In order to reduce hospitalisation through UTI’s (Urinary Tract Infections) HD developed a novel digital urine sensor health tracking device, to provide early detection of poor hydration or urinary tract infections in catheterised patients. This is achieved by analysing changes in the visible reflectance profile of urine to detect changes which may indicate underlying health issues. Ongoing identification of potential issues ensures earlier interventions can be implemented before more serious conditions develop.
Early lab trials of this device have proven highly successful so it is possible to quickly identify changes in urine colour, detect cloudiness and blood within the urine too. This novel sensor clips to the entry tube of a drainage bag and detects changes in urine output by measuring the magnitude of wavelengths at specific visible and NIR points. It can consistently detect slight colour changes or inclusions which the human eye cannot
The monitor gives users and/or carers a fast and clearer picture of bladder and urinary tract health so quick action can be taken if a health risk is detected.
When it comes to urine, colour is a very good indicator about what is going on in your bladder and your body, however, there could be confounding factors that cause it to darken other than simply poor hydration for example. Currently, quick tests involve dipsticks to overcome the limitations of colour by analysing the specific chemical make-up of the sample – exaggerated presence or lack of certain compounds can therefore be used to analyse the patient’s current condition. But the time at which these tests are taken, in general practice for example could be long after a UTI has formed. Using some form of real time sensor, could detect common issues sooner for catheter wearers in care or in the home.
HD created two physical prototypes for a real time monitoring device:
If the sensor was to be simple to use, not messy to get set up then a clip-on device which does not contact the urine itself was essential. Any inline sensor, in contact with urine, would mean using barb fittings, or connectors, whilst requiring changes to existing devices and systems. This would prevent this device from being universal and easily reusable. This left colourimetry as the primary option. Understanding the limitations as previously outlined HD discovered a multi-channel wavelength sensor, dark urine – is a combination of wavelengths, but slightly different combinations of certain wavelengths can still produce the same colour (this is how TV’s work!), we questioned whether spikes in the specific wavelength on a set of test data would give us better fidelity than what the overall colour is.
After reviewing use cases, we conducted further testing to identify detectable elements and key factors to focus on.
Using the works like prototype, we were able to develop a classified dataset based on a set of mock urine samples. Using ‘test tubing’ made from a drainage tube, we created several ‘fake’ urine sample which could be passed in front of the wavelength sensors to create a classified dataset.
Passing each colour sample through the device gave us a large dataset of training data so that we could create a simple machine learning model to predict the status of the urine passing in front of the sensors. Using a form of decision tree, the model takes the sensor reading and weights the likelihood that the collected variables belong to a tested classifier. On our initial test data, the accuracy was in the region of 95%. As we were looking at only a small number of samples, it was important that the model was not over fitted to the handful of test samples.
The model trained was then tested both with second sets of test tubes, passing fluid through a tube to replicate the use case. In both scenarios the device was accurate and approximated what it was seeing when fluid was passed from a syringe through a tube in front of the sensors.
With this promising sign, we began to think more about the usability of the device in situ. How were the alerts going to be presented to users? We considered haptics from the device or sound; haptic motors were inserted into the works-like device but due to the number of things which could be detected we felt that this approach would not provide enough information.
We therefore developed UI mock-ups for a mobile app as part of a connected system to represent a patient and healthcare professional portal to help inform catheter users and district nurses on the risk of condition being prevalent so that the correct actions can be advised by a healthcare professional.
Due to the complex nature of urine testing, the device cannot diagnose a condition – but it can certainly alert if it suspects a health risk. With more work and research, it is feasible to refine the technology further, should an appropriate application be identified. It can consistently identify bubbles for instance.
Compatibility across brands – many drainage bags have different tubes which may confuse the model – data would need to be gathered from multiple common drainage tubes and specified in set-up.
Drips – droplets in the tube can currently confuse the device so cause it to change its output ready rapidly.
Dataset – A large study is required to improve the accuracy of the dataset with real samples.
Power – The sensor cannot always be on and continuously detecting, a form of ‘wake-up’ function would need to be implemented such as a capacitive sensor on the tube to only detect when fluid is passing through.
Overall, there are still several barriers to overcome with a device like this before being launched. For connected measuring and analysis products, Haughton Design has the capability to help you create novel solutions to difficult use cases. Please get in touch to find out how we can help your device development.