Strange output regarding VPA (clipping?)

[Nina Vansweevelt]

Hi Jairo and Vincent,

I processed data from wrist-worn Actigraphs and I found something strange in my output. I had a participant with 200 min of VPA on one day (this is an adult). I checked the summary plot and plot to check data quality and apparently there was a period of a couple of hours with very high acceleration on the first day. I wonder whether this could be true VPA or might this be something like clipping that is not picked up by the algorithm?

Best wishes,

Nina

[Vincent van Hees]

Hi Nina,

Did you talk to the participant and ask what they were doing during those hours?

If they cannot remember than ask whether they have occupations or hobbies that could also this? For example, endurance sport.

Further, it may be worth inspecting the raw data by using

https://cran.r-project.org/web/packages/read.gt3x/vignettes/batch_loading_gt3x.html

based on the read.gt3x. You could for example load the first 2 days and plot each individual axis of raw data.

If you see clear flat lines near the maximum acceleration values then that would indicate consistent clipping, but if so then it would be surprising that GGIR did not detect it. If you see more fluctuating signals then I think it is more likely that it is either:

• True movement/vibrations
• Malfunction of the accelerometer => Try to trace other recordings with the same device and see whether this patterns occurs in other recordings.

Best,

Vincent

--
You received this message because you are subscribed to the Google Groups "R package GGIR" group.
To unsubscribe from this group and stop receiving emails from it, send an email to RpackageGGIR...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/RpackageGGIR/f77fdcdd-4ce9-49e1-82c0-346fa75117efn%40googlegroups.com.

[Nina Vansweevelt]

Hi Vincent,

I asked the participant and he/she did some gymnastics in the morning every day and some soccer in the evening now and then for one hour, but not that intense. But the measurement is quite some time ago so he/she is not sure if there was something special that day. But the extreme values are in the afternoon from 12-15 o 'clock so that does not seem to match with the usual sports timing of this participant.

I did plot the raw values for these hours and this is the result:

Is this pattern in the z-axis possible?

I also checked other measurements with the same device (csv files and file summary reports) but I don't see this kind of output in these results. But there have only been a couple of measurements with the same device since this one.

Best wishes,

Nina

Op zaterdag 13 juli 2024 om 17:36:44 UTC+2 schreef Vincent van Hees:

[Jairo Hidalgo Migueles]

Hi Nina,

Thank you for the additional information. For me, it is not that surprising that the z axis gets to these high values, but it is that only the z axis seems to be affected. See the picture below, I recorded it during one of my running sessions (including warm-up, 10x500m intervals, and cool down as annotated below). Y axis gets close to 8 g, but you can see there the running and the intervals clearly in the 3 axes. Also, the accelerations go up and down and do not get sustained at high values as in your plot. So, either the participant was doing a different activity that only affected the z axis (I cannot think of any activity lasting a few hours that would produce this pattern in the accelerations), or it might be a malfunctioning of the device.

Best,
Jairo

[Nina Vansweevelt]

Hi Jairo,

Thanks for the explanation! I also cannot think of an activity that would only affect the z axis... Probably a malfunctioning of the device then.

Best wishes,

Nina

Op woensdag 17 juli 2024 om 11:08:55 UTC+2 schreef Jairo Hidalgo Migueles:

[Vincent van Hees]

Hi Jairo and Nina,

In Nina's figure I find it suspicious that the z-acceleration is consistently above 2.5g: Not even a fraction of the data is low.

Jairo's plot does not show this despite his vigorous activity, so I think there is an issue with the accelerometer.

The clipping scoring algorithm does not detect this because it only looks at signals that are consistently close to the maximum value.

Possible next steps:

• We update the clipping detection with an additional check for whether the values at any point cross the 1g line.
• Nina reports back that this happened to ActiGraph and ask them to investigate and provide a clarification.

Best wishes,

Vincent

To view this discussion on the web, visit https://groups.google.com/d/msgid/RpackageGGIR/d6da19cf-534c-468e-90e8-ecf85210b581n%40googlegroups.com.

[Nina Vansweevelt]

Hi Vincent,

Thank you as well for giving your opinion. I took contact with Actigraph to ask for clarification. I checked the other measurements with this device and I did not find such extreme values as the one I sent you. However, I found a few things that make me doubt. Would it be possible to give your opinion on whether the graphs below are possible true accelerations or not? The time between +/- 10:10 and 10:50 is labeled as MPA by GGIR.

Best wishes,

Nina

Op vrijdag 19 juli 2024 om 17:57:10 UTC+2 schreef Vincent van Hees:

[Vincent van Hees]

Hi Nina,

It is hard to see the values. If you zoom in on that window you should be able to see the average acceleration for each axis. If your take the vector magnitude (square root of the sum of the squared values for each axis). If it is far a away from 1 g then that is suspicious. Maybe the sensor are poorly calibrated and GGIR struggles to address the calibration error.

A few more things you could try:

• Run GGIR with do.cal = FALSE to see how results look like if we autocalibration is used.
• Use ActiLife to export the csv and inspect that visually. This to ensure it is not read.gt3x that causes the issue.

Best,

Vincent

To view this discussion on the web, visit https://groups.google.com/d/msgid/RpackageGGIR/0e0dfc60-cfb8-45e3-8b7e-40cee5e62231n%40googlegroups.com.