Hi everyone,
I am currently analyzing MEG data from pediatric subjects and I would like to ask for advice regarding head motion thresholds for subject exclusion.
For adult datasets, a threshold of around 5 mm is often considered acceptable. However, in my case, almost all children exceed this threshold.
Do you have any recommendations on appropriate motion thresholds for pediatric data?
Hello
In such a case, the idea may not be to exclude runs or subjects but to analyze the best way you can and get as much out of the data you have. The idea is then to try to get enough “stable” segments with good enough data. I’d suggest to still start with 5 mm as the threshold to split your data, but if it’s splitting or rejecting too much, try 1 cm. You can also shorten the “minimum duration” parameter to avoid rejecting too much. How small that can be however depends on the methods you use. E.g. beamforming will need enough to properly estimate the data covariance matrix, whereas minimum norm doesn’t care and therefore can be split in smaller chunks. But in the end, you have to pick values that gives you enough data to do the analyses you want. Of course, movements will often be accompanied by artefacts which may also force you to reject some segments.
Cheers,
Marc
Thank you very much for your suggestions, Marc.
So, just to make sure I understood correctly: in pediatric MEG data, using 1 cm as a motion threshold can still be acceptable if 5 mm would lead to rejecting too much data, and the main goal is to retain enough stable segments for the planned analysis. Is that correct, or would you consider 1 cm already too permissive?
I also wanted to ask about how to handle the good segments in practice. Suppose there are several bad segments interspersed with good ones during the recording, and the good segments correspond to relatively stable but slightly different head positions. Would you recommend keeping the data as is and simply marking the bad segments, or actually splitting the recording into separate files / epochs based on the stable segments?
I am asking because in the past I tried splitting the data for a functional connectivity analysis, but then I had trouble merging the different segments afterward in Brainstorm, which gave me errors.
Thanks again for your help.
There’s no clear answer to your first question. It depends on what you want to do with the data and whether you have the luxury of not using it; e.g. if it’s a retroactive research project where you can pick which subjects to use vs a clinical case where you need to extract as much info as you can from what you have.
In general you want to have the correct head position for each segment. This however means you can’t simply stick the segments back together, but you have to analyze them separately and “regroup” the results later when it’s possible. The specifics again depends on what you’re doing with the data. One classic example is an evoked response. Just like in the main series of tutorials there are two runs that are analyzed separately until we get at least to source space, then the results can be averaged across runs. It would be similar with segments with different head positions. The sensors are not in the same place, so you have to wait until your in source space before you can put the data together again. And you can’t just concatenate segments together if you’re doing any type of frequency analysis (including many connectivity methods), because the jumps in the time series where the segments were joined could cause false results. So often you’ll need to keep them separate and average the results at a later stage.
If you think it’s valid to merge segments in a way that Brainstorm won’t let you, then I’d suggest to post another question about that specifically.
Cheers