AI Breath: biosignals and machine learning for better breath analysis
I guess you heard about Neuralink’s Brain-Machine-Interfaces, BrainCo’s focus headband or at least have a fitness tracker around your wrist 24/7. As you might know, these products are just at the tip of the iceberg of the global trend towards individual healthcare, healthy living and “biohacking”.
Biosignals, basically all signals in living beings that can be measured and monitored, are at the core of this trend. Although the data is often hard to collect, and is very noisy, biosignals offer a huge space of possibilities for understanding, healing, and improving our bodies.
And while the amount of available biosignal sensors has been growing immensely, there are still very few really human-centered devices out there. So, in our recent Motius Discovery round, we decided to dive deeper into biosignals and develop a cutting-edge device using biosignals.
Why we chose to build a wearable spirometry device
Possibilities to measure biosignals are endless. Whether it is your brain, eyes, muscles, skin, blood — you can always measure something somewhere. Here’s just a short list:
- Brain: electromagnetic fields, anything your brain does
- Nose: airflow, depth and duration of breath
- Eyes: tear fluid, pupil dilation, gaze tracking
- Muscles: deformation, electric signals
- Skin: body temperature, heart rate, pulse, blood pressure, sweat, skin color
- Blood: minerals, vitamins, hormones, pH value
- Tissue composition: organs, skin, blood vessels
As we looked at current state-of-the-art sensors for various biosignals, spirometry devices got our attention. Spirometry is the classic way to do breath analysis and get many insights into the patient’s health, for example:
- lung volume
- overall breathing rate
- fitness
- stress level
- heartrate (indirectly)
However, as you can see in the picture below, there are major drawbacks with modern spirometry: it is expensive, complicated, not portable and does not have a differential measurement. We decided to change that and build a device that can improve the way spirometry tests are conducted.
AI Breath for a better breath analysis
Our idea was to apply IoT sensors in a new context of our body’s biological signals. We wanted to:
- measure breaths with low-cost devices via the nose
- be able to record data on the go
- compare outflow from nostrils
Further, our goal was to find data patterns that would allow us to develop a machine learning algorithm which would then improve the breath analysis. We called our idea “AI Breath”.
Building and testing AI Breath
We used low-cost off-the-shelf pressure sensors and microcontrollers to build AI Breath. After setting them up, we also needed to set up our data analytics IoT toolkit. To do so, we used motius.io. In this case, we used:
- RabbitMQ as our data broker
- Telegraf as our data forwarder
- Influx as our database
- Grafana to build our dashboard
Next, we wanted to set up the live visualisation of our sensor. But first, we needed to 3D-print the nozzles and a holder that can be placed close to the nose. After that, it was finally time to assemble AI Breath.
For our test session, we set it up right in front of the nose to enable optimal breath measurements. Surprisingly, we obtained high-quality data within only one afternoon. To display our results, we used fast Fourier transform (FFT), a common way to analyze different frequency spectrums.
The most important result was that AI Breath works, as the picture below shows. We measured the breath and were able to label different measurements to corresponding situations and activities. This might sound unspectacular to you but it shows that we are able to build a low-cost breath measurement device that solves the problems of spirometry and therefore enables various healthcare use cases. If you just think about the current pandemic and the role that lung capacity plays, you might grasp the importance and potential of this idea.
Although we obtained good enough data, we did not get enough to be able to train a machine learning algorithm. However, this is not the end of the road — we just need to do further research.
A product for the future of measuring biosignals
In this Motius Discovery project, we have built a low-cost breath measurement device that has the potential to significantly improve how breath measurement is done these days. We obtained high-quality data that need further research to realize our initial idea of including a machine learning algorithm.
Looking at current developments in the medical and consumer industry, AI Breath has high potential to make a real impact. Just let us do some more research, iterate on our idea and then develop it all the way through. That is what we do at Motius — we R&D.
