Skip to main content
Wiley Open Access Collection logoLink to Wiley Open Access Collection
. 2025 Oct 1;137(1):112–119. doi: 10.1111/bju.70011

Can unsupervised machine learning gain new insights into urodynamic pressure flow pattern analysis?

Wouter van Dort 1,✉, Peter FWM Rosier 1, Thomas RF van Steenbergen 1, Bernard J Geurts 2, Laetitia MO de Kort 1
PMCID: PMC12690339  PMID: 41031874

Abstract

Objectives

To explore the use of unsupervised machine learning (UML) to analyse segments of the pressure flow study (PFS) curve after maximum flow, and subsequently to analyse the urodynamic and patient characteristics of men in the detected clusters.

Subjects and Methods

In this study, we considered 1650 PFSs of men with lower urinary tract symptoms, without relevant interventions in the past. After datapoint reduction and normalisation of the PFS curve segments, the k‐Shape clustering algorithm was used to identify different pattern clusters. Differences in patient and urodynamic characteristics among those clusters were explored.

Results

The UML approach identified four prominent clusters, with significantly different patient and urodynamic characteristics. Two pairs of these clusters were visually similar, and included similar urethral resistance values; however, they differed with regard to detrusor voiding contraction (DVC) and prostate size. In two clusters, the PFS curve pattern was significantly different from the commonly assumed ‘normal’ urethral resistance pattern in elderly men.

Conclusion

In males, PFS patterns are considered to be uniform in shape. However, this study shows that UML can help to identify clusters of pressure–flow urethral resistance subtype patterns in men. We found that these subtype patterns were associated with DVC strength and prostate size. This feasibility study has shown that UML clustering of urodynamic PFSs in men holds promise for improving the diagnosis of urethral resistance and DVC properties and dynamics.

Keywords: urodynamics, artificial intelligence, machine learning, pressure flow study, male LUTS, bladder outflow obstruction


Abbreviations

DCI

detrusor contractility index

ML

machine learning

pdet

detrusor pressure

PFS

pressure flow study

Qmax

maximum urinary flow rate

UDS

urodynamic study

UML

unsupervised machine learning

Introduction

Machine learning (ML) is a mathematical tool increasingly used for advanced analysis in a wide range of fields. ML enables computers to learn from large amounts of data and improve their performance in tasks without being specifically programmed. In healthcare, ML is being applied with the aim of improving diagnosis and treatment [1].

Urodynamic study (UDS) is a method used to analyse lower urinary tract function in patients with LUTS. During UDS, data are collected, displayed and identified in terms of visually recognisable patterns. ML algorithms could be used to enhance the consistency of the pattern analysis and diagnosis and to help relate variations in such patterns to the patient's pathophysiology [2]. Greater detail, more objective diagnosis and better stratification of patients’ urinary dysfunction could improve patient outcomes. The current application of ML in UDS is mainly focused on additional measurement techniques [3], extending the usability of non‐invasive UDS [4], or on automated analysis of the urodynamic traces [5]. Unravelling the pathophysiology or defining sub‐classes of dysfunction is an area that is still largely unexplored. Unsupervised ML (UML) holds great promise as it can detect and cluster previously unknown relationships, thus, it becomes a tool for the exploration of undescribed patterns that are not easily identified by humans [6]. The patterns identified by UML potentially contain clinically valuable information. In contrast to supervised ML, UML does not need labelled data, thereby eliminating the inconsistency that can be introduced by human expert labelling. This comes with a trade‐off, as the performance of the algorithm cannot be described in terms of compliance with a ‘gold standard’. However, UML performance in terms of clinical relevance can be resolved separately, using investigational or clinical outcomes in the subsequent phases of the research.

Current measures of urethral resistance included in urodynamic pressure–flow studies (PFSs) are all based on a simplified representation of the relationship between pressure and urinary flow after maximum urinary flow (Qmax) [7, 8, 9, 10]. This results in a one‐dimensional measure of BOO that neglects the actual pressure–flow relationship that is present during the course of voiding. The currently recommended sub‐classification of BOO is based on this simplified representation [9, 11]. UML can help to identify more details from the PFS curves, while including time dependency, thereby extending the current measures of BOO.

In this study, we explored the application of UML to the segments of the PFS curve after Qmax is reached, and analysed the urodynamic and patient characteristics of men in the clusters detected by this UML. Through this UML analysis, we aimed to explore the potential relevance of UML in fundamental urodynamics and to explore novel insights into the existence and clinical relevance of pattern variations in the PFS curves.

