[Eeglablist] Repairing non-stereotyped artifacts using using the Automatic Artifact Removal (AAR) toolbox

Makoto Miyakoshi mmiyakoshi at ucsd.edu
Mon May 19 19:23:56 PDT 2014


Dear Arno,

Would you mind telling us what the difference between these two?

*'baseline'   *
Spectral baseline end-time (in ms). Use NaN for no baseline removal{0}
*'powbase'   *
Baseline spectrum to log-subtract. 'baseline' parameter is ignored if this
parameter is used {def|NaN->from data} This is useful only when you want to
use a known baseline spectrum (e.g. from another condition) instead of
using the actual mean baseline spectrum of the data. Otherwise, leave this
out or specify as 'NaN' (not a number).

Makoto

2014-05-17 7:01 GMT-07:00 <lars.rogenmoser at psychologie.uzh.ch>:

> Dear Makoto
>
> I think including a power baseline should be done while pre-computing
> component measures (before pre-clustering). According to the std_spec,
> 'timerange' is suggested:
>
>     'timerange'  - [min max] use data within a specific time range before
>                    computing the data spectrum. For instance, for evoked
>                    data trials, it is recommended to use the baseline time
>                    period.
>
> However, there is furher information on FFTs discussing the baseline:
>
> http://sccn.ucsd.edu/eeglab/allfunctions/timefdetails.html
>
>
> *'baseline'   *
> Spectral baseline end-time (in ms). Use NaN for no baseline removal{0}
> *'powbase'   *
> Baseline spectrum to log-subtract. 'baseline' parameter is ignored if this
> parameter is used {def|NaN->from data} This is useful only when you want to
> use a known baseline spectrum (e.g. from another condition) instead of
> using the actual mean baseline spectrum of the data. Otherwise, leave this
> out or specify as 'NaN' (not a number).
>
>
> Is this not updated? or is this meant for plotting subject-wise?
>
> best, Lars
>
>
> -----Makoto Miyakoshi <mmiyakoshi at ucsd.edu> schrieb: -----
> An: lars.rogenmoser at psychologie.uzh.ch
> Von: Makoto Miyakoshi <mmiyakoshi at ucsd.edu>
> Datum: 16.05.2014 17:33
>
> Kopie: EEGLAB List <eeglablist at sccn.ucsd.edu>
> Betreff: Re: [Eeglablist] Repairing non-stereotyped artifacts using using
> the Automatic Artifact Removal (AAR) toolbox
>
> Dear Lars,
>
> > By default, I would pre-cluster with "3 25". However, I am confused
> since the whole frequency range (0-250 Hz) is alway calculated anyway.
>
> Imagine your experimental paradigm is SSVEP or Kanizsa triangle
> presentation targeting at gamma-band activities. You want to use 40-80Hz in
> this case, in stead of 3-25Hz. This allows users to focus on specific
> frequency ranges for clustering criteria.
>
> > Do you know how to include a mean baseline power deriving from a
> specific time period (at the level of clusters)? I am not sure whether its
> 'baseline' or 'powbase'. Both are discussed in terms of spectral analysis.
>
> I checked std_preclust() help but did not see 'baseline' or 'powbase'...
> aren't they options for ERSP?
>
> Makoto
>
> 2014-05-16 7:37 GMT-07:00 <lars.rogenmoser at psychologie.uzh.ch>:
>
>>  I am just interested in the spectra (no ERSP). thats why I am
>> pre-computing that only. Concerning pre-clustering, it is also possible to
>> specify the freq. range [Hz]. By default, I would pre-cluster with "3 25".
>> However, I am confused since the whole frequency range (0-250 Hz) is alway
>> calculated anyway.
>>
>>
>> Do you know how to include a mean baseline power deriving from a
>> specific time period (at the level of clusters)? I am not sure whether its
>> 'baseline' or 'powbase'. Both are discussed in terms of spectral analysis.
>>
>> best, Lars
>>
>>
>>
>>
>> -----Makoto Miyakoshi <mmiyakoshi at ucsd.edu> schrieb: -----
>>  An: lars.rogenmoser at psychologie.uzh.ch
>> Von: Makoto Miyakoshi <mmiyakoshi at ucsd.edu>
>> Datum: 15.05.2014 01:21
>>
>> Kopie: EEGLAB List <eeglablist at sccn.ucsd.edu>
>> Betreff: Re: [Eeglablist] Repairing non-stereotyped artifacts using using
>> the Automatic Artifact Removal (AAR) toolbox
>>
