Automatic Classification of Audio Uroflowmetry with a Smartwatch

Area of Science:

• Biomedical Engineering
• Urology
• Machine Learning

Background:

• Voiding dysfunctions are common urological conditions.
• Uroflowmetry is a standard diagnostic tool, but requires specialized equipment.
• Machine learning has shown promise in analyzing medical data.

Purpose of the Study:

• To investigate the use of smartwatch audio for classifying voiding signals.
• To develop an automated system for detecting normal versus abnormal voiding patterns.
• To assess the feasibility of using ubiquitous audio capture devices for urological diagnostics.

Main Methods:

• Utilized the UroSound platform to collect smartwatch audio data.
• Applied classical machine learning techniques for signal classification.
• Trained and evaluated multiple classification models, including ensemble methods.

Main Results:

• Achieved a maximal test accuracy of 86.16% in classifying voiding signals.
• Demonstrated the effectiveness of ensemble methods for this classification task.
• Validated the potential of audio analysis for detecting voiding abnormalities.

Conclusions:

• Smartwatch audio analysis is a viable method for classifying voiding signals.
• This approach can supplement traditional diagnostic tools in urology.
• The UroSound platform offers a promising solution for urology telemedicine, particularly in resource-limited settings.