Subjects and Methods

Dataset Description

All 5657 PFSs performed in men between 2002 and 2020 were initially included. PFSs from patients with a history of bladder, prostate or urethra surgical interventions and those with neurogenic lower urinary tract dysfunction or urethral strictures were excluded (n = 3559, 62.9%). In addition, PFSs showing urodynamic values presumed to be non‐physiological (Qmax <2 mL/s or >35 mL/s, detrusor pressure [pdet] >200 cmH2O and voided volume <100 mL) or that included substantial artefacts were also excluded (n = 448, 7.9%). Substantial artefact presence was assessed by a computer algorithm, as described in an earlier publication [12]. This algorithm assesses the variations in both intravesical and abdominal pressure, and rejects measurements when those are outside specified cut‐off limits, details of which can be found in the supplementary materials of the cited publication. This resulted in a final dataset of 1650 high‐quality PFSs from 1374 unique patients. All PFSs were performed in accordance with the International Continence Society ‐ Society of Urodynamics, Female Pelvic Medicine & Urogenital Reconstruction Good Urodynamic practices [13, 14]. The investigations were performed using an 8‐F urodynamic double‐lumen catheter on a water‐filled urodynamic Ellipse system with AUDACT software (Andromeda Medizinische Systeme GmbH).

Data Preparation

Detrusor pressure and urinary flow rate (Q) were filtered with a 2‐s moving average filter [8], and a correction for the flow measurement delay of 0.75 s was applied [15]. Normally, Pdet and Q are measured as dependent variables against time as independent variable. As the pressure–flow relationship was of interest, a conversion from time as an independent variable to Q as an independent variable was made, resulting in a direct relation between Pdet and Q. The resolution of this new flow axis was a 30‐point, equally spaced scale, starting from Qmax to 0. This was the optimal scale, balancing the availability of a sufficient number of datapoints available for accurate ML, while preventing the need for up‐sampling in the case of shorter voids. Furthermore, use of a larger point scale would also result in a bigger model that would not be expected to provide more insightful results.

For each flow axis point, the average corresponding pdet around that point was calculated based on the original time‐based connection of pdet and Q. In addition, both pressure and flow were normalised to a 0–1 scale, where 1 referred to pdet.max or Qmax and 0 to pdet.min or Q = 0. In this way, the UML was focused on the shape of the curve and the differences between the values, instead of the actual measured values.

Unsupervised Machine Learning

The k‐Shape clustering algorithm was used, which preserves the shape of the given data series and produces homogeneous and well‐separated clusters [16]. This algorithm was found to be superior to other well‐known methods for time series analysis with regard to both accuracy and efficiency [16]. In this case, the pressure–flow relationship was studied. For this UML, this normally used time base was changed to a flow base, with pressure as the dependent variable, resulting in a ‘flow series’. The number of clusters was determined by a modified silhouette score, which gives a score between −1 and 1 based on the similarity within a cluster compared to the difference with other clusters. The standard silhouette score could not be used as it is based on Euclidian distances, while normalised cross‐correlation was used in the k‐Shape algorithm. By plotting the silhouette scores for a variety of cluster numbers, we decided that four clusters was the best balance between not having too many clusters and still being able to perform a potentially meaningful classification.

During UML, the algorithm optimises itself to minimise the differences between the graphs and the proposed centre within a cluster, which is called inertia, while maximising the differences between the clusters [11]. As the centre of the clusters is initiated randomly, the UML was run 1000 times to find the best‐performing initiation. The initiation with the lowest inertia was reported.

Analysis

To investigate the potential relevance of the observed clusters, patient characteristics (age and prostate size) and PFS characteristics (urethral resistance, detrusor voiding contraction [DVC], voided volume and post‐void residual urine volume) were compared among the clusters and statistically tested using the non‐parametric Kruskal–Wallis test.

To investigate differences in clinically relevant distribution of the described characteristics, several characteristics were grouped into clinically relevant clusters. For urethral resistance, the BOO index was used, with categories defined as follows: no BOO (BOO index < 20); equivocal BOO (20 ≤ BOO index < 40); and BOO (BOO index ≥ 40) [17]. The detrusor contractility index (DCI) was used to define the following detrusor contractility categories: weak (DCI < 100); normal (100 ≤ DCI < 150); and strong (DCI ≥ 150) [14]. Prostate size was categorised as small (<30 cm3), medium (30–80 cm3) and large (>80 cm3) [18]. Differences in distribution among the groups were investigated using the N‐1 chi‐squared test.