>> > I was curious about whether it is possible to include different
>> frequency ranges (delta, alpha...) in component clustering seperately.
>>
>> You can specify the freq range in ERSP when preclustering. However, if
>> you want to include multiple ranges, that's impossible. Also, if you want
>> to test the difference across freq ranges, you need to repeat preclustering
>> separately for each range, and the results should be treated as separate
>> too. You can't do multiple frequency range clustering with different
>> weights etc.
>>
>> > I am confused since the spectra (plot spectra) remains the same
>> independently of the 'frequrange'(?) I add. However, on the matlab file
>> (Study-cluster, specdata) the frequency-bins change correspondingly.
>>
>> I don't understand this. Is this a spectrum curve or ERSP? Please tell us
>> more details.
>>
>> > By the way, what information does the file "specfreqs" contain?
>>
>> I don't to know off the top of my head...Which function takes this
>> option, do you know?
>>
>> Makoto
>>
>> 2014-05-14 14:02 GMT-07:00 <lars.rogenmoser at psychologie.uzh.ch>:
>>
>>> Dear Makoto
>>>
>>> Thanks for the insightful inputs.
>>>
>>> I guess 'timerange' [min max] is the spectopo parameter for including a
>>> power baseline.
>>>
>>> Thanks for referring to the code for exporting grouped frequency values.
>>> However, I was curious about whether it is possible to include
>>> different frequency ranges (delta, alpha...) in component clustering
>>> seperately. This might be advantageous since less tests (per Hz) have
>>> to be done, decrasing the number of multiple corrections. I am confused
>>> since the spectra (plot spectra) remains the same independently of the
>>> 'frequrange'(?) I add. However, on the matlab file (Study-cluster,
>>> specdata) the frequency-bins change correspondingly.
>>>
>>> By the way, what information does the file "specfreqs" contain?
>>>
>>> Lars
>>>
>>>
>>>
>>>
>>>
>>> -----Makoto Miyakoshi <mmiyakoshi at ucsd.edu> schrieb: -----
>>> An: lars.rogenmoser at psychologie.uzh.ch
>>> Von: Makoto Miyakoshi <mmiyakoshi at ucsd.edu>
>>> Datum: 12.05.2014 17:52
>>> Kopie: EEGLAB List <eeglablist at sccn.ucsd.edu>
>>>
>>> Betreff: Re: [Eeglablist] Repairing non-stereotyped artifacts using
>>> using the Automatic Artifact Removal (AAR) toolbox
>>>
>>> Dear Lars,
>>>
>>> > For example, how do I run a spectro-analysis with a 90% overlapping
>>> Hanning window of 1 s (512-point short-time Fourier transform)
>>> including a baseline (mean baseline power) within the stimulus, and
>>> with log-converted power values (dB)?
>>>
>>> Go to http://sccn.ucsd.edu/pipermail/eeglablist/2014/thread.html and
>>> find this thread
>>> [Eeglablist] How to change FFT parameters?
>>> See what Arno says there.
>>>
>>> > How would I caluclate relative (vs. absolute) dB?
>>>
>>> In EEGLAB, 1uV peak-to-peak sine wave power is defined to be 0dB;, 10uV
>>> is 20dB, 0.1uV is -20dB etc.
>>>
>>> > Is it possible to run the spectro-analysis in grouped frequency ranges
>>> (e.g., delta, theta, alpha, beta, gamma) seperately? What would be the code
>>> I would have to add in the interface? I am a bit confused, since the
>>> Freq.range [Hz] on the "Build preclustering array"interface is by default
>>> "3 25" and by changing it would not change the output.
>>>
>>> Go to http://sccn.ucsd.edu/pipermail/eeglablist/2014/thread.html and
>>> find this thread
>>> [Eeglablist] Extracting data in frequency ranges from dataset
>>> Find my replies. You can find code for that.
>>>
>>> > And two questions concerning statistics: Does the sampling rate have
>>> an impact on (permutation) statistics? Currently, I am using a sampling
>>> rate of 500. I would probably have to down-sample if it increases multiple
>>> corrections.
>>>
>>> Although I don't have mathematical proof, I would say no.
>>>
>>> > And finally, I can't find a file with the statistical values. I assume
>>> I would just have to click on the cluster I am interested and select the
>>> statistics ("STATS") and plot spectra. Then matlab should create a new
>>> file, right?
>>>
