Clinical validation of SonoCurve, a machine-learning sound-based uroflowmetry algorithm, in adult men with lower urinary tract symptoms

Objectives To validate the clinical performance of SonoCurve, a machine-learning algorithm that provides uroflowmetry metrics from the sound of the urinary void, in adult men with lower urinary tract symptoms (LUTS). Design Prospective within-person equivalence validation study. Setting Single-centre specialist urology clinic. Participants Adult men undergoing uroflowmetry as part of LUTS assessment (September 2024–February 2025). Intervention Simultaneous conventional uroflowmetry and smartphone audio recording of the urinary stream. Recordings were processed using SonoCurve to calculate flow parameters from sound spectrograms. Outcome measures Primary: equivalence of maximum flow rate (Qmax) versus conventional uroflowmetry within a ±2 mL/s margin. The mean absolute difference (MAD) described pairwise variability. Secondary: agreement analyses for Qmax, average flow rate (Qave), voided volume (VV) and voiding time (VT), and curve-shape similarity. Results 62 men (median age 55 years, range 19–85) were included. SonoCurve achieved equivalence for Qmax (mean difference −1.19 mL/s; SD 3.11; 90% CI −1.85 to −0.53 mL/s; p=0.022). MAD was 2.69 mL/s. Lin’s concordance correlation coefficients (95% CI) were: Qmax 0.81 (0.71–0.88), Qave 0.90 (0.85–0.93), VV 0.95 (0.92–0.97) and VT 0.97 (0.95–0.98). Bland-Altman plots showed minimal bias and narrow limits of agreement. Flow-curve similarity was high (root mean square error (RMSE) 2.88±1.23 mL/s, dynamic time-warping (DTW) RMSE 0.84±0.49 mL/s). No adverse events occurred. Conclusions SonoCurve accurately measured uroflowmetry parameters from voiding sound in men with LUTS, achieving statistical equivalence for Qmax and excellent concordance across all secondary measures. Findings support use of SonoCurve as a precise, accessible alternative to clinic-based uroflowmetry and highlight its potential role in remote male LUTS assessment.