Data selection and preparation were performed using Matlab R2022b (The Mathworks Inc.). ML was performed using Python 3.9.13, including the tslearn 0.6.3 package, and statistical analysis was performed in SPSS, version 27 (IBM).

Results

In total, 1650 PFSs of men (aged 17–93 years, mean age 59 years) referred with LUTS were included in this study. Based on the silhouette score, UML indicated that four clusters provided an optimal representation. Subsequently, the clinical analysis was performed based on these four clusters.

The UML results are shown in Fig. 1. The largest cluster (Cluster 1 in Fig. 1) consisted of 834 PFSs (51% of total), starting with higher pressure/higher flow, steadily declining to lower pressure/zero flow. Cluster 2 consisted of 614 PFSs (37%), which showed a slight initial increase in pressure after Qmax but ended at low pressure/low flow. Cluster 3 with 105 PFSs (6%) showed an initial decrease in pressure, which increased later during the voiding and decreased again at the end of voiding. Cluster 4 consisted of 97 PFSs (6%) and was similar to the third cluster, with the main difference being a maximum pressure more concentrated at the end of voiding.

Fig. 1.

Fig. 1

Results of the unsupervised machine learning algorithm applied to the pressure–flow relationship after maximum urinary flow (Qmax). The graphs are oriented according to the latest pressure flow study (PFS) standard [19]. Cluster 1 consisted of 834 PFSs (51%), Cluster 2 of 614 PFSs (37%), Cluster 3 of 105 PFSs (6%) and Cluster 4 of 97 PFSs (6%). All axes show normalised values with 1 representing the maximum detrusor pressure (pdet) or flow rate and 0 the minimum pdet or flow of the pressure–flow curve after Qmax. Each graph can be read from the upper right corner to the lower left corner to relate to the course of micturition from Qmax to end of flow. The red line represents the centre of the cluster, while all the grey lines represent the individual traces.

A significant difference in patient and urodynamic characteristics among the clusters was found for almost all assessed variables (Table 1). No difference was found for PFS‐voided volume and PFS post‐void residual urine volume (Kruskal–Wallis P > 0.05). In association with the higher grades of BOO, patient age was slightly higher in Clusters 1 and 2. Clusters 1 and 2 showed similar values for urethral resistance, a result which was also observed when comparing Clusters 3 and 4. Differences between Clusters 1 and 2 or Clusters 3 and 4 were expressed in terms of differences in DCI, describing the DVC. Prostate size was nonsignificantly different when comparing all clusters (Kruskal–Wallis P = 0.086), but was significantly different when comparing combined Clusters 1 and 2 with Clusters 3 and 4 (Mann–Whitney U‐test P = 0.022).

Table 1.

Results of the unsupervised machine‐learning clustering of the urethral resistance relationship after the maximum urinary flow has been reached.

Feature Cluster 1 Cluster 2 Cluster 3 Cluster 4 Kruskal–Wallis test
N Mean (SD) N Mean (SD) N Mean (SD) N Mean (SD) P
Age, years 834 59 (14) 614 60 (14) 105 58 (15) 97 52 (15) <0.001
Prostate size, cm3 395 44 (27) 272 44 (28) 39 36 (22) 41 38 (17) 0.086
PFS voided volume, mL 834 310 (130) 614 310 (150) 105 320 (140) 97 320 (140) 0.484
PFS post‐void residual urine volume, mL 399 70 (130) 246 70 (110) 49 50 (80) 40 60 (100) 0.465
PFS‐void, % 399 84 (22) 246 83 (22) 49 88 (18) 40 86 (21) 0.434
BOO index 834 42 (30) 614 41 (29) 105 24 (26) 97 24 (26) <0.001
Urethral resistance A 834 32 (15) 614 33 (16) 105 24 (12) 97 23 (12) <0.001
linPURR 834 2.5 (1.4) 614 2.5 (1.4) 105 1.7 (1.2) 97 1.7 (1.2) <0.001
PFS‐Qmax, mL/s 834 10.1 (5.0) 614 9.1 (4.6) 105 12.2 (5.7) 97 13.1 (5.8) <0.001
pdetQmax, cmH2O 834 62 (25) 614 59 (26) 105 48 (22) 97 50 (21) <0.001
pmuo, cmH2O 834 37 (18) 614 39 (20) 105 49 (26) 97 56 (26) <0.001
DCI 834 112 (30) 614 105 (32) 105 109 (33) 97 116 (33) <0.001
WFmax 834 10.3 (3.9) 614 9.4 (3.7) 105 9.8 (4.4) 97 10.6 (4.2) <0.001
WFQmax 834 9.1 (3.5) 614 8.4 (3.4) 105 7.7 (3.3) 97 8.4 (3.2) <0.001

