When a patient is receiving intravenous (IV) fluids, it is vital that air is not introduced into the line, as its presence can create issues such as excessive pressure or even embolisms that can be life-threatening. Normally, the level of fluids remaining within the bag is periodically checked by a nurse, but due to challenges related to staffing, this might not be enough. Therefore, Manivannan Sivan devised an automated monitoring system that uses computer vision along with machine learning to do this repetitive task instead of a person.

To begin the project, Sivan gathered a series of images that spanned three categories with an Arduino Portenta H7 and Vision Shield. Within each picture, the IV fluid bag is 15cm away and contains either an adequate, less than 50%, or low level of fluid remaining. Once trained, his model was able to successfully recognize the correct level around 96% of the time, although additional images taken at different light levels would help improve the accuracy even more.

From here, Sivan deployed his model back to the Portenta H7 to see how it performs in day-to-day conditions, with classification taking place once every two seconds. His plan is to take the resulting category and export it to an awaiting server over the network so that hospital staff can see the fluid bag’s status in real-time on a dashboard.

For more details on this project, you can watch Sivan’s explainer video below or read his write-up on the Edge Impulse docs page.