When You Gotta Go, Audioflow

Many types of important health diagnostics require expensive laboratory equipment and trained professionals to operate and interpret the results. These factors limit the ability of these tests to detect health problems and allow early intervention plans to be initiated. Fortunately, in recent years there has been a trend towards miniaturization of diagnostic devices that has allowed for the production of inexpensive equipment that is foolproof enough for virtually anyone to operate at any time. This offers many benefits, like constant monitoring, and recording measurements under normal conditions, rather than under the artificial conditions of a lab visit.

Thanks to a recent breakthrough, miniaturization of another diagnostic device may be on the way to help a whole new class of patients. Blockages of the urinary tract and issues with the bladder can be very painful and can also lead to bigger health problems. One of the first steps in diagnosing these conditions is uroflowmetry — this involves using laboratory equipment to measure the volume of urine released from the body, and also the speed and duration of that release. But because this test must be done in a clinic, it can be very inconvenient, and cannot be done on a regular basis.

A team of physicians and engineers at the Singapore General Hospital have developed a novel method to assess parameters related to the flow of urine. Coming straight from the “I sure am glad they built a machine to do this job” department comes Audioflow, a tool designed to capture audio recordings of urination, then use a machine learning data analysis pipeline to seek out the presence of any abnormal conditions. This small device is capable of pairing with a smartphone to analyze the data, making it suitable for use by anyone, without the need for any special expertise.

Using a dataset of 220 audio recordings, Audioflow was trained to estimate flow rate, volume, and time. These are critical parameters in determining if there is an obstruction, or if the bladder is not working as it should. It is important to note that the present data was collected solely from men. An entirely different dataset would need to be generated, and the model would need to be retrained, for the device to be effective in assessing female patients.

Conventional uroflowmetry machines are the present gold standard diagnostic devices, so Audioflow was compared to its results by capturing audio during normal testing. It was found that the new AI-powered solution agreed with the uroflowmetry machine 80% of the time. In some cases, the job of identifying abnormal flows falls to specialist urologists and external residents, so Audioflow was also compared with their assessments as well. In this case, an 84% rate of agreement was observed.

These results are very encouraging, however at this time all of the work has been done exclusively in soundproof environments. Before this technique can be used in real world situations, the model will need additional training, and perhaps additional refinements, so that it can learn to ignore background noises. With a bit more work, Audioflow may represent yet another case of machine learning improving people’s lives in a meaningful way.