sLORETA Source Analysis

Discussions
1

Dear experts,

I am trying to replicate the EEG source localization and functional connectivity pipeline described in the supplementary information of the paper “Creative experience and brain clocks” (2025). Specifically, I want to follow exactly the sLORETA procedure outlined on Page 2, section 1.3 :

“The standardized current density maps were generated using a three‑layer concentric sphere head model across a predefined source space of 6,242 voxels (each voxel being 5 mm³) based on the MNI average brain. Electrode configurations were mapped onto the MNI152 scalp coordinates.”

I am currently stuck at the step of choosing/computing the head model in my software (I am using [Brainstorm / EEGLAB / sLORETA standalone – please specify which you recommend]) and would be very grateful for concrete, step‑by‑step guidance.

My specific questions:

  1. Head model selection
  • The paper uses a three‑layer concentric sphere head model . How do I set this up to exactly match the MNI152 template? Should I use the default spherical model with three layers (scalp, skull, brain) and fit it to the MNI152 scalp positions?
  • In Brainstorm/EEGLAB, there are several options (e.g., overlapping spheres, boundary element model, finite element model). Which one corresponds to the “three‑layer concentric sphere” used in the paper?
  1. Electrode registration to MNI152
  • The paper says “electrode configurations were mapped onto the MNI152 scalp coordinates” . How is this done in practice? Do I use the standard 10‑20 system locations that are already available in MNI space (e.g., from the standard BEM meshes), or do I need to manually warp my own electrode positions?
  1. Source space of 6,242 voxels (5 mm³)
  • How can I obtain exactly 6,242 voxels at 5 mm³ resolution restricted to cortical gray matter and hippocampus (as mentioned)? The default MNI152 volume at 5 mm usually contains more voxels. Do I need to mask it with a specific atlas or restrict it using the AAL cortical mask?
  • Is there a standard template file (e.g., from the sLORETA software) that provides this predefined source space, and if so, how can I import it into other toolboxes?
  1. Regularization and SNR
  • The paper mentions: “A regularization method, set with a signal‑to‑noise ratio of 1, was utilized to derive the sLORETA transformation matrix” . Where exactly is this SNR parameter set? In the inverse modelling step?
  1. From current density to ROI time series
  • After computing sLORETA for each time point, they “averaged the voxels within each AAL region to produce a single mean time series” . Do I first export the full voxel‑wise current density time series (which would be huge) and then average, or is there a more memory‑efficient way (e.g., directly computing ROI averages via a weighting matrix)?

I have read the sLORETA documentation and various tutorials, but I cannot find a description that reproduces this exact 6,242‑voxel / 5 mm³ / three‑layer sphere / MNI152 combination. I suspect that many subtle settings (e.g., head model type, electrode projection, source grid definition) critically affect the final result, and I want to avoid hidden deviations.

What I have tried so far:

  • In Brainstorm, I used the ICBM152 template, generated a 5 mm grid, and applied the AAL cortical mask – this gives a different number of voxels (∼15,000).
  • In the official sLORETA software, the default voxel count is 6,239 (or 6,242?) – but I am not sure how to replicate that outside the standalone program.

Could you please provide a detailed, step‑by‑step workflow (menu selections, parameters, any necessary code snippets) to achieve exactly the pipeline described in the paper? Any clarification on how the authors technically implemented this would be immensely helpful.

Thank you very much for your time and expertise!

2

Dear Yuting,

It is recommended to use sLORETA software to replicate the experiment, as some of the operations are hard to achieve using other toolboxes.

The paper uses a three‑layer concentric sphere head model. How do I set this up to exactly match the MNI152 template? Should I use the default spherical model with three layers (scalp, skull, brain) and fit it to the MNI152 scalp positions?

You don’t have to do that, as the three‑layer concentric sphere head model was obtained from the template. The template and the head model are registered intrinsically.

In Brainstorm/EEGLAB, there are several options (e.g., overlapping spheres, boundary element model, finite element model). Which one corresponds to the “three‑layer concentric sphere” used in the paper?

None of them corresponds.

overlapping spheres: each magnetometer fits a sphere, this method is for MEG source reconstruction.

BEM and FEM: these are used for EEG source reconstruction.

Actually, the concentric sphere head model will result in some source localization bias, as this model is too simple. BEM and FEM are recommended but they are time-consuming.

The paper says “electrode configurations were mapped onto the MNI152 scalp coordinates” . How is this done in practice? Do I use the standard 10‑20 system locations that are already available in MNI space (e.g., from the standard BEM meshes), or do I need to manually warp my own electrode positions?

Just use the coordinates in the template, as you don’t have individual MRI for registration.

How can I obtain exactly 6,242 voxels at 5 mm³ resolution restricted to cortical gray matter and hippocampus (as mentioned)? The default MNI152 volume at 5 mm usually contains more voxels. Do I need to mask it with a specific atlas or restrict it using the AAL cortical mask?

Apply a gray matter mask that includes cortex + hippocampus, typically derived from the AAL atlas, excluding subcortical nuclei but including hippocampal regions (AAL labels 37–40: Hippocampus_L/R, Parahippocampal_L/R). Or use the gray matter probability map from SPM or MNI ICBM152, thresholded and manually augmented to retain hippocampus. However, I’m not sure this procedure will obtain your desired result. It is recommended to directly use the template in sLORETA, they provided the voxel coordinates, if we can replace the coordinates in brainstorm template with that, then we can use that template. However, the best way is to use sLORETA software.

The paper mentions: “A regularization method, set with a signal‑to‑noise ratio of 1, was utilized to derive the sLORETA transformation matrix” . Where exactly is this SNR parameter set? In the inverse modelling step?

run→sources→compute sources [2018]

QQ_1770871226673
QQ_17708712266731175×754 37.9 KB

After computing sLORETA for each time point, they “averaged the voxels within each AAL region to produce a single mean time series” . Do I first export the full voxel‑wise current density time series (which would be huge) and then average, or is there a more memory‑efficient way (e.g., directly computing ROI averages via a weighting matrix)?

You don’t have to obtain the full result.

run→sources→compute sources [2018]→Kernel only: one per file

Then drag the source result file into process space: run→extract→scout time series, choose the scout you want, set the method to average, then you will get the source result on each ROI.

May be you should try to use sLORETA since I’m not familiar with that software, however, please let me know if you have any problems.

Best regards,

Jiameng