BOO index, bladder outflow obstruction index; DCI, detrusor contractility index; linPURR, linearised passive urethral resistance relation; pdetQmax, detrusor pressure at maximum flow; PFS, pressure flow study; pmuo, minimal urethral opening pressure; Qmax, maximum urinary flow rate; WF, Watts factor.

Significant differences were found among the clusters in the clinically relevant distribution of urethral resistance, bladder voiding contraction and prostate size (Fig. 2). Clusters 3 and 4 included significantly more men without BOO compared to Clusters 1 and 2 (N‐1 chi‐squared test P < 0.001). The number of men with weak DVC was significantly lower in Cluster 1 compared to Cluster 2 (P < 0.001) and also in Cluster 4 compared to Cluster 3 (P = 0.008). Cluster 3 included significantly more men with a small prostate compared to all other groups (P < 0.026). Interestingly, those men had a similar grade of obstruction to the men in Cluster 4, and were also of comparable age with the men in Clusters 1 and 2.

Fig. 2.

Fig. 2

Distribution of (A) urethral resistance, (B) detrusor voiding contraction (DVC) and (C) prostate size in the four clusters. For urethral resistance, BOO index (BOOI) was divided into no BOO (BOOI < 20), equivocal BOO (20 ≤ BOOI < 40) and BOO (BOOI ≥ 40). Detrusor contractility index (DCI) was divided in weak (DCI < 100), normal (100 ≤ DCI < 150) and strong (DCI ≥150). Prostate size was divided into small (<30 cm3), medium (30–80 cm3) and large (>80 cm3).

Discussion

We showed that the application of UML in the field of urodynamics can be used successfully to extract potentially clinically meaningful clusters segregating the PFSs of a large male patient population. UML resulted in distinguishable clusters, which differed visually as well as in terms of urodynamic features and patient characteristics. We are aware that the associations cannot, in this phase of our studies, imply causation; however, we can interpret these observed differences as a sign that previously rarely assessed aspects of the urodynamic graph may hold additional diagnostic information, providing a way of improving the assessment of both urethral resistance and detrusor voiding contractility.

We limit our discussion here mainly to the adopted techniques and to the possibilities, limitations and challenges associated with the use of UML for UDS data. An important step that must be taken before application of UML to the collected time series is data preparation [20]. Although some steps are included in common research practice, for example, source data cleaning for technical quality, or artefact and outlier removal/investigation, steps more specific to ML also need to be considered, such as normalisation and scale reduction. When focusing on curve shapes instead of absolute values, normalisation is an indispensable step in achieving reliable UML results [15]. The outcome of an algorithm based on normalised data should however always be combined with non‐normalised data if used for clinical purposes, as the absolute values contain clinically significant information.

Often, ML is performed on either time‐based data or on images. In this study, UML was applied using flow rate as an independent variable, as would be the case in clinical PFS analysis. Although the k‐Shape UML method used is normally applied to time‐based signals, there is no reason to suppose that the method will not work in a non‐time‐based environment as the underlying metric (normalised cross‐correlation) is generally applicable. This conversion from time to flow rate as an independent variable also solves the problem of the individual traces being unequal in length. A 30‐point scale was chosen as a ‘sweet spot’, with enough accuracy for performing UML while preventing the need for up‐sampling in the case of shorter micturition. This also has the side effect of reducing the number of datapoints, thereby lowering the amount of computational resources needed for ML. Although some information is lost in this step, the lower resolution would not be expected to affect the results, as the differences in original and lower resolution fall well within the expected test–retest variation [21]. An important challenge in the validation of UML is the two‐stage approach that is required to translate the results into a clinically meaningful measure. Supervised ML provides an ‘immediate’ measure of the accuracy of the method by comparing the outcome to a (gold) standard. This suggests that inferences can be made about clinical relevance. This downside of UML, lacking immediate validation, should however be considered in the context of urodynamics, which is the gold standard by itself. Studies considering UDS frequently include expert opinion clinical diagnosis, which is subjective by nature. In UML, a reference standard is neither required nor provided, thereby removing the need for a clinically labelled database obtained by consensus of multiple experts. Performance of UML is rather expressed in terms of initiating qualitative understanding of generic classes in a database. This makes UML a fundamental research tool for explorative studies of concepts that are testable in second study phases, which is very useful for (UDS)‐PFS scientific studies.

