[Faster-eeg-list] A few questions and answers

[Hugh N. <no...@tc...>]

Dear all,
a few interesting questions have been sent to us, so we thought we
would share them along with our answers.

Question 1:
> Hi Hugh and Rob,
> I've recently read through your FASTER manuscript in the journal of neuroscience methods and downloaded the code.  The report makes a lot of sense, and I appreciate a less subjective approach that moves away from the "expert inspection" method of ICA component removal.  I work in a schizophrenia EEG/fMRI lab, and am working on applying this tool to one of our data sets.  While it was easy to set something up in the GUI, we have data from many different systems in the lab, and I'd like to get some things scripted.  Looking at your FASTER_process.m function, it looks like *_properties.m functions do most of the number crunching in steps 1-5 of your figure 2, and min_z.m identifies the trouble makers.  Does that sound about right?  I'll obviously spend a lot more time with the code and trying to figure things out, but I wanted to be able to cite your method and would like to be confident that in moving away from the gui, I do not butcher the FASTER process.
> Will you have an e-mail list for questions about the tool?  Is there any way I could be added to a list of those you'll inform about updates (if you're keeping one)?
> Thank you,
> Brian

Answer 1:
> Hi Brian,
> apologies for the late reply, we have been testing the new version of FASTER (v1.1) which has a nember of new features added and bugs fixed. We moved to a new host at https://sourceforge.net/projects/faster/ if you'd like to take a look.
> You are correct that the main functions used are the *_properties functions and the min_z function. The *_properties functions are somewhat idiocratic in that they each take different arguments, but their output is a standard format fed into min_z.
> The general approach for artifact identification is:
> (replace X with channel, epoch, component, etc)
> 1) list_properties = X_properties(EEG, further, inputs); % The further inputs can be seen in each file, they
> 2) exceeded_threshold = min_z(list_properties); % min_z also takes further arguments to allow you to turn on or off specific testing properties or use a Z-score threshold other than 3
> 3) bad_X = find(exceeded_threshold); % Exceeded threshold also details how many properties of each channel, epoch, component, etc exceeded the threshold, so you could use "bad_X = find(exceeded_threshold >= 2)" to tighten the rejection conditions to only reject Xs that are considered artifact-contaminated by 2 tests, if desired.

> Scripting shouldn't be too difficult, although we didn't write the code with easy scripting in mind, sorry about that! If you follow the faster_process.m file, you will be able to see what occurs between each processing stage (for example, re-referencing before channel rejection). As you say, the meat is in the _properties and min_z files; the rest is mostly bookkeeping, logging, GUI interface stuff, and tests for specific cases. If you use them, you should be following the process pretty well.
> We have just started a mailing list for FASTER discussion and updates; would you mind if we posted your question and my response onto the list? You can sign up at https://lists.sourceforge.net/lists/listinfo/faster-eeg-list if you want.
> Thanks,
> Hugh

Question 2:
> Dear Robert,
> i am reading your paper on FASTER and and I am so impressed that i would like to use it immediately with my data sets within an oddball auditory experiment.
> I have 3 EEG data sets (each with 20 subjects) from three groups: younger controls, older controls, and parkinsonian patients. Especially with the latest, a manual eog rejection through templates does not yield good results. We cannot apply the Schloegl method since we did not collect separate data files on eye movements.
> However, I used a 25 scalp-electrode (+ 4 eye, 2 mastoids and 1 nose reference), so I wonder whether in your opinion/experience this is going to be a major hindrance wrt  the use of FASTER.
> My very best regards,
> Alessandro

Answer 2:
> Hi Alessandro,
> thank you for your kind comments about FASTER.
> I think -- in theory -- it should be ok to run FASTER on your data. However, you should check the data after running the default settings. All of the parameters in FASTER are modifiable, so you may want to change the default settings to achieve a better results. I'm not really sure what settings might need to be changed for 25 electrodes, but in particular you should check that the EOG is being removed. If not, you could change some of the EOG settings.
> By the way, FASTER v1.1. is available on http://sourceforge.net/projects/faster/files/ - this version has some extra features, such as use of a trimmed mean in the Grand Averaging, which might be better for your data.
> Hope that helps,
> Rob

> Hi Alessandro,
> further to Robert's comments, I would recommend that you turn off the "Channel rejection" and the "Epoch interpolation" options if only using 25 channels, as the effectiveness of the bad channel detection measure dropped a lot when we tested on 32 channels. I don't know how effective the ICA decomposition will be, so do check with a few test datasets that good data isn't being removed!
> All the best,
> Hugh

In other news, hopefully FASTER 1.1.1 will be up later today.
Hugh