Privacy-Preserving Automatic Collection of Acoustic Voiding Events.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Urology Diagnostics

Background:

• Uroflowmetry is a key non-invasive test for urinary tract function.
• Current uroflowmetry faces challenges with high data variability and on-demand voiding.
• These limitations hinder accurate and convenient patient monitoring.

Purpose of the Study:

• To develop a novel, low-cost ultrasonic platform for automatic, in-home uroflowmetry.
• To leverage machine learning for detecting natural voiding events without user intervention.
• To enable privacy-preserving urinary flow analysis in daily routines.

Main Methods:

• Development of a low-cost ultrasonic sensing platform operating outside human-audible frequencies.
• Implementation and evaluation of various machine learning models for event detection.
• Validation of the Multi-layer Perceptron classifier for its performance in classifying voiding events.
• Assessment of voiding flow envelope integrity using inaudible frequencies.

Main Results:

• The Multi-layer Perceptron classifier achieved 97.8% accuracy with a 1.2% false negative rate.
• Machine learning models, including lightweight SVM, demonstrated robust performance.
• The voiding flow envelope, crucial for diagnostics, was preserved using inaudible frequencies.
• The system enables automatic, privacy-preserving uroflowmetry during natural voiding.

Conclusions:

• The developed ultrasonic platform with ML effectively overcomes traditional uroflowmetry limitations.
• This technology offers a viable solution for remote, continuous urinary tract monitoring.
• It holds significant potential for urology telemedicine, especially in resource-limited settings.