Dear Francois:
I am going to do both event-related analysis and connectivity analysis on the same BST protocol file and need to find an epoch length that will work for me to run both event-related and connectivity analyses with no issues. Based on the valuable information described in this tutorial: https://neuroimage.usc.edu/brainstorm/Tutorials/Epoching#Epoch_length
It says "The minimum duration between two stimuli (minimal ISI) defines the maximum length you can consider analyzing after the stimulus." My minimal ISI is 1000ms, so does that mean that I can only epoch a maximum of 1000ms including the baseline, like -200ms-800ms or -300ms-700ms, etc?
My frequency filters were applied on the continuous data before the data were epoched, so I assume it wouldn't create the transient effect on my epoched data, right?
But the tutorial also says the normalization step for filtered source averages would create some edge effects for the baseline used and therefore it recommends using longer baseline epoch window. I am not sure what it means by "filtered source averages", does it mean the event-related source analysis averages? I don't remember using any filters on that step? The only filtering step I did so far is on my continuous data using 1-50 Hz high/Low frequency filters.
So should I use epochs like: -300ms-700ms, -400ms - 600ms, or I should extend my epoch to over 1000ms regardless, like -1000ms to 1000ms?
It says "The minimum duration between two stimuli (minimal ISI) defines the maximum length you can consider analyzing after the stimulus." My minimal ISI is 1000ms, so does that mean that I can only epoch a maximum of 1000ms including the baseline, like -200ms-800ms or -300ms-700ms, etc?
It can be a bit longer if you want to do some time-frequency analysis, but then you need to be aware that there might be some overlap between the end of an epoch and the baseline of the next one...
My frequency filters were applied on the continuous data before the data were epoched, so I assume it wouldn't create the transient effect on my epoched data, right?
No, but if you applied frequency filters, it creates a form of temporal smoothing of the signals.
But the tutorial also says the normalization step for filtered source averages would create some edge effects for the baseline used and therefore it recommends using longer baseline epoch window. I am not sure what it means by "filtered source averages", does it mean the event-related source analysis averages?
This is only when you filter the source maps after source estimation.
If you filtered the continuous files from the beginning, you don't need to (and should not) apply any additional filter.
So should I use epochs like: -300ms-700ms, -400ms - 600ms, or I should extend my epoch to over 1000ms regardless, like -1000ms to 1000ms?
You have understood all the recommendations and limitations, this is up to you to design your analysis pipeline. If you're not sure for a parameter, try different options and see what you obtain...
Dear Francois:
Since I won't be focusing on any time-frequency analysis but will just focus primarily on my connectivity analysis and event-related analysis protocols, and that I don't apply any filtering other than the beginning one time filtering on the continuous data only, I assume using an epoch of -200ms to 800ms would work fine me for both my event-related and connectivity analyses pipeline and it wouldn't give me trouble for the connectivity part of the analysis, right? Thank you so much for your most helpful information for me to make this decision.
So I assume once I filtered the continuous data in the very beginning, for the rest of the steps whenever I see an option of checking the "Remove DC Offset", I shouldn't check it and should just leave it unchecked, right? These places are: during fixing the stimulus delay (the 'Detect analog triggers" step), and during importing the epochs.
But I did the fixing of my stimulus delay step first thing in the begging, before I applied the 1-50Hz High/Low pass filters on the continuous data, so I had the "Remove DC offset" box checked in the fixing stimulus delay ('Detect analog triggers'). Is that alright?
I know for sure when importing the epochs later, I should leave the "Remove DC Offset" unchecked.
No I didn't. So the first thing after I import my raw MEG data (using the 'review' function actually), I followed the tutorial of fixing the stimulus delay by using the Process 1 Box's Event > Detect analog triggers and I use my stimulus channel (UTRG001) as the reference for the actual timing, and I "checked" the Remove DC offset box to generate a new event trigger with correct onset timing. After this step, I applied the 1-50Hz High/Low frequency filters. Does this order sound correct, and is my initial removing of the DC offset fine for the next step's frequency filtering?
I was also wondering if I had done the processing differenty, say if I did the frequency filtering as my very first step followed by fixing stimulus delay, should I leave the Remove DC offset unchecked when I fixed my stimulus delay using the Detect analog triggers?
Now I got it! And I agree that the analog triggers detection works better with the DC offset removal so there is no need to uncheck that box! I earlier thought when I applied for 1-50Hz band-pass on all my channels, it also affected the stimulus channel (the one I used for my analog triggers detection to fix my stimulus delays), which is not the case! Thank you so much for making sense for all this!!