Skip to content

Commit c232600

Browse files
authored
Update RunICA.md
1 parent 414f20b commit c232600

File tree

1 file changed

+3
-1
lines changed

1 file changed

+3
-1
lines changed

tutorials/06_RejectArtifacts/RunICA.md

+3-1
Original file line numberDiff line numberDiff line change
@@ -379,6 +379,8 @@ This is not to say that using PCA should be done systematically. In
379379
general, PCA will slightly corrupt the data by adding nonlinearities, so
380380
it is better to use the full rank data matrix whenever possible.
381381

382+
A full discussion of corrupted ICA components, also called Ghost ICA components, is available in this [paper](https://www.frontiersin.org/articles/10.3389/frsip.2023.1064138/full).
383+
382384
Issues with data rank deficiencies
383385
---------------------------------------
384386

@@ -399,7 +401,7 @@ case, the user should manually reduce the number of components. If the ICA decom
399401
For example, when using 64 channels, enter "'pca', 63" in the option
400402
edit box. If you do not do this, the activity of one of the
401403
components that contributes the most to the data might be duplicated (as explained
402-
in the previous section), and you will not be able to use the ICA decomposition.
404+
in the previous section), and you will not be able to use the ICA decomposition. <font color=red>Note that because of small numerical innacuracy inherent to digital signal processing, we recommend people compute ICA, then compute average reference (see this [paper](https://www.frontiersin.org/articles/10.3389/frsip.2023.1064138/full).</font>
403405

404406
Issues associated with aggressive automated artifact removal before ICA
405407
---------------------------------------

0 commit comments

Comments
 (0)