I am encountering several key issues while using sLORETA for EEG source analysis, and I have doubts about the correctness of my methodology. I sincerely request your guidance:
Huge Disparity in Voxel Count : The voxel count I obtain is very large (e.g., several hundred thousand), whereas the classic literature I reference uses only around 6000 voxels. Neither of us has individual MRI data; we both use the same standard brain template (e.g., MNI152). My questions are:
What is the most likely reason for this order-of-magnitude difference? (e.g., different default software resolution settings, different ROI definition methods?)
Given the use of a standard template without individual MRIs, what is a common or reasonable range for the voxel count?
Feasibility of Large-Scale Network Output : I aim to calculate the time series for 78 brain regions (e.g., using AAL or Destrieux atlas) based on the source results, and further compute a 78x78 Pearson correlation matrix (i.e., functional connectivity). Based on information I found, if I choose to output "all results" or the time series for each voxel, the estimated data size for a single subject could be as high as tens of GB, which seems impractical for storage and computation.
Is this an expected outcome, or is there an error in my operational workflow?
To obtain the average time series for the 78 brain regions (for calculating functional connectivity), what should be the correct and storage-efficient output/processing steps ? Should I perform spatial dimensionality reduction first (e.g., extract the mean of voxels within each ROI) before output, rather than directly outputting whole-brain voxel data?
Verification of Methodological Workflow : Based on the above issues, is there a fundamental error in my overall process? What are the typical steps for a standard sLORETA analysis without individual MRI data, aimed at obtaining functional connectivity at the brain region level?
The difference is in how the head-model is computed. This is to say:
What is the source space, and how are dipole location defined.
E.g. in Brainstorm two common source spaces are cortical surface and brain volume:
Cortical surface: Dipoles are placed on the vertices of the cortical surface
Brain volume: Dipoles are placed on the point of a 3D grid that covers the brain volume
(Since you are talking on voxels, I guess you are working with a volume head model)
In sLORETA the volume head model is not the entire brain, but constrained to the gray matter with some anatomical constrains, this leads to 6340 source locations ("voxels"). Check the full information on this head model in the sLORETA technical details.
As mentioned above, even if the same anatomical template is used the number of sources ("voxels") would depend on how the head model is computed.
This is well documented in these Brainstorm tutorials. Please give it a read, and let us know if something remains unclear:
In short, yes there is need of spatial dimension reduction. However, this can happen before or after computing the connectivity metric. As you can see, if the reduction happens after, the processing will require a huge amount of RAM.