[Eeglablist] Makoto's preprocessing pipeline: ICA matrices transfer

Scott Makeig smakeig at gmail.com
Fri Oct 15 20:48:25 PDT 2021


Ya -  I believe that ICLabel model was generally trained on ~1-Hz
high-passed data. If so, data in the 0-1-1.0 Hz low frequency band may well
function like 'noise' in ICLabel classification.

Scott

On Fri, Oct 8, 2021 at 8:44 AM Ya Zheng <123975520 at qq.com> wrote:

> Thanks, Scott.
> I agree with that the upper limit is best left open. My furhter question
> is whether IClabel should be applied to the 1-Hz high-pass data or the
> 0.1-Hz high-pass data after the ICA matrices transfer? I found that the
> confidence of classification by IClabel is lower for the 0.1-Hz high-pass
> data than for the 1-Hz high-pass data.
> Ya
>
>
> ------------------ 原始邮件 ------------------
> *发件人:* "smakeig" <smakeig at gmail.com>;
> *发送时间:* 2021年10月8日(星期五) 凌晨1:09
> *收件人:* "Ya Zheng"<123975520 at qq.com>;
> *抄送:* "eeglablist"<eeglablist at sccn.ucsd.edu>;
> *主题:* Re: [Eeglablist] Makoto's preprocessing pipeline: ICA matrices
> transfer
>
> Ya -
>
> Why do you bandlimit 1-35 Hz before applying ICA decomposition? The upper
> limit is best left open, as ICA decomposition can use spatially
> differentiating information contained in higher frequency portions of the
> source signals. The (~1 Hz) lower limit is justified by the understanding
> that very-low-frequency contributions may be dominated by non-brain
> artifact processes that would compete with brain sources for
> places/separate dimensions in the decomposition. If the same were to occur
> at high frequencies (e.g., hypothetically, say, *spatially* complex and
> varying broadband 100-150 Hz noise), then using a high-pass cutoff (e.g., <
> 100 Hz) would be justified by the same reasoning.
>
> Likely your finding that ICLabel returns more EMG sources when applied to
> the 0.1-Hz high-passed data (without low pass filtering) arises because
> ICLabel looks for the spectral signature of EMG (conceptually, a noise
> plateau from 20 Hz  to max Hz) as well as for scalp map characteristics,
> and doesn't see the spectral signature it is looking for in the 35-Hz
> low-passed data...
>
> Scott
>
> On Thu, Oct 7, 2021 at 12:21 PM Ya Zheng via eeglablist <
> eeglablist at sccn.ucsd.edu> wrote:
>
>> Hi Makoto,
>> I calculated ICA weight matrix and sphereing matrix with 1-35 Hz
>> band-passed data, then appled it to 0.1-Hz high-passed data. Then, I
>> applied the ICLabel to identify the artifact-related ICs. My question is
>> which dataset should be applied to? The 1-35 band-passed data or the 0.1-Hz
>> high-passed data? I found that the ICLabel resulted in some different
>> results, e.g, more muscle ICs found for the 0.1-Hz high-pass data. Thanks!
>> Best,
>> Ya
>> _______________________________________________
>> 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
>>
>
>
> --
> Scott Makeig, Research Scientist and Director, Swartz Center for
> Computational Neuroscience, Institute for Neural Computation, University of
> California San Diego, La Jolla CA 92093-0559, http://sccn.ucsd.edu/~scott
>


-- 
Scott Makeig, Research Scientist and Director, Swartz Center for
Computational Neuroscience, Institute for Neural Computation, University of
California San Diego, La Jolla CA 92093-0559, http://sccn.ucsd.edu/~scott


More information about the eeglablist mailing list