[Eeglablist] Newbie needs help with evoked potentials and ICA
malteanders at gmail.com
Wed Aug 23 07:55:53 PDT 2017
I'm trying to keep this short and simple.
First I have to say that there I have nobody around me to help me with the
matter, so I am out on a limb. I really have no idea where else to turn to
than this mailing list. There is, however, no pressure on me to deliver any
results at all - so it is definitely not as horrible as it sounds, I
volunteered for this.
I have a good background in statistics and MATLAB, but unfortunately no
background in neurology or anything EEG-related at all, besides the stuff
I've been researching for the past weeks. I am currently doing an
internship in a pain research group here in Germany in the senior year of
my Masters (pharmaceutical sciences). I also have solid backgrounds in
physics and scripting languages.
We have a 64 channel g.HIamp EEG with a 64 channel *active *electrode cap.
Experimental setup: nociceptive evoked potentials using CHEPS on the left
forearm. Two sessions per subject, one with 20 trials with baseline 32°C to
peak temp. 51°C and the other with 20 trials with baseline 32°C to peak
temp. 49°C. CHEPS electrode is moved after each trial to avoid
sensitization, subjects are blindfolded and have earplugs in their ears. 9
electrodes connected: 1: Fp1; 2: Fz; 3: T7; 4: C3; 5: Cz; 6: C4; 7: T8; 8:
Pz; 9: A2. The study design can be found in many already published studies.
I am expecting results like in the many other studies, e.g.: Granovsky,
Yelena: Objective correlate of subjective pain perception by contact
heat-evoked potentials, The Journal of Pain (2008). When averaging Cz
Electrode after post-processing, it should approximately look like this:
I had 4 volunteers, each doing the measurements with 49°C and 51°C peak
During recording, I haven't applied neither a common reference (that is
possible when you're using active electrodes) nor any filters. I have left
that for post processing.
I have then imported the data into eeglab and added channel locations. You
can download the data here:
The subjects are called a, b, c and d for anonymization purposes. Each set
is the imported raw data, I only added the channel locations.
This is what I have done for post processing. I have spent countless hours
to figure out EEGLABs functions and understanding what they do. I have also
studied this: https://sccn.ucsd.edu/wiki/Makoto%27s_preprocessing_pipeline
, which helped me a lot. Some steps are equivalent to the study linked
(0). Imported with loadhdf5()
(1). Added location data (already done in the sets linked above)
(2). Downsampled to 250 Hz
(3). High-Pass filter at 1 Hz (basic FIR Filter, eegfiltnew())
(4). As CleanLine Plugin unfortunately doesn't work and I don't want to
notch filter, Low-Pass Filter at 40 Hz.
(5). Tried to remove bad channels. Christian's clean_rawdata messes up my
events so I used "automatic channel rejection" with standard settings
(kurtosis). Sometimes single channels were rejected.
(6). Re-Referenced to channel 9 (A2, earlobe). I didn't want to use average
reference, as I have not and equivalent amount of channels on the left and
right side of my head (e.g.: I only have one earlobe connected) and from my
understanding, average reference shouldn't be used then.
(7). Removed channel 9 from "channel locations"
(8). Epoched from -1 to 2 seconds, no baseline correction.
(9). Removed epoch 1 and 22 as the trigger there only stands for the
CHEPS-machine being started and stopped, no heat stimulus was applied there.
(10). Run ICA
(11). Tried many plugins to reject eye artefacts (didn't suceed with
those). Tried to remove artifacts with "Reject components by map", but I
don't think I'm experienced enough for this, even with the ICA tutorial.
This is how "c 51.set" looks like there:
I don't think I'm experienced enough to spot artifacts there. 1 looks like
the component I'm looking for (N2 and P2 peaks in the Cz region), 7 and 6
maaaaay be ocular artifacts but I'm really not sure. 2 and 4 may have to do
something with the evoked potential (not sure), 3 as well. 5 looks totally
(12). Still "c 51.set": Channel 2 has been rejected in (5). Components map
is visible in (11). When I don't remove any components whatsoever, I get
this for Cz after averaging:
This doesn't look like the graph from the study I mentioned above at all,
considering I didn't remove and artifacts or bad epochs at all.
If I do the same steps for "a 51.set" and "b 51.set" (when removing epochs,
a message "removing mean of each data channel" pops up, no idea what that
is supposed to do. I just clicked okay) and "d 51.set" I get this, where a
is "meh", b is beyong-wrong and d looks "okay":
*Now comes the questions-part. *Please consider that I'm an actual novice
and figuring out everything above took me a couple of weeks. The EEG wasn't
even connected when I arrived, Cheps and Trigger didn't work, no subjects
were available on the spot so I just recruited some friends etc... If you
sticked with me to this part: thank you so much!
-Do you think my data is corrupt and I have some errors in my measurements
so I need to rethink the design of the process (recording eeg +
thermostimulation using CHEPS)?
-Do you think there can't be made any assumptions about the data quality
before artifact removal? (I think I can answer this question for myself...)
-If I do need more post-processing: how should I continue? The ICA-concept
makes sense to me, although I don't think I'm experienced enough to use it.
Are there any other good options for automatic artifact rejection in
EEGLAB, especially for muscle movement and ocular artifacts? We will order
EOG electrodes in the upcoming week but they aren't available yet. Or is
there anybody willing to go through the process of ICA with me in the
presented data set?
-If my data isn't sufficient: What could I do to get better recordings? I
thought that it would make sense to connect more electrodes, e.g. 20
electrodes or 32 electrodes equally on both sides and then use an average
reference? Or would that make no big difference? Impendance has been "LOW"
on all Electrodes during impendance measurements, so impendance was mostly
way below 5 kOhm.
-Is it wrong to use only one earlobe as a reference? Does this maybe mess
up my data?
I know this is a long text, so if you read everything and are keen to help
me, you're more than invited to grab a coffee with me whenever you're in
Germany. I would be so grateful for any input, even if it's just "you're
doing everything wrong, start over". I am also happy to discuss this via
phone/skype with somebody or arrange a TeamViewer session.
Thank you so much!
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the eeglablist