I need some advice on what to do with a very noisy subject. I should preface this by saying I do not have ECG or EOG or EMG recordings, so I'm flying with the help of ICA decomposition.
This is the subjects' original recordings with a 1Hz high pass (and notch) filters applied:
I had considered marking the offending channels as bad, but there are several channels implicated so that would result in me discarding a significant amount of data.
I also came across this post regarding muscle artifact, and @ebock recommended using an SSP to capture the spikes: Muscle artifacts
My question is as follows:
Is this subject irredeemable? And if not, would it be advisable to use the 17 ICA projectors, or should I give Dr. Bock's suggestion a try? Would I be losing too much data with the application of the 17 ICA projectors?
I don’t recommend you remove any ICA or SSP component you are not sure about the origin. In your first ICA experiment, I’m not sure I understand the reason for which you remove the first three components: the first one could be something like 50Hz residuals, but the 2nd and 3rd ones are probably some brain components (central and occipital sources).
It looks like your subject moved a lot: all the darker area you see in the figure where all the channels are displayed are probably muscle artifacts. You can either mark these segments as bad, try to compute and SSP or ICA projector to remove this activity, or apply a low-pass filter that will remove all the high-frequency activity. The most appropriate solution depends on the question you are trying to answer with this data. Keep in mind that if the subject moved a lot, the MEG probably didn’t much brain activity. If you have recorded the head localization coils continuously, you could could try looking to identify the periods where the subject was moving his/her head.
Hi Aquila,
It appears this artifact is muscle. Most muscle artifacts occur between 40-240Hz. Since we don’t always want to analyse this large spectrum, you should focus only on the frequencies that you are interested - don’t worry about removing artifacts in a frequency range that you are not going to analyse.
There are few things to try before removing so many dimensions…17 is quite a lot. Normally I don’t use ICA for this type of artifact, but instead I find the PCA (SSP) does a better job. For this high freq noise, you can try marking those segments as an event (i.e. muscle). Then you can compute an SSP projector on that event (Artifacts->SSP Generic). Try to narrow the freq range to capture the artifact the best (i.e. 40-90Hz or 40-150Hz). This will depend on what your frequencies of interest are for analysis. You can also follow the cookbook example that you cited. I also suggest you look at the topography and the time series of any component and be sure you are only removing those dimensions that you are confident define the artifact.
Beth
