As everyone knows, there are four archetypal microstates in normal people that explain most of the global topographic variance. These four maps have right-frontal left-posterior, left-frontal right-posterior, midline frontal-occipital, and midline frontal topographies, and are labeled A, B, C, and D. However, I cannot get typical topographic maps when I compute EEG microstates in subject level. Some are close to the typical map, and some are very different. Here are two examples:
subject A
subject B
It seems that subject A is close to typical topographic maps. I wonder if there is any criterion to tell me the result is acceptable or not. I found that the atypical subject can influence group-level analysis.
Hey Jack,
Thanks for the question.
It is true that in the literature, 4.5 canonical topographies are usually reported. However, it is important to point out that it is quite possible to find differences (more or less states, or different topographies) depending on the conditions/populations studied. The analysis of EEG microstates is completely data-driven.
Concerning your results, I see on the topomaps that you use a small number of electrodes ( 6 ?). It’s therefore normal not to obtain the canonical topographies, as they are often calculated for a larger number of electrodes (32+).
Generally speaking, and in view of the various tests I’ve carried out, microstate analysis gives consistent results with at least 19 electrodes. Below this level, there is much more variability in topographies and segmentations. The main advantage of microstate analysis is to take advantage of the full spatial resolution of the EEG… so with only 6 electrodes, we’re entering the unknown… and it therefore becomes a little harder to interpret the results, and even harder to compare them with the literature.
If you do’nt have access to higher density EEG recordings, I would recommend to use the method lots of caution. In particular, I’d inspect all the topographies of each individual (as you’ve probably already done). I’d also check whether the duration of the states obtained after backfitting makes sense (usually around 100ms). I’d also check that the correlations of the microstates after backfitting make sense. for all subjects ( > 70% ?)
segmentation.compute_parameters()['XXX_mean_corr']
I hope this helps, and don’t hesitate to continue this thread if you need to clarify certain points or if you have issues at another stage of your analysis.
2 Likes