I have a dataset sorted by condition; approximately 180 trials of one and 130 of another. When I load the dataset in mne_analyze and compute the field patterns, I find a very nice peak in a particular neural region. I now want to test for significance. A colleagues of mine, a statistician, suggested that I randomly assign the trials to one condition or another, retrieve the resultant field pattern, and then do this another thousand times or so. This way, each timepoint will have its own confidence band, and I can then see whether the peak that appears when sorting by condition falls outside the confidence bands, thus indicating significance.
My question is, can this be scripted with MNE? I couldn't find a way to either load an fif onto a mesh or access the resultant timecourses (by label) via matlab. Can this be done? On a similar note, is there an easier way to do this? Thanks -
There are some great matlab utilities for doing more advanced analysis
methods found in chapter 10 of the latest MNE documentation: http://www.martinos.org/meg/manuals.php
What you described sounds sort of like bootstrapping, but I am confused
about getting confidence intervals if you simply have two condition choices.
Any particular epoch would then randomly get a 1 or a 2, resulting in only 2
possible field patterns for a particular timepoint within.
If perhaps you meant to do a more general randomization by permuting the
entire timeseries, then you might get a more robust type of confidence
interval. But you'd need to be careful to decide what that confidence
interval meant - i.e. the range of usual values of derived brain activity at
a point during that particular session may be different than the range under
other conditions, and certainly is different than no brain activity.
If you do build some tools in matlab for something like this, perhaps others
would like to use it too. There are all sorts of ways to share the code you
write, including this email list. Another random option is my website that
many MEG users tend to visit from around the world (I guess this is google's
fault?)... I'd be more than happy to host any useful software you might put
together that you want to share.