[Eeglablist] source localization problem

[Makoto Miyakoshi]

Hi Jinwon,

> For subjects with psychiatric disorders, is it valid to apply equivalent
dipole fitting on brain components using a template based on healthy
individuals.


You mean post-ICA dipole fitting to ICs.
The validity depends on whether we can assume more or less the same (1)
volume conductor model and (2) neocortical dynamics. (1) could be affected
if your patients show severe sulcal opening due to drug effects or
malnutrition etc.. (2) is much more difficult to confirm, as neocortical
dynamics itself remains poorly understood, as well as how it is affected by
each disease condition.

However, practically speaking, you can apply the identical methods to both
healthy control and patients, and in the limitation section you mention the
above two problems and say 'Care must be taken when interpreting the group
difference...' I think it is ok. Basically, group difference you report can
be contributed by all of these experimentally uncontrolled factors which
you can imagine. You don't need to be perfectly knowledged when you write a
paper. No one can publish any papers if that were a requirement. Instead,
you make it clear that what is known, what is unknown, what is observation,
what is model/assumption.

>  Also, dipole fitting sometimes fails (and frequently) to find dipoles
in ROIs in some subjects, leading to ignoring these data from final
analysis. This could lead some biases on interpretation of resultant
dipoles, and many clinical studies are more vulnerable due to low sample
sizes from uncommon disorder types.


This is 'post-ICA inter-subject inconsistency', a classical ICA problem at
the group level analysis. I proposed groupICA based on Nima's 'Network
Projection'. If you are interested, download groupICA, feed it to your
ChatAI, and ask it how groupICA addresses this issue. It's rather
complicated. That said, there is no definitive solution to this problem. It
is easier to change your analysis philosophy: you give up the result's
uniformity across subjects. Instead, you only pick up the strongest results
from each subject which may or may not overlap among them. But thinking
about this solution is typically too heavy a burden for psychologists and
clinical researchers. I personally think that this is the reason why
ICA-based 'EEG signal analysis' did not become very popular. Group-level
analysis becomes messy, and you have to depend on probabilistic description
all the time (which reviewers do not like very much).

Makoto




On Wed, Aug 26, 2026 at 7:03 PM 장진원 via eeglablist <eeglablist at sccn.ucsd.edu>
wrote:

> Dear all,
>
> I wonder how to address dipole fitting in clinical models. For subjects
> with psychiatric disorders, is it valid to apply equivalent dipole fitting
> on brain components using a template based on healthy individuals.
>
>  Also, dipole fitting sometimes fails (and frequently) to find dipoles in
> ROIs in some subjects, leading to ignoring these data from final analysis.
> This could lead some biases on interpretation of resultant dipoles, and
> many clinical studies are more vulnerable due to low sample sizes from
> uncommon disorder types.
>
> Best Regards,
> Jinwon Chang
>
> Department of Psychiatry, Beth Israel Deaconess Medical Center
> _______________________________________________
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu or visit
> https://sccn.ucsd.edu/mailman/listinfo/eeglablist  .
>