Hello everyone:
I was wondering about narrow band filtering the EEG (ALPHA, BETA, THETA) and doing a source analysis with sLORETA in a particular set of ROIs.
Filtering before (EEG electrodes activity) or after (ROI sloreta activity) gives different ressults.
At a first sight, it seems to me maybe more correct narrowband filtering the sLORETA output (doing a wide band source analyisis and then filter the ROIs signal) but i couldn´t find any explicit paper to look at this without going deep into the formulation of LORETA.
Here, for example I supose they filtered before but I´m not sure:
Clemens, B., Bessenyei, M., Fekete, I., Puskás, S., Kondákor, I., Tóth, M., & Hollódy, K. (2010). Theta EEG source localization using LORETA in partial epilepsy patients with and without medication. Clinical Neurophysiology, 121(6), 848-858. https://doi.org/10.1016/j.clinph.2010.01.020
I want to obtain a filtered time series (ALPHA, BETA, THETA) in each ROI, not the cross spectrum.
Thank you very much in advance!
P Diaz.
My idea is, filtering the signal before is much better for source analysis with some methods like sLORETA. Why?
The relationship between source dynamics and observed dynamics, at the level of EEG sensors is linear. I you were to solve this linear problem (estimating the source activity) with a simple pseudo-inverse (eLORETA) usually denominated linear solvers, filtering before and after would be completely equivalent. But in case you use any method that performs statistical learning (sLORETA) it will not.
The estimation of EEG sources is a severely ill-conditioned Inverse Problem (meaning it the space of possible solutions is huge). Any method in the state-of-the-art maximizes a cost function representing probabilities of the source configurations and dynamics. These are biased to use the simplest configuration of sources that achieves to explain the observations (Occam's Razor). Meaning, performance is better as simpler are the features present in the data: like those of narrow band signals or signals ICA/PCA components. Pablo Aleixandre Díaz Brage
Thank you very much for your answer.
I agree with you about filtering before but, If I filter before sLORETA the signal I get changes its spectrum.
If I calculate with ALPHA (8-12 Hz) band after sLORETA localization, in my ROIs I get a (16-18 Hz) with a strong DC component...
Do you recommend to try eLORETA instead?
Thanks again!
eLORETA is said to have the greatest performance in the LORETAs family, some say is the best M/EEG source imaging method. See final part of page 7 in Pascual's paper https://arxiv.org/ftp/arxiv/papers/0710/0710.3341.pdf. Although way more advanced methodological research is nowadays in progress, mostly on Beamformer or Bayesian learning methods.
About the filtering, the spectral composition of the source signal estimated with any method will be similar to the input signal. Therefore, if filtered before sLORETA this will we reflected in the spectra of the source signal. Basically, you can repeat this process for every band (delta, theta, alpha ...) and get filtered source signal for each one of them. The standard procedure is to do statistics with these filtered signals in every band, mostly on their variances or connectivity metrics. To retrieve the full spectral properties you need to add up together the filtered signal for all bands, but is a bit problematic because of possible border artifacts of the bandpass filters.
Thanks again, Deirel.
Sorry if it is a stupid question but, is there any (free) software to compute eLORETA like sLORETA?
I have the LORETA key software v20190617 but I didn´t see any option to do this.
There is an implementation by Guido Nolte ( https://scholar.google.de/citations?user=iDXuEAoAAAAJ&hl=en ) that the FieldTrip team (http://www.fieldtriptoolbox.org/) uses and I have used as well. It can stand alone easily in any package, is called "mkfilt_eloreta".
[A,Wout] = mkfilt_eloreta(L,regu);
L - the MEG/EEG lead field obtained by any means
reg - regularization parameter that can be adjusted with cross-validation for example
A - a matrix that you multiply by the data to obtain the source activity
W - represent variances
Here I attach the code, you can get the whole FieldTrip package from (https://github.com/fieldtrip/fieldtrip).
BC-VARETA team ( https://github.com/CCC-members/BC-VARETA_Toolbox) has it implemented with some further guarantees, cross-validation, deformed graph laplacian, field rotational invariance. But is merged to spectral analysis of the signals, time domain implementation in the package is still pending.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Electrodes