>>> If you want to extract statistics results, you need to run it from
>>> command line. I strongly believe that this should be fixed so that users
>>> can access result values. Let me show you a temporally solution for a
>>> workaround. In STUDY edit/plot GUI, show spectra for all clusters. Press
>>> 'ok' and close the edit/plot GUI to make EEGLAB updated. In the Matlab
>>> workspace, click 'STUDY-cluster(1,x) where x is your desired cluster -
>>> spectra (or something like that). You can find stacked arrays of spectra in
>>> the order of conditions (conditions are separated as cell), frequency, and
>>> ICs. You can run statistics on your own using these values.
>>>
>>> Makoto
>>>
>>> 2014-05-12 4:20 GMT-07:00 <lars.rogenmoser at psychologie.uzh.ch>:
>>>
>>>> Dear Makoto
>>>>
>>>> Thank you very much for your help on the time-frequency analysis. However,
>>>> since EEGlab has been updated recently, the problem ("Inner matrix
>>>> dimensions must agree") did not occur anymore. Thus, the problem is solved.
>>>>
>>>> Currently, I am analyzing spectra on the level of IC clusters. However,
>>>> even though I studied the manual carefully, I don t fully understand how to
>>>> use these spectopo parameters provided on the interface “Precompute
>>>> component measures ”(Power spectrum). I was wondering whether you could share
>>>> your knowledge on this. I would be really thankful!
>>>>
>>>> For example, how do I run a spectro-analysis with a 90% overlapping
>>>> Hanning window of 1 s (512-point short-time Fourier transform)
>>>> including a baseline (mean baseline power) within the stimulus, and
>>>> with log-converted power values (dB)?
>>>>
>>>> How would I caluclate relative (vs. absolute) dB?
>>>>
>>>> Is it possible to run the spectro-analysis in grouped frequency ranges
>>>> (e.g., delta, theta, alpha, beta, gamma) seperately? What would be the code
>>>> I would have to add in the interface? I am a bit confused, since the
>>>> Freq.range [Hz] on the "Build preclustering array"interface is by default
>>>> "3 25" and by changing it would not change the output.
>>>>
>>>> And two questions concerning statistics: Does the sampling rate have an
>>>> impact on (permutation) statistics? Currently, I am using a sampling rate
>>>> of 500. I would probably have to down-sample if it increases multiple
>>>> corrections. And finally, I can't find a file with the statistical values.
>>>> I assume I would just have to click on the cluster I am interested and
>>>> select the statistics ("STATS") and plot spectra. Then matlab should create
>>>> a new file, right?
>>>>
>>>> I hope these arent too many questions. Thousand thanks!!!
>>>>
>>>> best, Lars
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> -----Makoto Miyakoshi <mmiyakoshi at ucsd.edu> schrieb: -----
>>>> An: lars.rogenmoser at psychologie.uzh.ch
>>>> Von: Makoto Miyakoshi <mmiyakoshi at ucsd.edu>
>>>> Datum: 15.04.2014 17:34
>>>>
>>>> Betreff: Re: [Eeglablist] Repairing non-stereotyped artifacts using
>>>> using the Automatic Artifact Removal (AAR) toolbox
>>>>
>>>> Dear Lars,
>>>>
>>>> Sorry for delay.
>>>> I found you sent me the document that says your size(EEG.icaact,1) ==
>>>> 0, which is a problem. This means that you don't have independent component
>>>> activation.
>>>>
>>>> Do you have ICA matrix? Check your main EEGLAB GUI to see whether ICA
>>>> is yes. If it's yes AND size(EEG.icaact,1) == 0, that is pathological. You
>>>> should try this code
>>>>
>>>> EEG = eeg_checkset(EEG, 'ica');
>>>>
>>>> although I don't understand why you don't have EEG.icaact... and if you
>>>> don't have EEG.icaact you should not have been able to compute component
>>>> cross coherence.
>>>>
>>>> Could you check you EEG.data also? Does it say 'mmo' (memory mapping
>>>> object) by any chance? If that's the case, then you should do
>>>>
>>>> EEG.data = EEG.data(:,:,:);
>>>>
>>>> to retrieve all data into RAM. I hate this memory mapping object since
>>>> this is not compatible with many EEGLAB functionis and causes troubles
>>>> everywhere. This memory mapping option comes with (I believe) Mobilab. In
