Hi Francois, and Brainstormers
Hope you're doing well,
I'm analysing MEG data with Brainstorm and I have very basic questions regarding the SSP procedure. For EEG processing in EEGlab it is often adviced to eliminate non-stereotyped artefacts from the data to get the most of ICA. In this way artefacts such as eye movements and cardiac are better captured in the first components. I wonder if it is the same for SSP with MEG data. After detection of blink and cardiac should I look into these to be sure they are prototypical so SSP can do a good job. I would need to eliminate blink and cardiac that are not so prototypical (looking for instance the topography). I'm asking this because I have a subject with very weird ECG. So, I wonder if I should select only some ECG events. If this is case should I also pay attention to the polarity for cardiac artefacts?
As an aside question. I came across a discussion in Brainstorm about the order of SSP and filtering, it seems I would need to filter first and then compute the SSP for apply it when importing data. Is this correct? What if after eliminating artefacts such as eye blinks and cardiac I wanted to eliminate other types of artefacts such muscle, for which it would be better to have unfiltered data?
Thank you in advance for any hint,
José Luis
For EEG processing in EEGlab it is often adviced to eliminate non-stereotyped artefacts from the data to get the most of ICA. In this way artefacts such as eye movements and cardiac are better captured in the first components. I wonder if it is the same for SSP with MEG data
The logic is the same, identifying spatial topographies you want to remove from the recordings. ICA ensures additionally that the components are independent in time, and more or less equivalent in amount of power captured by each component. ICA on MEG signals also gives good results, but typically very long to run and difficult to automate (you need to manually pick many components out of 300).
https://neuroimage.usc.edu/brainstorm/Tutorials/ArtifactsSsp
https://neuroimage.usc.edu/brainstorm/Tutorials/Epilepsy#Artifact_cleaning_with_ICA
https://neuroimage.usc.edu/brainstorm/Tutorials/SSPCookbook
I would need to eliminate blink and cardiac that are not so prototypical (looking for instance the topography). I'm asking this because I have a subject with very weird ECG. So, I wonder if I should select only some ECG events.
Select all the ECG events that look alike in one category are try to compute SSP for them. Then make sure 1) one or several components are clearly specific of the artifact, 2) it removes the artifact, 3) it doesn't alter the signal too much away from the artifact.
https://neuroimage.usc.edu/brainstorm/Tutorials/ArtifactsSsp#Evaluate_the_components
You can also right-click on a 2D topography figure showing a specific artifact topography > Snapshot > Save as SSP projector. This would remove that specific topography from the recordings. Do the same checks as for the computed SSP.
If this is case should I also pay attention to the polarity for cardiac artefacts?
No, polarity is not relevant. Blue and red parts can be exchanged, it doesn't change the correction applied to the signal.
it seems I would need to filter first and then compute the SSP for apply it when importing data. Is this correct? What if after eliminating artefacts such as eye blinks and cardiac I wanted to eliminate other types of artefacts such muscle, for which it would be better to have unfiltered data?
The order of the corrections you apply depends on your signals. Both orders could make sense, but if you filter out most of the activity you want to get rid of (low-pass filters would remove most eye movements and high-pass filters most muscle activity), then the SSP algorithm will not work efficiently.
1 Like