Intracranial "distant" sources


I am currently working on an intracranial monitoring case for an epilepsy patient. The biggest drawback with sEEG is the limited sampling area and it seems artificial as we "filter out" low voltage and evaluate only for large signals (the result of LFP) of the seizure area. My questions. 1/is there a difference between the scalp and intracranial EEGs when it comes to detecting "distant" EEG signals like spikes and seizure rhythms. 2/If the answer is "no difference", how do I evaluate the "distant" signal, is the reference and the depth electrodes are all "contaminated" by that "distant" source. Is there a way to extract that "distant" signal for localization?
Thank you

sEEG signals are sensitive to local sources, essentially. The local signal is so strong that it tends to mask contributions from distant areas.

1 Like

There is no way to subtract local signals and keep distant sources?If I reference the depth electrode to a silent reference and then compared the signal to an earlier sample of the same electrode, will it work?

I am not sure of what you propose exactly, but I don't think it will work.

Think EEG or MEG: they often use as many electrodes as possible to ensure signal detection as close as possible to underlying sources.

The biggest porblem with sEEG is localizing the true seizure inset zone. In many cases, we determine that region depending on the first derivative showing seizure signature but given the limited area of sampling, the locally detected seizure onset signature might reflect a propagation sone rather than an onset zone. My hope is to think of a way to map the seizure activity in a targeted region by surrounding it with depth electrodes. Once a seizure happens, all electrodes can be used to localize the activity. My understanding from you, this is not possible due to the masking of the distal signal by the high amplitude of the local activity.

Unfortunately, I don't think it is possible to subtract easily "distant effects" from local effects in sEEG. EEG electrodes all see more or less the same things from the brain activity, we have a continuous sampling of the scalp and we can use all the electrodes together to identify where the signal is coming from. The signals picked up by the sEEG electrodes are too local for that, and not sampled uniformly.

What some people do, to remove the ambient noise (and some of the distant sources), is to re-reference all the contacts of the same sEEG probe or within the same area to an average reference: this subtracts the average of all signals, and therefore makes much more salient some local effects (eg. origin of a spike). This can be done with Brainstorm.

More promising markers of identifying the spatial extent of the seizure onset zone come from the frequency analysis of the signals, the closer to the SOZ the stronger the fast oscillations. You can find three detailed tutorials on how to explore this dimension in Brainstorm:

Thank you Francois

That is unfortunate. Then the question that comes to mind, can we use the waveform of the seizure(may be the initial 1 second) and apply it to the time series from the other electrodes to "extract" that waveform and its power?

Thank you

I'm not sure I understand what you describe as "applying a a waveform to another waveform".
Do you want to compute a convolution of all the data with the first second of the seizure?

Maybe, but I haven't seen this approach documented anywhere...
I think you should rather target an increase of power in a specific frequency range rather some random signal convolution. Please start by reading the "SEEG / Epileptogenicity tutorial", and get back to us if you have questions about this method.

@odavid FYI

Hi @Francois,

Can you please provide some information about how the coverage of SEEG is calculated? I mean, how do you know how far from an SEEG electrode contact you should see the significant effect (e.g., epileptogenicity index)? Because based on visual inspection of EI it seems to me the the significant area around a contact is very much like a perfect circle of specific diameter. How does BS come up with the diameter and why is that so uniform in every direction?


The SEEG recordings need to be interpolated on the MRI volume or surface mesh for various purposes.
The extension of the neighborhood that is considered around each contact depends on the application.

See this article (search for "sphere" and "neighborhood"):
Advances in human intracranial electroencephalography research, guidelines and good practices - ScienceDirect

Some more information in this section of the SEEG/Epileptogenicity tutorial:

For the epileptogenicity computation:

  • On a surface: it starts from all the vertices within a sphere of 10mm each the contact, then extends this area with all the vertices that are connected in the cortex tesselation at a maximum distance of 20 connections. It creates a large neighborhood, possibly much larger than the actual brain area that is directly recorded by the contact. When interpolating between the SEEG recordings and the surface, a weight of 1/D^2 is applied (D=distance contact-vertex), so the impact of a contact on a given vertex in the interpolation decreases fast when the distance to this vertex increases.
    Code: brainstorm3/ImaGIN_spm_eeg_convertmat2ana_mesh.m at master · brainstorm-tools/brainstorm3 · GitHub
  • On the volume: the interpolation is first done on the cortex surface, and then from the surface to the volume. Therefore the spatial extent between the two methods should be similar.

In both cases, the distance between an SEEG contact and a possibly significant voxel or vertex is difficult to establish, as it depends on the density of the cortex mesh. You should not rely on this information to evaluate the spatial extent of any observed epileptic activity (e.g. seizure onset zone).
The key information to retain from these EI maps is the location of the area with the highest values in each significant blob.

@odavid Is this correct?


Thanks @Francois for the clarification.