The UML performed on the pressure–flow relationship after Qmax showed that 51% of the curves follow the theoretical strict monotonic relationship between Q and pdet, [7, 8] depicted in the mean of Cluster 1 in Fig. 1. Combining Clusters 1 and 2, which include 88% of the PFSs, shows a relationship as described by further mathematical explorations regarding the urethral resistance dynamics after Qmax as an extension of the strict monotonic relationship [22, 23]. Cluster 2, showing a deviation from the expected ‘ideal’ pressure–flow relationship, includes significantly more PFSs with a weak DVC while showing similar urethral resistance. This further refinement made possible by UML suggests that this deviation from the ‘ideal’ relationship holds additional information about the DVC, as hypothesised in earlier research [24, 25].

Clusters 3 and 4 were substantially different from Clusters 1 and 2, with a clear increase in pdet during voiding. This pattern has been described previously as ‘spikes’ in both patients with detrusor‐sphincter dyssynergia and benign prostate hypertrophy [26]. The pattern clustered in Cluster 4 could represent an active sphincter contraction at the end of voiding, the premature closing of the urethra due to a high bladder neck, or the regain of bladder neck closure. Earlier observations led to the postulation in a previous study that the curve shape, as found in Clusters 3 or 4, could be the result of lower urinary tract dysfunction caused by bladder neck hypertrophy or prostate median lobe enlargement, although that study was based on a small number of patients [27]. Our study cannot confirm or refute this but does support the hypothesis that additional voiding dynamics observed in a PFS could be clinically relevant.

Some variation can be seen among the individual graphs shown in Fig. 1. The clusters are created based on what the UML algorithm supposes to be the best distribution of the graphs. The red centre line shown also depends on the included graphs, and changes during learning. The larger variation seen in Clusters 3 and 4 is the result of the choice for the k‐Shape algorithm. This algorithm is not time‐ (and in our case, flow‐) dependent. This results in less consistent graphs for the human eye. For example, almost all graphs in Cluster 3 show a decrease in pressure, followed by an increase. The exact flow rates corresponding to these events can differ due to the k‐Shape algorithm.

Although UML is largely operator‐independent, the decision on how many clusters to use for the UML is not fully objective. A limited number of clusters will be more likely to have a clinical impact as too many sub‐classes of diagnoses will map towards an individual patient instead of a patient group. However, if the number of clusters is too low, the variance within a cluster will create difficulties in assessing the relevance of the sub‐classification. The methods that we used to obtain a modified silhouette score provided some insights into the optimum number of clusters, resulting in the decision to use four clusters. This number is partly subjective and a different number of clusters could result in a more clinical meaningful outcome. However, as we found significant differences between all clusters, the decision to use four clusters seems justified.

We found significant, non‐overlapping differences in various urodynamic parameters between the four clusters extracted on the basis of UML, suggesting that the curve shape holds potential additional clinical value in addition to the currently used PFS measures. The current study is limited, however, by being a single‐centre study with no data on treatment effect available. Therefore, these findings should be further strengthened using clinical data and outcomes, ideally in a prospective study. Furthermore, we believe that filling‐phase UDS could also be studied using UML, for example redefining the detrusor overactive contraction pattern sub‐classification. [28]. In addition, the entire PFS with regard to DVC patterns and whole voiding urethral resistance patterns can be analysed using the proposed UML [29, 30]. We postulate that the use of UML for standard UDS results can provide more fundamental insights in the pathophysiology of lower urinary tract dysfunction and that this will ultimately lead to better stratified and individualised management.

