hello,
I’m having trouble finding simultaneous EEG-MRS papers-- could anyone recommend any methods papers that have developed a way to clean EEG data collected during MRS scans? or has anyone here attempted to clean EEG data informally and could offer advice?
Hi,
How far along have you come? Very broadly, you need to have good control of the timing to do this effectively – depending on your hardware, this could well mean synchronizing the EEG recorder to the scanner’s reference clock, and recording some sort of per-TR trigger alongside the EEG data to inform whatever filtering algorithm you’re using.
In principle the EEG-MRS case should be a little cleaner to work with than the fMRI case. There are a couple of EEGLab plugins which should handle this for you; one of these is the Bergen plugin, which was developed in our lab many years ago (and unfortunately has not evolved much since); if you’re using EEGLab, I should be able to forward you a copy of that (if you haven’t found anything newer).
That particular algorithm is discussed here: 10.1016/j.neuroimage.2009.01.024 and builds on the approach of 10.1006/nimg.2000.0599
The Bergen implementation can incorporate realignment parameters to inform the filtering; not sure how important this would be in the EEG-MRS case, but perhaps you could try using the residual water frequency to infer when the subject might have moved…
Hope this helps a bit…
Alex.
Hi Alex,
thanks so much for your reply! yes sorry I should have clarified that our scanner doesn’t output ‘triggers’ for our MRS sequence in the same way that it does for our fMRI sequences, and that is the part I am having trouble with. Is that unusual? do others see ‘triggers’ from MRs sequences? I also wasn’t sure whether there would be something different about the gradient artifacts created by the MRS sequence since there would be editing pulses in addition to the RF pulse
Yes thank you for those references! We are very familiar with cleaning EEG during EEG-fMRI ; our current set up is to use moving window average artifact subtraction to remove gradient artifacts, but we need the triggers that you mentioned. To remove ballistocardiogram artifacts we regress out artifact recorded by carbon wire loops embedded in the EEG caps. We do synchronize our EEG system (BrainProducts) with the scanner clock as you mentioned (using a BrainProducts ‘syncbox’).
thanks so much!
Hi,
This is quite usual – we have a local patch which adds trigger generation to the GE advanced MRS sequence, on the same channel as the standard EPI sequence; I believe there are similar patches out there for other platforms. In principle we’re happy to share our patch, but this would need to be mediated by the vendor… Otherwise (on GE), the signal on the Scope Trig port of the exciter board can sometimes be useful in this context.
We use a couple of other gadgets (clock divider, RS485->TTL buffer) to make sure our scanner, BrainProducts syncbox and NNL syncbox (*financial interest) are all communicating properly, I guess you have that sorted already from the fMRI case but can give pointers if needed.
I haven’t tried without triggers, but if at least the clocks are well synchronized then I would think:
- Do a coarse initial alignment (per TR) based on the peak amplitude, then
- Refine this with a correlational approach (in some ways similar to spectral registration): so for a small range of offsets relative to the coarse alignment, find the offset which gives the best correlation with the mean of the other signals. Try oversampling/interpolation if this is too granular.
You could either do this on a single channel of EEG data to infer precise onset times to inform your existing filtering algorithm, or on all channels then directly subtract those aligned elements from your signal.
I would expect phase cycling and (probably) editing to confound this; you’d likely see better results by treating each step separately (so if you’re using 8-way phase cycling and MEGA-editing, you’ll want 16 separate models)
Alex.
Thinking about this some more: of course the gradients are doing the same thing regardless of phase cycling/editing, so you’ll only have an issue if significant amounts of RF are getting through the bandpass. This might be a risk with dual-band editing (HERCULES), otherwise it’s probably okay.