SEEG and clustering

Hi everyone,

I am working with iEEG and I have reached a point that I need to perform group analysis.
I try to explain the problem as best as I can.
We have multiple patients, each patient has multiple channels. some of these channels, even though they have different names and belong to different electrodes, record from the same region ( i.e. hippocampus ).
I cannot find any option in Brainstorm that would allow me to select these channels in one subject and combine them together and produce one unique cluster called " hippocampus subject 1" so that I can Z-score them and perform a grand-average across different subjects.
Is this possible in Brainstorm, without going through all the MRI things? If not, what can be done?
Thank you for your time

No, unfortunately there are no solutions yet for selecting sEEG contacts by anatomical region.
These would be an interesting feature but this is not part of our immediate development plans.

Localizing the contacts automatically would require to load an anatomical atlas for the subject, either as a volume or as set of surfaces ( and then write Matlab code for identifying to which region belong the various SEEG contacts. This is not an easy task...
If you are interested in contributing to these tools, we would be happy to give you additional entry points in the code and database.

Otherwise, the simplest would be a manual identification and selection of the contacts in your region of interest. Then you can extract the values of these contacts (and possibly average them) with the process "Extract > Extract values", typing the names of the contacts separated with commas in the option "Sensor types or names". You can average your cluster for a given subject by enabling the option "Average selected signals".

1 Like

Maybe @as_dub or @juangpc have better suggestions?

Thanks, I will try it!
I thought of a way my self in which I saved the pre-processed file as .BST and Imported the same file of the patient, several times. Then each time, I renamed one of the channels to " Hipp" and I changed the problem from one patient with 10 Hippocampal recording to 10 patients with 1 Hippocampal recording. Then I proceeded with normal epoching and eventually permutation test.
It felt logical in my head, but it didn't give significant results in the end so I'm not entirely sure about its validity.

So you are considering at the same level multiple electrodes of the same patient and different patients?
This would introduce a bias towards the patients that have the most SEEG recording sites in the hippocampus. I think it would make more sense to average all the hippocampal channels for each subject and then do your statistics across subjects.

This makes more sense yes.
The challenge here would be that some patients ( due to their performance) have much more trials in a certain condition in my memory task ( i.e. "hit" ) than other patients. bottom line is that it feels biased if I just grand average patient's intrasubject average. However, maybe a weighted average would do the trick.
Thank you, it sparked some ideas in my mind.

Hi, yes. I agree with @Francois, It seems more prudent to average all trials in each sensors and then do the inter-subject statistics, averages, etc... I would follow that path and then, try to compare area under the averaged curves instead of the actual peak amplitude differences between groups, conditions, etc... because when you have probable bias towards a certain condition, areas seem to be more stable.

Some parts need to be done "outside" bst some parts can be done through a normal processing box. One thing I do is to create new types of channels editing the channel spreadsheet in brainstorm (channel-> edit good/bad channels). This way you can very easily test around different configurations to see what is going on in each file.

1 Like