In conclusion, the use of UML for analysis of urodynamic graphs is technically feasible after adequate data preparation. UML does not require a gold standard comparator to achieve results, which facilitates the in‐depth analysis of urodynamic test graphs. UML can add relevant and testable new information by including curve shape analysis of urodynamic patterns in addition to the existing one‐dimensional measure of BOO. This feasibility study has shown that UML clustering of urodynamic PFSs in men holds promise for improving the diagnosis of urethral resistance and DVC properties and dynamics.

Funding

This study was funded by the local institution.

Ethics Approval

This study was exempted from the local law on medical studies, as patients were not actively involved. The principles of the Helsinki Declaration were followed.

Informed Consent

This study was exempt from individual patient consent, due to local regulations.

Disclosure of Interests

No conflict of interest was reported by any author.

Author Contributions

W.D.: conceptualisation, methodology, software and formal analysis. P.F.W.M.R.: conceptualisation. T.R.F.S.: methodology. B.J.G. and L.M.O.K.: supervision; All authors were involved in the writing of the article.

Data Availability Statement

The datasets generated during this study are available on reasonable request.

References

  • 1. Obermeyer Z, Lee TH. Lost in thought – the limits of the human mind and the future of medicine. N Engl J Med 2017; 377: 1209–1211 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2. Gammie A, Arlandis S, Couri BM et al. Can we use machine learning to improve the interpretation and application of urodynamic data?: ICI‐RS 2023. Neurourol Urodyn 2024; 43: 1337–1343 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3. Lee HJ, Aslim EJ, Balamurali BT et al. Development and validation of a deep learning system for sound‐based prediction of urinary flow. Eur Urol Focus 2023; 9: 209–215 [DOI] [PubMed] [Google Scholar]
  • 4. Bang S, Tukhtaev S, Ko KJ et al. Feasibility of a deep learning‐based diagnostic platform to evaluate lower urinary tract disorders in men using simple uroflowmetry. Investig Clin Urol 2022; 63: 301–308 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5. Wang HS, Cahill D, Panagides J, Nelson CP, Wu HT, Estrada C. Pattern recognition algorithm to identify detrusor overactivity on urodynamics. Neurourol Urodyn 2021; 40: 428–434 [DOI] [PubMed] [Google Scholar]
  • 6. Celebi ME, Aydin K eds. Unsupervised Learning Algorithms. Switzerland: Springer International Publishing, 2016. 10.1007/978-3-319-24211-8 [DOI] [Google Scholar]
  • 7. Griffiths DJ. The mechanics of the urethra and of micturition. Br J Urol 1973; 45: 497–507 [DOI] [PubMed] [Google Scholar]
  • 8. Schäfer W. The contribution of the bladder outlet to the relation between pressure and flow rate during micturition. In Hinman F, Boyarsky S eds, Benign Prostatic Hypertrophy. New York, NY: Springer, 1983. 10.1007/978-1-4612-5476-8_44 [DOI] [Google Scholar]
  • 9. Schäfer W. Analysis of bladder‐outlet function with the linearized passive urethral resistance relation, linPURR, and a disease‐specific approach for grading obstruction: from complex to simple. World J Urol 1995; 13: 47–58 [DOI] [PubMed] [Google Scholar]
  • 10. Lim CS, Abrams P. The Abrams‐Griffiths nomogram. World J Urol 1995; 13: 34–39 [DOI] [PubMed] [Google Scholar]
  • 11. van Dort W, Rosier PFWM, van Steenbergen TRF, Geurts BJ, de Kort LMO. Constrictive versus compressive bladder outflow obstruction in men: does it matter? Neurourol Urodyn 2024; 43: 2178–2184 [DOI] [PubMed] [Google Scholar]
  • 12. van Dort W, Rosier PFWM, Geurts BJ, van Steenbergen TRF, de Kort LMO. Quantifying bladder outflow obstruction in men: a comparison of four approximation methods exploiting large data samples. Neurourol Urodyn 2023; 42: 1628–1638 [DOI] [PubMed] [Google Scholar]
  • 13. Schäfer W, Abrams P, Liao L et al. Good urodynamic practices: uroflowmetry, filling cystometry, and pressure‐flow studies. Neurourol Urodyn 2002; 21: 261–274 [DOI] [PubMed] [Google Scholar]
