Hi Brainstorm,
When I ran Morlet wavelet on data of four subjects with the option of "Remove evoked responses ~~" as below, it gave me an error. The error box indicated that the size across data did not match. After running ICA, the number of electrodes saved to the final data are different across participants. When I tried it without the option, it worked though. Could you please let me know whether there is a way to fix this? Thank you so much.
I think I can apply Pre-process > DC offset to replace the option of removing evoked responses in TF analysis. In the tutorial, it was explained that DC offset can be replaced by running a high-pass filter with a very low frequency (for instance 0.3Hz). Though I am wondering about the logic behind it. Could you please let me know why DC offset could be replace with specific filtering? Thank you so much.
When I ran Morlet wavelet on data of four subjects with the option of "Remove evoked responses ~~" as below, it gave me an error. The error box indicated that the size across data did not match.
With the online averager ot the TF processes, you can't average files that have different dimensions together. If you select the option "Remove evoked response from each trial", it first averages the recordings, subtracts the average to all the input files and then proceed to the TF computation. As it needs to average the files, it requires all the files in input to have the same dimensions. One solution is to mark as bad all the electrodes that are bad in at least one file, or physically remove them (process "Standardize > Uniform list of channels").
Note that this option "Remove evoked response" was designed to remove the ERP from the list of trials, in order to bring the signals to a more stationary state, to help with the evaluation of the frequency contents. See: https://neuroimage.usc.edu/brainstorm/Tutorials/TimeFrequency#MEG_recordings:_Single_trials
But this is not the case in your example: these are 4 file from 4 different subjects. I don't think you should use this option here. And in general, you should not compute the TF on already averaged ERPs, but on single trials.
I think I can apply Pre-process > DC offset to replace the option of removing evoked responses in TF analysis.
Removing the average of multiple trials and removing the average of the baseline of one single file are two very different operations, with very different use cases. Please get back to following the introduction tutorials if you are not familiar with what "Remove DC offset" is.
Could you please let me know why DC offset could be replace with specific filtering?
Hi Francois,
Thank you so much for detailed explanation. I also noticed that I confused DC offset with the option of removing evoked responses. The research design that I am working with it unique so that only one stimuli was presented per subject so I realized that I did not need to remove evoked responses. As for the DC offset, I noticed that I could do that by running it as part of pre-processing (under pre-process) or as part of standardizing process (DC offset correction under baseline correction). Would the two return different results? If my goal is to run TF analysis, I think I should run it for normalization. Could you please let me know what you think? Thank you so much.
As for the DC offset, I noticed that I could do that by running it as part of pre-processing (under pre-process) or as part of standardizing process (DC offset correction under baseline correction). Would the two return different results?
No, the two lead to the same results.
In terms of workflow, the process "Pre-process > Remove DC offset" is better indicated for "cleaning" imported epochs, for readability of the processing pipeline. When using the interface to import epochs, this is process that is called when selecting the option to remove the DC offset:
https://neuroimage.usc.edu/brainstorm/Tutorials/Epoching
If my goal is to run TF analysis, I think I should run it for normalization.
The DC level should very little impact on the TF estimation. The average of the signal corresponds to the component at 0Hz of your signal, which is not part of your TF analysis. Try it and see by yourself.
This pre-processing step has much more importance when computing an ERP in time domain.
Hi Francois,
I see! Thank you so much for your clarifications! Thanks!
