Sorry for what may be a naive question (or perhaps inappropriate for this forum), but I am using BST for source localization of some resting state EEG data. My goal is to use some methods of leakage correction from the literature on this data. However, one such method (https://www.sciencedirect.com/science/article/pii/S1053811915002670) requires full rank from my understanding.
Using both MNE and LCMV beamforming for source reconstruction, I then extract the data from each participant using the scouts from the DK atlas. However, inspection of each participant's resulting matrix shows them to be varying levels of rank deficient.
Is this to be expected? Or is something wrong with how I am doing my source reconstruction?
Using both MNE and LCMV beamforming for source reconstruction, I then extract the data from each participant using the scouts from the DK atlas. However, inspection of each participant's resulting matrix shows them to be varying levels of rank deficient.
There is no particular reason for which you could expect your scouts signals matrix to have a rank equal to the number of scouts.
Many steps of preprocessing and the source estimation process reduce the rank of the data. The signals obtained with MNE/LCMV at all the sources are heavily correlated, averaging them by groups of a few tens or hundreds may have unpredictable effects on the final rank of your matrix.
If there is leakage between time series as columns/rows of an ROI matrix, then this latter cannot be full rank. I encourage you to contact the authors of that paper to clarify this aspect.