A Physics-Driven Smartphone Framework for Remote Uroflowmetry Using Computer Vision
Uroflowmetry is routinely used in urology to characterize urinary flow during voiding and to support assessment of lower urinary tract dysfunction. In conventional testing, the patient uses a dedicated collection device equipped with a calibrated transducer. This arrangement generally limits testing to clinical facilities, adds equipment and maintenance requirements, does not retain a direct visual record of the void, and may influence normal voiding behavior. This study presents a physics-informed video analysis framework in which three-tier computer-vision architecture is used instead of dedicated flow-rate acquisition hardware. A smartphone recording is scaled using a fixed 26 cm physical reference and passed through a processing pipeline comprising background subtraction (MOG2), region-of-interest (ROI) locking, width-based gating, morphological cleanup, vertical dilation, and geometric (aspect-ratio) filtering to distinguish the stream from the surrounding body, hands, and background. Velocity tracking and volume normalization convert the isolated stream into a clinically meaningful flow curve, from which metrics including peak flow rate (Qmax), average flow rate, flow time, voiding time, time to Qmax, and hesitancy are derived. A confidence-weighted ensemble fuses top- and side-view recordings for improved robustness. The resulting workflow is implemented as an end-to-end clinical web platform: a React-based frontend for patient intake, video upload, and report viewing; a FastAPI backend providing JWT-authenticated, role-based endpoints for patient management, analysis orchestration, and report/email delivery; and cloud-based media storage. This work is presented as proposed and implemented system architecture; quantitative validation results, accuracy figures, and dataset-level evaluation are not yet available and are identified as required future work. The principal contribution is a smartphone-based acquisition approach that avoids dedicated uroflow transducer hardware that extends existing AI-based uroflow curve-classification research, which assumes a transducer-derived signal, into the signal-acquisition stage itself.
Authors
- Bhargavi Naik
- Ramya B N
- Bhavana Mohandas sangalad
- Anusha K
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.5281/zenodo.24272202
- Primary Topic
- Urinary Bladder and Prostate Research
- Type
- article
- Field-Weighted Citation Impact
- 0.00