[FieldTrip] MEG Time-Frequency testing (ft_freqstatistics)

[Tavabi, Kambiz]

Greetings Fieldstrippers - Would someone savvy in cluster permutation testing kindly take a look at the attached plots. Briefly, Top left upper & lower panels are conditional grand averaged (n=10), time-frequency representation (TFR) of single virtual electrode. The TFR shows %change in power relative to baseline (i.e., baseline corrected power spectrum). The top right panel is the p-value map of sample-wise paired t-testing (two-tailed) of the data.

To deal with the multiple comparison problem, I used ft_freqstatistics to carryout out permuation tests on the source space time-frequency representations from my virtual sensors (2), as such the statistics were done without cfg.minnbchan and cfg.neighbours=[ ]. Testing was done on the time-frequency bins throughout the epochs of wide band data (i.e., no averaging over frequencies).

In the attached file the bottom panel is the plot of the permutation result for the same data and channel in the grand averages. As per advice from other fieldtrip users I created the plot by:
stat.stat2=stat.stat.*(stat.posclusterlabelmat == 1);
cfg.zparam='stat2';
ft_singleplotTFR(cfg, stat);

When I use ft_clusterplot(cfg, stat); for obvious reasons I get the following error. As I have no apriori freqs interval of interest I am not sure how to get around this.:
??? Error using ==> ft_clusterplot at 75
stat contains multiple frequencies which is not allowed because it should be
averaged over frequencies

My problem with the results in the attachment is that in the grand averages there is an obvious event related desynchrony (cold colors, negative values) difference between 0.4-0.8sec; 10-40hz which somehow doesn't get picked up, but the small blip of noise/event-related synchrony (hot colors, positive values) c. 0.5sec at 50Hz is revealed as different between conditions by the permutations! To carry out the permutations I used 
cfg.method           = 'montecarlo';
cfg.statistic        = 'depsamplesT';
cfg.correctm         = 'cluster';
cfg.clusteralpha     = 0.05;
cfg.clusterstatistic = 'maxsum';
cfg.neighbours       = [ ];
cfg.tail             = 0;
cfg.clustertail      = 0;
cfg.alpha            = 0.05;
cfg.numrandomization = 500;
So I presume that the result is not due to a signs (pos/neg values) issue, if I am using a two-tailed test. Changing cfg.tail & cfg.clustertail to 1 makes little to no difference, and a value of -1 results in errors. Basically, I'd appreciate if someone, based on what is observable in the grand averages, could confirm that the permutation result makes sense? And if not then some advice on what I am doing wrong. Thank you. -Kambiz

-K.
------------------------------------------------------------
Kambiz Tavabi PhD
Biomedical Imaging Laboratory
The Children's Hospital of Philadelphia
34th Street and Civic Center Boulevard
Philadelphia, Pa. 19104
Tel: 267.426.0302
email: tavabik at email.chop.edu<mailto:tavabik at email.chop.edu>
------------------------------------------------------------

-------------- next part --------------
A non-text attachment was scrubbed...
Name: TFR_testing.pdf
Type: application/pdf
Size: 103055 bytes
Desc: TFR_testing.pdf
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20101215/b50013c5/attachment-0001.pdf>