I am a psychologist turned professor who is passionate about studying learning processes based on EEG and other biometric data (eg, GSR, eye tracking). My first fascination is however EEG analysis.
I wrote some software for stimuli/data capture and have spent a good bit of time working with EEGLAB (eg, attended EEGLAB workshop in fall 2015). Have conducted two studies (one with grad student). For me though, EEGLAB focus on ERP does not fully capture where I want our program to be (eg, ICA/time-frequency for continuous data).
Short description - I ran across Brainstorm last week and have spent the last two days working through your Intro and Epilepsy tutorials. Wonderful.
Right now, I am trying to figure out how to set up a Brainstorm study (eg, anatomy) with the Emotiv device. I have the Emotiv .ced file (and EDF data generated by Emotiv Testbench), but am stuck. How can I set up so impoverished a device given no MRI etc.?. Any advice you may have as to how to proceed will be greatly appreciated.
I have taken a quick look at the Brainstorm forum on these matters but have not found anything yet. I’ll keep looking.
Thanks in advance for any and all help.
Hello,
You can convert your Emotiv recordings into something that can be read and processed with Brainstorm (EDF for instance).
If you can find some Matlab code that reads directly the Emotiv recordings, I could also add a direct support for them in Brainstorm.
However, it might not a good idea to try to estimate brain sources based on less than 10 noisy electrodes.
Technically you can, nothing will stop you from doing it in the interface. But with such a low coverage it is not possible to obtain accurate brain maps, therefore you may end up with a distorted image and incorrect interpretations.
It might be safer to work at the sensor level with these devices, using machine learning and classifiers to explore the differences between experimental conditions.
Cheers,
Francois
Thanks Francois. I certainly understand the need for caution/conservatism in working with such a low density. I would like to step up in density at some point ($$). I am very interested in using classifiers, any advice on how to get started with that? The only related tool I am aware of is BCILAB; are there others you might recommend?
Thanks.
Rich
PS The EDF files (from Testbench) did not read as expected however in Brainstorm (of course I could have been doing something wrong) but .set files which I created from the EDF via EEGLAB worked well.
EDF: Can you describe better what you mean by “did not read as expected however in Brainstorm”?
Classifiers: You can start from this tutorial:
http://neuroimage.usc.edu/brainstorm/Tutorials/Decoding
[QUOTE=ingramre;11869]
Right now, I am trying to figure out how to set up a Brainstorm study (eg, anatomy) with the Emotiv device. I have the Emotiv .ced file (and EDF data generated by Emotiv Testbench), but am stuck. How can I set up so impoverished a device given no MRI etc.?.[/QUOTE]
You can use a standard anatomy template if you don’t have individual MRIs: http://neuroimage.usc.edu/brainstorm/Tutorials/DefaultAnatomy
The next step up if you don’t have individual MRIs, but have digitised head points (gathered with a Polhemus digitiser or similar device), is to warp an anatomical template to the individual’s head shape: http://neuroimage.usc.edu/brainstorm/Tutorials/TutWarping
I would however agree with Francois, that it’s probably advisable to stay in sensor space with so few (and unevenly distributed) electrodes. EEG source imaging is typically done with 64-256 channel systems. Here’s a figure showing the theoretical sensitivity of the Emotiv Epoc based on 10-20 positions and simulated data:
I’m not really interested in source estimation beyond 2D scalp. ICA results suggest that at least the top few ICs are identified reliably (usually blink artifacts). I have an ‘emotiv_epoc.pos’ channel file downloaded from this forum that works nicely. Tried to attach here but says invalid file.
BTW - I don’t see the Attachment 1527 noted in the previous post. Any ideas on how to get it? Would really like to see it.
The 2D scalp representations are still in sensor space (i.e. they only interpolate values between sensors). We were cautioning about using low-density EEG headsets for EEG source imaging (i.e. estimating brain sources from scalp EEG). I would still recommend you display the sensor positions when you plot 2D topographies of the Emotiv Epoc, because the uneven distribution of sensors may be somewhat misleading (Ctrl + E or Right click on the figure -> Channels -> Display sensors). The attachments only show up when you’re logged in I think, but here’s another link if it still doesn’t work: https://box.bic.mni.mcgill.ca/s/8VAokhbtyNlGFhF
Thanks greatly. Couldn’t see it although logged in. Saw the linked file - thanks. That would seem to indicate that Emotiv is less sensitive in the frontal lobes (as opposed to say Occipital)? Does this hold even if ICA Blinks taken into account?
The figure gives an idea of the relative sensitivity of of the Emotiv Epoc to various cortical regions. The ICA doesn’t matter in this context; these results are solely based on the position of the sensors relative to the brain. ICA is useful for removing artefacts from your data, but reliable frontal blink components don’t indicate that the device is particularly sensitive to frontal locations since the amplitude of EOG signal is much greater than any physiological EEG activity. I wouldn’t say the frontal lobes as a whole are less well captured than occipital lobes however, but certainly some areas of frontopolar as well as anterior and basal temporal cortex are perhaps not ideally recorded. I think the results are rather intuitive—they show that the Epoc is more sensitive to areas underlying the sensor locations. Some areas like parietal and dorsal motor cortex are poorly captured for instance. What you also need to keep in mind with the Emotiv Epoc is that the sensors are fixed to rigid arms, so depending on the size of your participants’ heads, the position of the sensors can vary non-negligibly. With the standard 10-20 system electrode positions are based on proportions, and sensors are therefore more consistently placed over the same anatomical structures between individuals.