Laplacian filter

Hi Mike.

I have some issues with applying the laplacian filter (PerrinX). For starters, I use 512 Hz EEG data (BioSemi). My experiment consists of a reaction time task involving various different possible responses (i.e., coordination patterns).
I bandpass filter the data (1 - 35 Hz), clean the continuous data and remove bad channels (using Cleanrawdata and ASR), run ICA, manually remove artefact components and epoch the data (-1 to 2s) (with baseline removal of -200 to 0ms). Then, I remove bad epochs. Finally, I apply the laplacian filter.

When I create time-frequency plots prior to laplacian filtering, these plots are different across channels (as expected). However, after laplacian filtering, my time-frequency plots become nearly identical across different channels (e.g., time frequency plots of O2, C3 and C4 look identical). The topoplots after laplacian seem fine. This is the case in all my data.

Any idea as to why this is the case? All help would be greatly appreciated since I want to investigate connectivity and ersp and would like to use Laplacian for this.

Kind regards
Sybren

I don’t know, but it sounds like a bug in the code. If you plot some example time series from different channels, are they also nearly identical?

Thanks for the response! Yeah, they also look very similar. Intrestingly, the topoplots look alright as aforementioned.

The IC’s after laplacian look somewhat odd as well (compared to spatially unfiltered data). I use data in the eeglab format and plot the data in the eeglab GUI as well (there seem to be some new ICs that are either completely red or blue in all subjects). Regarding the laplacian, I just copy the laplacian output over the eeg data output. I use the eeglab X, Y and Z coordinates as input. Could any of these steps cause the bug? If not, is it maybe the fact that I use the laplacian filtered data in the eeglab GUI?

You shouldn’t apply the IC weights to the Laplacian data. That could be causing the problem. So, clean the data with ICA, then apply the Laplacian, then analyze the data.

yeah I cleaned it with ICA prior to Laplacian. I just wanted to redo ICA to see if I could find the bug. The first screenshot is after laplacian the second is before laplacian. Could it be attributed to noisy data (judging from these screenshots)?

Thanks for the respons!