My project involves looking at phase estimation via the Hilbert transform in publicly available datasets. I’ve seen many papers use the Laplacian montage, so I was considering that to get spatially filtered data.
My issue is that I want to center on POz, but many of the datasets are missing that electrode. Here are some options I came up with
- Still use POz but with interpolated data. Seems weird because POz is my electrode of interest and I’m not looking at ‘raw’ data then
- Use Pz, and a Laplacian montage with large spacing (center Pz, neighbors Cz, P3, P4, Oz). I’m concerned that the spacing between the electrodes is too large and the filtering wouldn’t be done correctly
- Use an average re-referencing scheme
I’m leaning towards the second option but wanted to hear a second opinion.
Hi Brian. Interesting question. The original papers on the EEG Laplacian were done with only 16 electrodes, although that was more for methods development than application. So of course you can compute the Laplacian; the question is whether it’s really advantageous with so few electrodes.
I’m honestly not sure what I would recommend. Do you have a dataset with more electrodes? One thing you could try is applying the Laplacian on 64 channel data, and then selecting the same number of electrodes that you have and applying the Laplacian, and seeing if the two sets of results really differ.
All that said, I don’t think there is anything theoretical reason not to use the second option; my hesitation is only that I don’t have experience working with spatial filters on low-density EEG.
I do have 64 channel data so I’ll try comparing the two approaches by computing the circular correlation of phases between the high and low resolution montages. Thank you Mike!