In our study, we have two conditions and each condition has two levels (A: A1vs A2, B: B1 vs B2). We interested in the theta band in our study. And we have computed the time-frequency analysis in the single trial, and found a significant interaction between these two conditions. A1B2 has the highest the power than other conditions.
We want to do source-localization analysis to explore the source in a specific frequency band and specific time. As mentioned before, the significant theta power was presented in averaged frequency band (4-8 Hz) and averaged time (300-500ms). I imported the pre-processed data into Brainstorm and segmented the data into 4 conditions. Next, I estimated the source by using the Minimum norm imaging and opt for āunconstrainedā dipole orientation. After this step, each trial has a source map. Then, I do the time-frequency analysis in source space in single trial level. And I choose the options āGroup in frequency bands (4-8 Hz)ā, I understand that this option meets my specific frequency needs and averages this frequency band. Then, the time-frequency source maps were averaged across trials for each condition, and normalized based on Brainstormās implemented zscore transformation relative to the -500 to -200 ms pre-trial baseline. Finally, I used āaverageā option to average the specific time (300-500ms) for each condition. So that's the end of the data processing, right? We get the z-score normalized time-frequency source map of each person on each condition.
The group averaged time-frequency source map on each condition was presented below. It seems that the A1B2 activates more brain regions and more strongly than the other condition.
Next, we used nonparametric cluster based permutation tests to explore the source difference between these conditions. I selected the process2 in brainstorm, and put all subjectsā z-score time-frequency source map file of A1B2 condition into āFile Aā , put one of the other conditionās files in āFile Bā, and run ātestā ā fieldtripļ¼ft_sourcestatisticsā. No significant difference cluster was found between A1B2 condtion and other conditions. Matlab window shows āall values are nullā.
Is this result possible? For example, in our case, we have a significant difference in the Theta power, but nothing in the source-localization result. Is there any omission or error in my operation?
No significant difference cluster was found between A1B2 condtion and other conditions. Matlab window shows āall values are nullā.
Is this result possible? For example, in our case, we have a significant difference in the Theta power, but nothing in the source-localization result. Is there any omission or error in my operation?
We have no experience with the FieldTrip cluster-based permutation tests on sources, I didn't even know it was working when called from Brainstorm (and maybe it doesn't).
Try using the brainstorm non-parametric statistical test with an FDR correction instead.
Thanks for you reply.
I tried brainstorm non-parametric statistical test, but no difference was found.
I wonder if the two operations in the preprocessing will affect the result of the source.
Is it reasonable to use Laplacian transformation before computing source? by the way, the process "Standardize > FieldTrip: ft_scalpcurrentdensity" is a permanent transform data into current source density? (not only for viewing)
Are noisy electrodes an important factor affecting source results? We have some noisy electrodes in our EEG data, we used interpolation method to interpolate noisy channels. Is this a good way to deal with noise electrodes? Or, for noisy electrodes, what should we do before calculating the source?
Another concern was that we used bilateral mastoid as a re-reference during preprocess. These two electrodes were removed after the re-reference. Therefore, the data imported into Brainstorm does not contain these two electrodes. Before computing source, we performed an average reference on the pre-processed data, would the missing two electrodes make this step wrong? Whether the average reference is a must for computing source?
Is it reasonable to use Laplacian transformation before computing source?
No, it is not. Don't do it.
the process "Standardize > FieldTrip: ft_scalpcurrentdensity" is a permanent transform data into current source density?
Yes, it is. Don't use it.
Are noisy electrodes an important factor affecting source results?
we used interpolation method to interpolate noisy channels.
Of course.
Mark noisy electrodes as bad, do not replace them with local interpolations. Interpolating doesn't add any information to the model.
Another concern was that we used bilateral mastoid as a re-reference during preprocess. These two electrodes were removed after the re-reference. Therefore, the data imported into Brainstorm does not contain these two electrodes. Before computing source, we performed an average reference on the pre-processed data, would the missing two electrodes make this step wrong? Whether the average reference is a must for computing source?
If you have additional questions, please post them on the FieldTrip mailing list or other FieldTrip support channels.
Could you elaborate on why Laplacian transform should not be applied before computing sources?
This method computes a spatial derivative of the EEG, therefore the output data is not EEG anymore, the physics laws used in the forward models computed in Brainstorm don't apply anymore.
I do have a follow-up question. I first asked the FieldTrip mailing list, but as I'm using the Fieldtrip plug-in within Brainstorm, they thought the question was better directed here.
When I visualized Laplacian spline transformed ERPs within Brainstorm, they were scaled approximately between -150 to 150 (the units were labeled microvolts, but this must be a mislabeling)... Anyway, when I export my ERPs, the values were 10^6 times smaller. Should I rescale them to match what I see when plotting in Brainstorm, or is V/m^2 actually expected to be a very small number? Thank you!
Brainstorm saves everything in international units (meters, Volts, Ampers...)
If you have values in the range of the microVolt (or Ī¼V/m^2), this corresponds to values ~1e-6 V saved on the hard drive.
