Urinary Bladder Monitoring with a Wearable Device Based on Bioimpedance Measurements and Machine Learning

Neurogenic lower urinary tract dysfunction (NLUTD) can lead to an inability to recognize the need to empty the bladder. To the best of our knowledge, there is currently no established technological solution to assist patients. Prior research has demonstrated the potential of bioimpedance (BI) measurements combined with machine learning (ML) for the non-invasive monitoring of bladder filling. However, research on the processing of BI measurements under real-world conditions is scarce. In this paper, we built an ML-based prototype of a wearable BI device to validate and build on the results of existing work. We contribute to the body of knowledge by presenting an ML approach to leveraging BI measurements collected by a 16-electrode wearable device for bladder monitoring under real-world conditions. We validated our prototype with in-vivo experiments and evaluated its performance on a real-world dataset. The presented prototype employs a deep hybrid learning model based on convolutional and recurrent neural network architectures, which reaches a mean absolute error of 87.9 ml. In contrast to the current gold standard of time-based voiding, our prototype marks an important step toward a technological solution for NLUTD patients that is demand-driven and suitable for everyday use.