Removing evoked response for AEC functional connectivity analyses

Dear Brainstorrm community,

I have some ERP data. I am interested in calculating the functional connectivity between a scout and the remaining regions using amplitude envelope correlation.

Reading up on the idea of removing the evoked response from event-related data when wanting to calculate connectivity (such as the Wang et al. 2008 ' Estimating Granger causality after stimulus onset: A cautionary note'), it seems as though there are mixed thoughts as to doing so.

Could anyone suggest as to what the current opinion is on this process? With regards to AEC, I could only find resting state papers, so there's no paper I can reference to with regards to task-based connectivity and the removal of the evoked response.

I plan on trying both and comparing the results, but without much precedence, I am thinking going forward with this analysis, I may stick with not removing the evoked response.

Any input would be greatly appreciated.

I don't believe there is consensus on the question, because the neural mechanisms involved are still unknown. Be aware of edge effects due to using band-pass filters when applying AEC: they can confound the outcome when included in the AEC time window. This can be critical is you apply AEC on short ERP segments, such as trials. You may want to consider expanding the duration of ERP epochs, if possible, or better: bandpass filter your raw data in the frequency bands where you want to compute AEC and then re-epoch the data into trials -- more cumbersome but less prone to method artifacts.

Keep us posted!

@Sylvain, it's been some time since you addressed @pdhami's question.

I'm eager to hear your current thoughts on the matter. Additionally, I wonder if @pdhami has since applied the bandpass filter to the raw data in the specified frequency bands and would be willing to share his experiences.

My research focuses on studying brain connectivity between the insula and cingulate cortex in individuals with chronic pain. Each participant received an electrical stimulation, resulting in 10s epochs [-3s, +7s]. I intend to analyze the alpha, beta, and gamma bands at the very least. I'm somewhat hesitant about the lower frequencies, mainly due to the stimulus artifact predominantly observed in these frequencies.

However, I'm now at the stage where I need to apply a bandpass filter during preprocessing. I previously used this data for a different analysis and filtered it between 1 and 200 Hz. I'm now uncertain whether to set my filter limits to the specific frequency bands of interest right away or to explore the frequency bands with the Hilbert transformation as demonstrated in the connectivity tutorial.

Thanks in advance! Really appreciate all the effort you guys are making in the tutorials and the answers on the forum.

My recommendation is to always start from your original research question and choose the methods and their parameters accordingly.The long epochs in your study seem sufficient to bandpass filter the data in the alpha and faster bands at the trial level. Of course, if a transient artefact is present at the moment o the stem, there may be ringing effects in some or all bands, esp. if your filter passbands are very narrow.