I am ultimately interested in calculating the amplitude envelope correlation (AEC) using scouts derived from the 100 parcels Shaefer atlas.
Because I am using the MNI template for all subjects, as suggested in the tutorials, I am using unconstrained sources when performing source localization (which I am doing with MNE).
At this point, I see a couple of options, and was wondering if my interpretation of them correct.
Four potential options I see for AEC calculation are:
Using the AEC function, apply the scout function "Mean" to the unconstrained source map
Using the AEC function, apply the scout function "PCA" to the unconstrained source map
First use "Unconstrained to flat map", then using the AEC function, and apply the scout function "Mean"
First use "Unconstrained to flat map", then use the AEC function, and apply the scout function "PCA"
I am trying to understand how these 4 options differ in certain regards, and if my interpretation is correct.
For each scout, take the mean value of each of the 3 dipole orientation. This would provide 3 values for each scout. Then (based on the thread linked above), calculate " the connectivity measure for each orientation and each scout (3x3=9 values for each pair of scouts), then get the maximum measure for each pair (max of 9 values)". This is how one value is provided.
For each scout, get the first mode of the PCA decomposition of all the dipoles. This would provide 3 values/signals for each scout. Then, calculate " the connectivity measure for each orientation and each scout (3x3=9 values for each pair of scouts), then get the maximum measure for each pair (max of 9 values)". This is how one value is provided.
First, for each vertex, get the first mode of the PCA decomposition of all 3 dipoles. This provides a single value instead of 3 at each vertex. Then, for each scout, take the mean value of all single dimension dipoles, and then compute AEC with these scouts.
First, for each vertex, get the first mode of the PCA decomposition of all 3 dipoles. This provides a single value instead of 3 at each vertex. Then, for each scout, take again the first mode of the PCA decomposition of all single dimension dipoles, and then compute AEC with these scouts.
Is this interpretation correct? Furthermore, is there a recommended option as to how to calculate connectivity when working with unconstrained sources?
There is indeed a requirement for dimensionality reduction when dealing with sources with unconstrained orientation. The steps described in #3 and #4 above seem reasonable options to me. Another important aspect is whether the Scout Function is applied before or after computing the connectivity measure. This is an option offered by all BST connectivity processes.
IMO, after transforming the unconstrained volume to "flap maps", it would be best to derive all pairwise connectivity measures between all possible source pairs across regions and then compute their average. If you apply the Scout Function first, the mean or PCA of possibly many time series within each region is first computed, before the connectivity measure is applied. This second option is obvisouly faster computationally but may not be necessarily the most sensitive.
Pragmatic considerations based on your computing resources will help you take the best possible decision in that respect.
Apologizes to bump, but I have a question that I haven't come across a clear answer for.
In the case of EEG source reconstruction (with MNE and current density map for method and measure respectively) , when one does have the individual MRIs, what would be (if there are any) the benefits of using unconstrained sources over constrained? One is the smoothness provided by unconstrained, which is more appropriate for EEG from my understanding (due to the limited spatial resolution we can get by it). But I was wondering more from a physiological "ground truth perspective".
I guess my question is: in the case of EEG, when using individual MRIs, are there any benefits/reasons to use unconstrained over constrained sources?
EEG scalp potentials are very smooth, due to a combined effect of the skull and volume currents.
You cannot expect to localize sources with a a precision higher than a few centimeters. The activity of one EEG source is often reconstructed over large patches of cortex. Using dipole orientations constrained to the normal of the cortex often give very patchy source maps, with disconnected peaks over several gyrii - ugly-looking and difficult to interpret, but easier to process.
Unconstrained maps give smoother maps, easier to read. But more complicated to process afterwards.
It mostly depends on your feeling, your capacity to understand how to process one or the other, the figures you expect to produce...
At the end, you should obtain very similar results with the two approaches.