutton S, Omar O, Shanyinde M, Tajar A, et al. External validation of multivariable prediction models: a systematic review of methodological conduct and reporting. BMC Med Res Methodol 2014;14:40.

38. U.S. Food and Drug Administration (FDA). Clinical Decision Support Software: guidance for Industry and Food and Drug Administration Staff Silver Spring (MD): U.S. FDA. September; 2022.

39. Lipton ZC. The mythos of model interpretability. Commun ACM 2018;61:36–43.

40. Rudin C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nat Mach Intell 2019;1:206–15.

41. Sutton RT, Pincock D, Baumgart DC, Sadowski DC, Fedorak RN, Kroeker KI. An overview of clinical decision support systems: benefits, risks, and strategies for success. NPJ Digit Med 2020;3:17.

42. Char DS, Shah NH, Magnus D. Implementing machine learning in health care - addressing ethical challenges. N Engl J Med 2018;378:981–3.

43. Bauer MS, Kirchner J. Implementation science: what is it and why should I care? Psychiatry Res 2020;283:112376.

Article information Continued

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Study	Year	Data type	Algorithm	Target	AUC/accuracy	Key findings
Hobbs et al. [14]	2022	UDS (805 files)	SVM	DO detection	0.919	Pediatric spina bifida cohort
Choo et al. [15]	2022	UFM	CNN	Pattern classification	0.90	Automatic voiding curve labeling
Liu et al. [16]	2025	UDS signals	YOLOv5	Real-time signal recognition	> 95%	Neurogenic pattern detection
Cho and Youn [17]	2023	Intravesical pressure mapping	DL hybrid	Abnormal voiding prediction	0.93	AI bladder model, clinical correlation

Study	Başaranoğlu et al. (2025) [20]	Li et al. (2022) [21]	Lai et al. (2023) [22]	Karmonik et al. (2019) [23]
Population	OAB patients (n = 847)	Children (≤ 18 years) with neurogenic bladder (NB) (n = 257)	LURN cohort (n = 1,064)	MS patients (n = 27)
Algorithm	Gradient boosting	nomogram	Graph Neural Network	Random forest (fMRI)
Outcome	Medication failure	Upper urinary tract damage	Medication response	Voiding dysfunction classification
Input features (n)	Demographics, symptoms, urodynamics (23)	Clinical + urodynamics	Patient-reported symptoms with relational encoding	fMRI connectivity (PMC, PAG, PFC, insula)
Primary performance	AUC 0.91 (95% CI, 0.87–0.95); accuracy 87%	AUC 0.83 (95% CI, 0.75–0.91)	AUC 0.76 (95% CI, 0.72–0.80); accuracy 71%	Accuracy 86%, sensitivity 88%, specificity 84%
Validation	10-Fold CV, single center	Internal bootstrapping, single center	Multicenter CV within LURN	Internal CV; single center
Key Notes	Outperformed clinicians (AUC 0.91 vs. 0.71, P < 0.001). Top predictors: voiding frequency, nocturia, max capacity, detrusor overactivity	Interpretable bedside tool with good calibration.	Modest improvement over logistic regression (AUC, 0.72). Symptom relationships contain predictive information but complexity trade-off	PMC-PFC connectivity most discriminative. Small sample raises overfitting concerns; external validation needed
Key Notes		Key predictors: leak point pressure, bladder compliance, reflux, management type (CIC)

Study	Sample size (N)	Algorithm	AUC/accuracy (%)	Group
Başaranoğlu et al. (2025) [20]	847	Gradient boosting	AUC 0.91 (95% CI, 0.87–0.95)	> 500 Cases
Li et al. (2022) [21]	257	Nomogram	AUC 0.83 (95% CI, 0.75–0.91)	100-500 Cases
Lai et al. (2023) [22]	1,064	Graph Neural Network	AUC 0.76 (95% CI, 0.72–0.80)	> 500 Cases
Karmonik et al. (2019) [23]	27	Random forest (fMRI)	Accuracy 86% (≈AUC, 0.86)	< 100 Cases