Sleep 2.4 Signal Processing Configurability: Parameter Settings

Actigraphy settings, including epoch window sizes, filtering methods, and frequency ranges, and their impact on data quality and research flexibility.

Video Breakdown:

00:00 - Introduction to Actigraphy Configurations

00:22 - Choosing Between ZCM, Energy, and TAT

01:28 - Optimal Epoch Window Sizes for Research

02:33 - Understanding Filter Configurations & Frequency Bands

04:09 - Expanding Research Capabilities with Custom Settings

Okay, in, in terms of configurability and parameter settings, I think these are some of the options, again, that you have, possible for you, that Garmin gives our Labfront platform gives along with Garmin's SDK.
In terms of choosing whether to do the ZCM, Energy, or the TAT threshold, actually you don't have to choose between all, like choose between the three because Garmin enables all three to be simultaneously obtained for this particular actigraphy one setting. However, if you're really interested, ZCM is, has been used for Cole-Kripke, Sadeh, and the UCSD algorithm.
The energy method has been used for ActiWatch, can also be used for a UCSD algorithm, and TAT is largely used only in the UCSD algorithms. So there is some opportunity here to use TAT in ways that hasn't been done before. We talked about the filter path and I'm not going to go into too much detail again about that, but you have the option of choosing FMpre and FMpost, but again recognizing that there are some compatibility issues between let's just say FMpre and ZCM.
Now when it comes to choosing epoch window size, whenever possible my opinion is to opt for shorter windows. And the reason is simple. While you can always aggregate data from smaller windows into larger ones, you can't do the reverse. Since our activity measures are linear and additive, it's really straightforward to combine multiple, let's say, 5 second impacts into 60 second windows if needed.
However, if you initially record, 60 second epochs, there's no way to back that down into more granular five second resolution later. So this flexibility in data analysis makes shorter epoch windows a more versatile choice for your research, but it has the disadvantage of having larger data volume and thus larger, longer data transfer times, um, for, uh, for the participants.
So it would take longer sync time when the person's on our mobile app.
Aagain, this just shows you that this applies to all the three windows, because it's simply additive. Um, you could choose the smaller windows if you wanted to.
Now finally, the filter configuration. The Butterworth filter enables us to choose the frequency range and it gets a little complicated here and again, this is something where we'll have to, you'll have to reach out to us to establish what settings or what parameters that you would need to, coefficients into this equation to choose the frequency bands that you're trying to, to go for.
But from my physiological understanding, when it comes to movement, It could probably be divided into two frequency bands. One that's lower frequency, which is from 0. 3 to 3 hertz, and this captures sustained physical activities like walking and slow movement, and it identifies sort of overall activity patterns and daily rhythms.
High frequency guards more like 2 to 12 hertz, and captures more rapid movements, tremors or shorter bursts of activity. And it, you know, it could be useful for specific types of movements or behaviors. Maybe something that, you know, has to do with some sort of periodic limb movements, or some type of run behavior disorders. Or people like acting out their dream, for instance.
So, if you have, if, you know, again, we have these three options, possible, I would have to say that the options are probably the combination of one with FMpre with lower frequencies, a second is FMpost with lower frequencies, and a third, option would be to have FMpre with higher frequencies to capture sort of that rapid, uh, specific behavioral movements.
Because our actigraphy features basically generates nine features, so three for each settings, you could really, utilize that type of data, and train it to your needs and apply your specific algorithm if you wanted to, but again, this does require some kind of gold standard to measure it against, but this does expand your option when it comes to, doing actigraphy research.

Course episodes

Sleep 2.1 Intro to Sleep Actigraphy

Sleep 2.2 Traditional Signal Processing: Sleep-Wake Classification

Sleep 2.3 Piezoelectric vs Capacitive Accelerometers

Sleep 2.4 Signal Processing Configurability: Parameter Settings

Sleep 2.5 Unexplored Research Opportunities