  • 14. Rosier PFWM, Schaefer W, Lose G et al. International continence society good urodynamic practices and terms 2016: urodynamics, uroflowmetry, cystometry, and pressure‐flow study. Neurourol Urodyn 2017; 36: 1243–1260 [DOI] [PubMed] [Google Scholar]
  • 15. Kranse R, van Mastrigt R, Bosch R. Estimation of the lag time between detrusor pressure‐ and flow rate‐signals. Neurourol Urodyn 1995; 14: 217–229 [DOI] [PubMed] [Google Scholar]
  • 16. Paparrizos J, Gravano L. K‐shape: efficient and accurate clustering of time series. ACM SIGMOD Rec 2016; 45: 69–76 [Google Scholar]
  • 17. Eckhardt MD, van Venrooij GE, Boon TA. Urethral resistance factor (URA) versus Schäfer's obstruction grade and Abrams‐Griffiths (AG) number in the diagnosis of obstructive benign prostatic hyperplasia. Neurourol Urodyn 2001; 20: 175–185 [DOI] [PubMed] [Google Scholar]
  • 18. Cornu JN, Gacci M, Hashim H et al. EAU Guidelines on Non‐Neurogenic Male Lower Urinary Tract Symptoms (LUTS) Update April 2024. European Association of Urology, 2025. Arnhem, The Netherlands: EAU Guidelines Office. ISBN 978‐94‐92671‐23‐3. [Google Scholar]
  • 19. Rosier PFWM, Gammie A, Valdevenito JP et al. ICS‐SUFU standard: theory, terms, and recommendations for pressure‐flow studies performance, analysis, and reporting. Part 2: analysis of PFS, reporting, and diagnosis. Neurourol Urodyn 2023; 42: 1603–1627 [DOI] [PubMed] [Google Scholar]
  • 20. Brownlee J. Data Preparation for Machine Learning: data Cleaning, Feature Selection, and Data Transforms in Python. Vermont, VIC: Machine Learning Mastery, 2020. [Google Scholar]
  • 21. Rosier PF, de la Rosette JJ, Koldewijn EL, Debruyne FM, Wijkstra H. Variability of pressure‐flow analysis parameters in repeated cystometry in patients with benign prostatic hyperplasia. J Urol 1995; 153: 1520–1525 [PubMed] [Google Scholar]
  • 22. Spångberg A, Teriö H, Ask P, Engberg A, Griffiths D. Pressure/flow studies preoperatively and postoperatively in patients with benign prostatic hypertrophy: estimation of the urethral pressure/flow relation and urethral elasticity. Neurourol Urodyn 1991; 10: 139–167 [Google Scholar]
  • 23. Teriö H, Spångberg A, Engberg A, Ask P. Estimation of elastic properties in the urethral flow controlling zone by signal analysis of urodynamic pressure/flow data. Med Biol Eng Comput 1989; 27: 314–321 [DOI] [PubMed] [Google Scholar]
  • 24. Kranse R, van Mastrigt R. Bladder outlet resistance. J Urol 2003; 169: 1007–1010 [DOI] [PubMed] [Google Scholar]
  • 25. van Mastrigt R, Kranse M. Computer‐assisted pressure‐flow analysis. Urol Clin North Am 1996; 23: 393–405 [DOI] [PubMed] [Google Scholar]
  • 26. Vírseda M, Salinas J, López A, Esteban M. Usefulness of dynamic urethral resistance relation (DURR) measurement for differential diagnosis between static and dynamic urinary obstruction in male spinal cord injury patients. Neurourol Urodyn 2012; 31: 549–555 [DOI] [PubMed] [Google Scholar]
  • 27. Rosier P. Initial observation of the urodynamic pressure flow study – characteristics of bladder‐neck or prostate median lobe dynamics during micturion. Continence 2023; 7: 101042 [Google Scholar]
  • 28. Van Steenbergen T, Van Dort W, de Lange R, Rosier P. Objective parameters to describe detrusor overactivity patterns. Continence 2023; 7: 100875. 10.1016/j.cont.2023.100875 [DOI] [Google Scholar]
  • 29. Van Mastrigt R, Rollema HJ. The prognostic value of bladder contractility in transurethral resection of the prostate. J Urol 1992; 148: 1856–1860 [DOI] [PubMed] [Google Scholar]
  • 30. Bosch JL, Kranse R, van Mastrigt R, Schröder FH. Dependence of male voiding efficiency on age, bladder contractility and urethral resistance: development of a voiding efficiency nomogram. J Urol 1995; 154: 190–194 [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Data Availability Statement

The datasets generated during this study are available on reasonable request.


Articles from Bju International are provided here courtesy of Wiley

RESOURCES