>>>> the EEGLAB set up option you may find an item for this memory mapping
>>>> object. TURN IT OFF if it is ever checked 'on'!
>>>>
>>>> Makoto
>>>>
>>>> Makoto
>>>>
>>>>
>>>> 2014-04-07 3:01 GMT-07:00 <lars.rogenmoser at psychologie.uzh.ch>:
>>>>
>>>>> Dear Makoto
>>>>>
>>>>> I pasted the following code and received a large message. I copied it
>>>>> in a word file and attached it. I further attached my dataset for
>>>>> toubleshooting.
>>>>> I am still struggling with this error. I really appreciate your help.
>>>>>
>>>>> Best regards,
>>>>> Lars
>>>>>
>>>>>
>>>>> -----Makoto Miyakoshi <mmiyakoshi at ucsd.edu> schrieb: -----
>>>>> An: lars.rogenmoser at psychologie.uzh.ch
>>>>> Von: Makoto Miyakoshi <mmiyakoshi at ucsd.edu>
>>>>> Datum: 05.04.2014 04:26
>>>>>
>>>>> Betreff: Re: [Eeglablist] Repairing non-stereotyped artifacts using
>>>>> using the Automatic Artifact Removal (AAR) toolbox
>>>>>
>>>>> Dear Lars,
>>>>>
>>>>> Check the following, copy and paste what you find. I may be able to
>>>>> find something there.
>>>>>
>>>>> EEG.icachansind
>>>>> size(EEG.icaact,1)
>>>>> EEG.nbchan
>>>>> EEG.history
>>>>>
>>>>> If this does not work... you may need to send me a sample data for
>>>>> troubleshooting. It's highly likely that this is not an EEGLAB bug, but I'd
>>>>> be happy to help you anyways!
>>>>>
>>>>> Makoto
>>>>>
>>>>>
>>>>>
>>>>> 2014-04-03 5:05 GMT-07:00 <lars.rogenmoser at psychologie.uzh.ch>:
>>>>>
>>>>>> As I mentioned, on channel level I am able to plot time-frequency and
>>>>>> also cross-coherence. In fact, I am also able to run component
>>>>>> cross-coherence, but unfortunately no component time-frequency which seems
>>>>>> to be quite strange. I receive the error "Inner matrix dimensions
>>>>>> must agree".
>>>>>>
>>>>>> I imported vhdr-files recorded from Geodesics-system. I used a
>>>>>> self-generated channel location (please find attached), however I read this
>>>>>> template over "read locations" and chose the "BESA or EGI-3D cartesian .sfp
>>>>>> file" format, ending up with 109 channels. My sampling rate is 500 Hz. I
>>>>>> tried lowering the sampling rate (250 Hz) but the problem remained. I
>>>>>> re-referenced using an averaged reference. I filtered 1-100 Hz (Basic FIR
>>>>>> filter) and used the cleanline algorithm. I removed segemnts on which the
>>>>>> subjects responded (select data using events), ending up with a dataset
>>>>>> with segemnts of interest. I ran ASR and then ICA (extended), of course
>>>>>> excluding noisy channels before. I then epoched the segemnts (extract
>>>>>> epoches) beloning to the same conditions. Precisely, I have 4 segements per
>>>>>> conditions however with a lenght of 1 minute (+5 sec pre-stimulus). I haven
>>>>>> t removed the baseline or corrected anything in this circumstance.
>>>>>>
>>>>>> Furthermore, I re-named the types of the triggers (Event values) and
>>>>>> also co-registered channel location with head mesh (colin27headmesh). I am
>>>>>> using the latest version of EEGlab and Matlab R2013b.
>>>>>>
>>>>>> I really hope this helps.
>>>>>>
>>>>>> Best, Lars
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>
>>>>
>>>> --
>>>> Makoto Miyakoshi
>>>> Swartz Center for Computational Neuroscience
>>>> Institute for Neural Computation, University of California San Diego
>>>>
>>>
>>>
>>>
>>> --
>>> Makoto Miyakoshi
>>> Swartz Center for Computational Neuroscience
>>> Institute for Neural Computation, University of California San Diego
>>>
>>
>>
>>
>> --
>> Makoto Miyakoshi
>> Swartz Center for Computational Neuroscience
>> Institute for Neural Computation, University of California San Diego
>>
>
>
>
> --
> Makoto Miyakoshi
> Swartz Center for Computational Neuroscience
> Institute for Neural Computation, University of California San Diego
>



-- 
Makoto Miyakoshi
Swartz Center for Computational Neuroscience
Institute for Neural Computation, University of California San Diego
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20140519/c65e7896/attachment-0001.html>


More information about the eeglablist mailing list