Privacy-Preserving Automatic Collection of Acoustic Voiding Events.

Uroflowmetry is a non-invasive diagnostic test used to evaluate the function of the urinary tract. Despite its benefits, it has two main limitations: high intra-subject variability of flow parameters and the requirement for patients to urinate on demand. To overcome these limitations, we have developed a low-cost ultrasonic platform that utilizes machine learning (ML) models to automatically detect and record natural in-home voiding events, without any need for user intervention. This platform operates outside of human-audible frequencies, providing privacy-preserving, automatic uroflowmetries that can be conducted at home as part of daily routines. After evaluating several machine learning algorithms, we found that the Multi-layer Perceptron classifier performed exceptionally well, with a classification accuracy of 97.8% and a low false negative rate of 1.2%. Furthermore, even on lightweight SVM models, performance remains robust. Our results also showed that the voiding flow envelope, helpful for diagnosing underlying pathologies, remains intact even when using only inaudible frequencies.Clinical relevance- This classification task has the potential to be part of an essential toolkit for urology telemedicine. It is especially useful in areas that lack proper medical infrastructure but still host ubiquitous embedded privacy-preserving audio capture devices with Edge AI capabilities.

Classification of the Physical Surface in Sound-based Uroflowmetry.

This work constitutes a first approach to automatically classify the urination medium for non-invasive sound based uroflowmetry tests. Often the voiding flow impacts the toilet wall (often made of ceramic) instead of the water. This causes a reduction in the amplitude of the recorded audio signal, and thus a reduction in the amplitude of the extracted envelope. Analysing the envelope alone, it is not possible to tell accurately if the reduction in the amplitude is due to a low voiding flow or an impact on the toilet walls. In this work, we carry out a study on the classification of sound uroflowmetry data depending on the medium where the urine impacts within the toilet: water or ceramic. In the analysis, a classification algorithm is proposed to identify the physical medium automatically based on the urination acoustics. The classification algorithm takes as input the frequency spectrum, the variance, and the kurtosis of the audio signal corresponding to a voiding event.Clinical relevance- Sound uroflowmetry has a strong correlation with the standard uroflowmetry. It is useful for the non-invasive detection of pathologies associated with the urinary tract as a support tool for information processing and screening. It consists of a characterization of the urinary flow patterns by capturing the sound generated when the urine stream impacts the water in the toilet. Identifying the medium which originates the sound is of paramount importance to better interpret the sound uroflowmetry.