- MNE version: 1.8.0
- operating system: macOS 14.4.1
Hi,
I am using eLORETA to derive the source time course of EEG epoch data (computed for every trial with the SNR parameter set to 1).
My goal is to perform multiple linear regression on parcels of the source time courses. The tutorial The role of dipole orientations in distributed source localization, warns about performing regression across trials on source data if only the magnitude of the dipoles are extracted instead of the 3D representation of the dipole (noise won’t be averaged out).
Precisely, linear regression models are fitted to every parcel and time point source activity (with dipole orientation). In this setting, would a good approach (to recover scalar beta estimates), be to perform PCA over time for the beta parameters of each ROI and project the betas on the first principle component?
Elsewhere, I am also motivated to do a connectivity analysis from source data. However, the same in the tutorial’s warning remarks that frequency and phase may be distorted for distributed source reconstruction that only extracts dipole magnitude. Does there exist a principled way of then computing phased-connectivity measures or fitting vector autoregression models on source activity with dipole orientation?
Thanks for the guidance!