We are planning to perform a PCA on selected ERPs using your toolkit on a consolidated EEG data from different labs. The data were collected using different recording systems, resulting in different sampling rates (500/512), electrode numbers (64/32) and electrode configurations. I was wondering what would be the best approach go with the PCA? Should I first standardize all the recordings (i.e., selecting only common channels, downsample the data etc.) before proceeding to performing the PCA or is there another way to deal with this?
Thank you for your help
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Yes, definitely standardize the recordings first. You can use the Edit function to do both. The Samples pane allows you to resample as desired and the Channel Coordinates button of the Overview pane allows you to interpolate the channels from one montage to another. Then you'll be ready to go!
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Thank you so much for your response,
I do have a small follow up question: 9 of the labs record with 64 channels, 4 can only record 32. Is it recommended to just use "similar" electrodes and then standardize them, or is it acceptable to interpolate the missing channels for the less dense setup?
Thanks for your help
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
That'd be a "depends" I'm afraid. So for example, a channel interpolation is only as good as the surrounding channels. Channels on the edge of the montage are more likely to have issues because of the missing information about what is outside of the montage. Choosing only channels in common avoids the interpolation issues but will rob the PCA of much of its power as temporal PCA relies heavily on the differences in topography across the channels to distinguish between ERP components, so the less channels, the less effective it will be. On the other hand, it's not a matter of how many channels so much as the informational content of each channel with respect to distinguishing between ERP components. So the most valuable channels are those that have one ERP component but not another. So you'll have to decide what your goals are for the experiment and then do some trial-and-error to see what works best for those goals. Sorry I can't give you a better answer!
Joe
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
We are planning to perform a PCA on selected ERPs using your toolkit on a consolidated EEG data from different labs. The data were collected using different recording systems, resulting in different sampling rates (500/512), electrode numbers (64/32) and electrode configurations. I was wondering what would be the best approach go with the PCA? Should I first standardize all the recordings (i.e., selecting only common channels, downsample the data etc.) before proceeding to performing the PCA or is there another way to deal with this?
Thank you for your help
Yes, definitely standardize the recordings first. You can use the Edit function to do both. The Samples pane allows you to resample as desired and the Channel Coordinates button of the Overview pane allows you to interpolate the channels from one montage to another. Then you'll be ready to go!
Thank you so much for your response,
I do have a small follow up question: 9 of the labs record with 64 channels, 4 can only record 32. Is it recommended to just use "similar" electrodes and then standardize them, or is it acceptable to interpolate the missing channels for the less dense setup?
Thanks for your help
That'd be a "depends" I'm afraid. So for example, a channel interpolation is only as good as the surrounding channels. Channels on the edge of the montage are more likely to have issues because of the missing information about what is outside of the montage. Choosing only channels in common avoids the interpolation issues but will rob the PCA of much of its power as temporal PCA relies heavily on the differences in topography across the channels to distinguish between ERP components, so the less channels, the less effective it will be. On the other hand, it's not a matter of how many channels so much as the informational content of each channel with respect to distinguishing between ERP components. So the most valuable channels are those that have one ERP component but not another. So you'll have to decide what your goals are for the experiment and then do some trial-and-error to see what works best for those goals. Sorry I can't give you a better answer!
Joe