Kyto 2935: a decent, cheap sensor for spot check measurements
As you probably know already, I’m a big fan of morning measurements to assess our resting physiology (e.g. heart rate variability, or HRV).
By measuring intentionally, we can control most confounding factors, and gain insights that we would otherwise miss with automated data collection or data collection during sleep (this is due to the better timing of morning measurements, and also to differences in body position, see some examples here).
On top of the reliability, the morning routine offers a great opportunity to stop, reflect, and improve our self-awareness, e.g. by filling in a subjective questionnaire before looking at our data, as in the HRV4Training app.
Often the opposite happens when looking at wearables data right after waking up, letting the data (and possible inaccuracies or confounding factors) decide how our day will be - and often turning tracking into unhealthy behaviors.
When it comes to assessing our body’s response to stress and using the data effectively, less is more.
In this context, we keep an eye on sensors that can make it easier to take spot-check measurements, as not everyone likes to use a chest strap (even though the chest strap - e.g. a Polar H10 - remains the only way to actually measure heart rate variability or the electrical activity of the heart, as opposed to pulse rate variability or changes in pulse rate at the periphery, such as the wrist, finger or else).
Unfortunately, most wearables do not implement the standard Bluetooth heart rate profile, meaning that you can use them only via their app, and the wearable cannot communicate with any third party apps, which is obviously a pity as much more could be done with these sensors (or maybe the raw data is so bad that it’s better that you never see it? food for thought).
An alternative that I recommend is the Scosche Rhythm24 (not to be confused with the newer Rhythm2 and Rhythm+, which cannot do HRV analysis). This is an older sensor that includes an HRV mode and therefore can be used with HRV4Training and other apps. Make sure not to move at all and not to contract any muscle when using optical sensors for HRV analysis, because otherwise the data is complete garbage, as I’ve shown several times (e.g. here). This is one of the reasons why it makes no sense to use HRV for continuous stress monitoring outside of a spot check or night data (when there is no movement).
Another alternative is the Kyto 2935 which can be connected to the ear or finger, depending on your preference:
Below is some data collected using the Kyto device. Sessions 001 and 003 were collected with the finger sensor, while 002 with the ear sensor, and compared against a chest strap, the Polar H10.
The protocol consisted of either normal breathing at rest, while seated, or deep breathing exercises. For data collection, I used the HRV Logger app to collect RR intervals from the chest strap, and another HRV Logger app to collect RR intervals using the Kyto (my apps are available here).
RR and PP intervals:
We can see that the PP and RR intervals track very well the different phases of the protocol (minutes 1-4 with regular breathing, and minutes 5-6 with deep breathing), and are rather similar.
However, when computing rMSSD from RR and PP intervals for the data above and another two tests, we do have an offset in the data:
Still, we can see that the relative changes within sessions are captured very well, even though for some reason, data from the Kyto sensor seems lower (which is somewhat odd, as typically we get higher variability with PPG). I wonder if there is some form of overcorrection in place.
As my recommendation is never to do anything with absolute values of HRV, but to track changes over time, I consider this device good enough for the task, given the PP intervals above, and the ability to clearly capture relative changes even in a short segment of data.
Just make sure not to compare data collected with this sensor, with data collected with a different device (this is the case for any sensing modality, as there can always be differences based on where the sensor is placed, the technology used, artifact correction, etc.).
Marco holds a PhD cum laude in applied machine learning, a M.Sc. cum laude in computer science engineering, and a M.Sc. cum laude in human movement sciences and high-performance coaching.
He has published more than 50 papers and patents at the intersection between physiology, health, technology, and human performance.
He is co-founder of HRV4Training, advisor at Oura, guest lecturer at VU Amsterdam, and editor for IEEE Pervasive Computing Magazine. He loves running.
Social:
Hi Marco,
Thanks for the interesting article!
I have been using the HRV4training app for several years now (with phone camera), do you think this sensor (or any other sensor) would provide a better quality of data over the phone camera?
The phone camera can sometimes be a little tricky, depending on how still you hold it, finger at the right location etc. I've also used the Polar H10 in the past, but this gave other issues - strap had to be moist otherwise data was not always good. Maybe the Kyto could provide better data without the issues?
Thanks in advance for your time to answer, and for writing these interesting articles!