[Eeglablist] Handling Abrupt Changes After ASR Preprocessing

[Makoto Miyakoshi]

John is absolutely right.
The original ASR, which is the core part of clean_rawdata, DOES address
this issue by using the exact solution John has suggested.
If I remember correctly, clean_rawdata (written by Christian) uses linear
blending, while CleanLine (written by Tim) uses sigmoidal curves. Users can
even specify the slope coeffs.
However, when Arno implemented the current version of clean_rawdata (I used
to manage it until 2017 ish), he added 'window rejection instead of ASR'
with the default 'on'. This disables ASR.

I once showed the time-frequency plot of a simulated signal with
discontinuities. You can see that demo in the following Youtube video from
NIH HBCD summer seminar in 2023.
https://urldefense.com/v3/__https://youtu.be/NHKIPU_T7-0?t=1861__;!!Mih3wA!Gc6HvV3RyZbgy0n8c_IucAWN5J5fZXKIU3ul5G4MFaNKLvl9yAF3i2RC5hfB8i_GE4ZRp7nVFEj6F8LXzpm4n0_XjrY$ 
Note that, as Arno explained, their filter functions are designed not to go
over the 'boundary' event markers. So as long as you use EEGLAB functions,
you do not experience the problem shown in the movie.

However, the basic problem is still there. When I process the data on my
own, writing extra lines to exclude data windows that contain 'boundary' is
cumbersome.
If you feel the same, I recommend you try the following steps.

   1. Use clean_rawdata with ASR 'on' (i.e. uncheck the window rejection
   instead of ASR)
   2. Disable the final window rejection

The step 2 means that clean_rawdata checks whether ASR could really clean
the data. When it finds the ASR cleaning was not effective enough, it
rejects that window (default 0.5 or 1.0 s, I forget).
However, you can disable it. The reason for disabling it is so that
clean_rawdata does not change data length or event structures.
If you perform event-related potential analysis, you can reject 'bad
epochs' after all of these preprocessing and filtering, which is more
streamlined in my opinion.

As I discussed the issue with Masa yesterday, I heard that he was concerned
with the possibility that ASR may throw in some artificial signal in the
process of interpolation. Well, if you see my demonstration in the video,
you will find out your concern is probably unnecessary.

There is a very brief commentary on ASR. If you are not familiar with ASR,
you might find it informative.
https://urldefense.com/v3/__https://academic.oup.com/sleep/article/46/12/zsad241/7275639__;!!Mih3wA!Gc6HvV3RyZbgy0n8c_IucAWN5J5fZXKIU3ul5G4MFaNKLvl9yAF3i2RC5hfB8i_GE4ZRp7nVFEj6F8LXzpm4pjvstZQ$ 

I have published several thousands of EEG datasets (see my schizophrenia
papers with UCSD Psychiatry--this alone should count >3000) so far that
were processed with ASR. I also published technical papers reporting ASR's
behavior.
For example, see this paper--this is a study on systematic comparison with
no cleaning, ASR, ASR+ICA_level1, and ASR+ICA_level2.
https://urldefense.com/v3/__https://www.sciencedirect.com/science/article/pii/S0920121121002643?via*3Dihub__;JQ!!Mih3wA!Gc6HvV3RyZbgy0n8c_IucAWN5J5fZXKIU3ul5G4MFaNKLvl9yAF3i2RC5hfB8i_GE4ZRp7nVFEj6F8LXzpm4w5a2_LU$ 

My post-doc Heyonseok and I have submitted a paper entitled "Juggler's ASR"
which is under review now. I advocate ASR. Without ASR, how do you process
EEG data recorded during three-ball juggling?
The Three-ball juggling EEG presentation (July 2023, Tel Aviv)
https://urldefense.com/v3/__https://youtu.be/t777R12DYhA__;!!Mih3wA!Gc6HvV3RyZbgy0n8c_IucAWN5J5fZXKIU3ul5G4MFaNKLvl9yAF3i2RC5hfB8i_GE4ZRp7nVFEj6F8LXzpm4L9sJd00$ 

Makoto

On Tue, Oct 8, 2024 at 1:21 PM Hanna Szakács via eeglablist <
eeglablist at sccn.ucsd.edu> wrote:

> In the case of a phase analysis, would the abrupt changes introduce noise,
> masking the phase alignments, or introduce false data?
>
> Arnaud Delorme via eeglablist <eeglablist at sccn.ucsd.edu> ezt írta
> (időpont:
> 2024. okt. 8., Ke 3:55):
>
> > If you are using the clean_rawdata plugin on EEGLAB, it will add
> > discontinuity events between segments (i.e. ‘boundary’ events). When you
> > perform spectral decompositions in EEGLAB, the spectral decomposition
> will
> > not cross these boundaries and will process each chunk individually (this
> > is true for both the dataset and the study level).
> >
> > Arno
> >
> > > On Oct 7, 2024, at 11:53, Hanna Szakács via eeglablist <
> > eeglablist at sccn.ucsd.edu> wrote:
> > >
> > > Hi, this is something that interests me as well and I haven't found a
> > clear
> > > answer online or on the EEGLAB forums so far. So I'm piggybacking on
> > Masa's
> > > message to ask the community for clarification.
> > >
> > > Best,
> > > Hanna
> > >
> > > 和田真孝 via eeglablist <eeglablist at sccn.ucsd.edu> ezt írta (időpont:
> 2024.
> > > okt. 6., Vas 3:22):
> > >
> > >> Hi all,
> > >>
> > >> I usually use ASR for preprocessing in resting-state EEG analysis.
> This
> > >> function removes segments of data that contain excessive noise. It
> works
> > >> quite well; however, after the rejection, it directly concatenates the
> > >> periods before and after the removed segments, resulting in abrupt
> > changes
> > >> in the waveform. I am concerned that these abrupt changes may
> introduce
> > >> additional noise, such as ripples, in subsequent steps of the
> analysis,
> > >> such as time-frequency or connectivity analysis.
> > >>
> > >> Is this approach acceptable for further analysis? Or do you know of
> any
> > >> good solutions to avoid this problem?
> > >>
> > >> Best,
> > >> Masa
> > >>
> > >> --
> > >> Masataka Wada
> > >> Postdoctoral Scholar, Brain Stimulation Lab
> > >> Department of Psychiatry and Behavioral Sciences
> > >> Stanford University
> > >> _______________________________________________
> > >> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> > >> To unsubscribe, send an empty email to
> > >> eeglablist-unsubscribe at sccn.ucsd.edu
> > >> For digest mode, send an email with the subject "set digest mime" to
> > >> eeglablist-request at sccn.ucsd.edu
> > >>
> > > _______________________________________________
> > > Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> > > To unsubscribe, send an empty email to
> > eeglablist-unsubscribe at sccn.ucsd.edu
> > > For digest mode, send an email with the subject "set digest mime" to
> > eeglablist-request at sccn.ucsd.edu
> >
> >
> > _______________________________________________
> > Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> > To unsubscribe, send an empty email to
> > eeglablist-unsubscribe at sccn.ucsd.edu
> > For digest mode, send an email with the subject "set digest mime" to
> > eeglablist-request at sccn.ucsd.edu
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to
> eeglablist-request at sccn.ucsd.edu