Urodynamic studies are the gold standard for assessing lower urinary tract dysfunction. Quality assurance is a key component of urodynamic testing however, significant variability in test quality has been reported in clinical practice and research settings.  Advances in artificial intelligence offer opportunities to support quality control processes. Currently, there is no automated method to assess study quality during or after acquisition [1]. This pilot study aims to develop an automated quality assessment approach for urodynamics, focusing on cough test events. We hypothesised that a rule-based signal processing algorithm could detect cough events across the urodynamic study timeline in a way consistent with standard urodynamic interpretation.

We retrospectively analysed urodynamic studies from 300 consecutive patients. Multichannel signals including intravesical pressure (Pves), abdominal pressure (Pabd), urine flow rate (Qura), and infused volume (Vin) were extracted from raw recordings sampled at 10 Hz. 

A clinically informed rule-based signal processing framework was developed to automatically detect cough events (Figure 1). The system was designed as an expert-driven inference model, in which predefined physiological rules derived from urodynamic practice [2] were encoded into a structured knowledge base of thresholds and constraints.  Signals were first smoothed using a centred rolling average, and detrusor pressure (Pdet) was calculated as the difference between intravesical and abdominal pressures. Pressure changes were evaluated relative to a patient-specific baseline defined from the initial recording period to account for inter-individual variability.

A valid cough event was defined as a sharp and symmetrical increase in both Pves and Pabd lines (20–200 cmH₂O), with minimal change in detrusor pressure (|Pdet| ≤ 5 cmH₂O). A true cough produces a comparable increase in both Pves and Pabd pressures, quantified using a symmetry ratio (smaller/larger change), with a threshold ≥0.8 indicating a physiologically consistent cough. Cough detection was implemented by applying combined physiological criteria, including amplitude thresholds, symmetry constraints, detrusor stability, and rapid rise characteristics, to the pressure traces. To improve robustness to signal noise, temporally adjacent detections were merged to form into continuous candidate events, allowing identification of cough episodes at the event level rather than as isolated signal points. 

The urodynamic timeline was anchored using two clinically defined events: permission to void and end of void. Detected cough events were classified according to their temporal position relative to these anchors, including pre-filling, during filling, pre-void (within 30 seconds before permission to void), and post-void (within 30 seconds after void completion) periods.

The algorithm was applied to all 300 urodynamic studies, enabling automated detection and temporal classification of cough events for all cases. Validation was performed on a randomly selected subset of 25 outputs, independently reviewed by a urologist. Detected events showed good agreement with clinically expected pressure patterns. Phase-specific performance demonstrated high sensitivity for pre-filling cough detection (0.95) and sensitivity of 0.818 for pre-void coughs, while post-void detection was lower (0.235). Overall intermittent cough detection (coughs during filling) showed high precision (0.985) and recall (0.736). Figure 2 shows good examples of detections (A), missed cases (B), and performance metrics (C).

The rule-based framework demonstrates strong clinical alignment, effectively identifying physiologically consistent cough patterns while maintaining high precision. Reduced performance in the post-void phase is likely related to increased signal variability and lower pressure amplitudes, as well as potential discrepancies in clinical timing. The approach offers an interpretable and reproducible method for automated quality assessment in urodynamics.

This study presents a rule-based framework for automated cough detection in urodynamics, demonstrating good agreement with clinician validation. The approach shows potential to support quality assessment system.

Yao, M. and Simoes, A., 2020. Urodynamic testing and interpretation.
Rosier, P.F., Schaefer, W., Lose, G., Goldman, H.B., Guralnick, M., Eustice, S., Dickinson, T. and Hashim, H., 2017. International Continence Society Good Urodynamic Practices and Terms 2016: urodynamics, uroflowmetry, cystometry, and pressure-flow study. Neurourology and urodynamics, 36(5), pp.1243-1260.