[FieldTrip] cluster analysis on time courses from virtual channels

[Loes Koelewijn]

Hi all,

I have a question about group analysis statistics of my MEG study. I've made
virtual sensors from group average images of an independent localiser run,
created at several time windows, based on the group inversion. I then
extracted dipole waveforms for all other conditions (6), for each subject
(10), based on those virtual sensors (the study has a repeated measures
design). I would now like to analyse group significance of any differences
between these 6 time courses, over all time samples, corrected for multiple
comparisons. I have tried to do this using FieldTrip's cluster based
permutation test (Monte Carlo), independently for each virtual channel, by
setting the neighbouring channels to 0. However, I get an error about the
design matrix when running ft_timelockstatistics.

Matlab's output is the following:

selected 1 channels
selected 251 time bins
selected 1 frequency bins
Using the gradiometer configuration from the dataset.
there are on average 0.0 neighbours per channel
using "statistics_montecarlo" for the statistical testing
using "statfun_depsamplesT" for the single-sample statistics
constructing randomized design
total number of measurements     = 60
total number of variables        = 2
number of independent variables  = 1
number of unit variables         = 1
number of within-cell variables = 0
number of control variables      = 0
using a permutation resampling approach
repeated measurement in variable 2 over 6 levels
number of repeated measurements in each level is 10 10 10 10 10 10
computing a parmetric threshold for clustering
Error using ==> statfun_depsamplesT at 78
Invalid specification of the design array.
??? Error using ==> statistics_montecarlo at 217
could not determine the parametric critical value for clustering

Error in ==> statistics_wrapper at 285
    [stat, cfg] = statmethod(cfg, dat, cfg.design, 'issource',issource);

Error in ==> ft_timelockstatistics at 117
[stat, cfg] = statistics_wrapper(cfg, varargin{:});

Could anybody help me in the direction of what is going wrong here? I
wondered earlier if my data structures were incorrect, but these were
created with ft_timelockgrandaverage, keeping the individual data. This is
what I ran:

[stat]=ft_timelockstatistics(cfg,cond1,cond2,cond3,cond4,cond5,cond6);

cfg =

             channel: 'Occipital_Inf_RR_X'
             latency: [-0.2000 0.8000]
              method: 'montecarlo'
           statistic: 'depsamplesT'
            correctm: 'cluster'
        clusteralpha: 0.0500
    clusterstatistic: 'maxsum'
                tail: 0
         clustertail: 0
               alpha: 0.0250
    numrandomization: 500
              design: [2x60 double]
                uvar: 2
                ivar: 1

cfg.design is 2*60, with the first row (ivar) [1:10,1:10,1:10 etc], second
row (uvar) [ones(1,10),ones(1,10)*2, etc].
I initially had ivar and uvar reversed, but I got the same error.

And this is the format of each data structure:

         label: {'Occipital_Inf_RR_X'}
       fsample: 250
           avg: [1x826 double]
           var: [1x826 double]
          time: [1x826 double]
    individual: [10x1x826 double]
        dimord: 'subj_chan_time'
           cfg: [1x1 struct]
          grad: [1x1 struct]

Apologies for the massive email, but I was hoping this is enough info for
someone to have more of a clue than I do?

Kind regards,

Loes

PS Do let me know if you think this approach is wrong in the first place. I
was trying to avoid pre-setting time windows for image-based statistics, as
we do not really have strong a priori expectations for times.

-- 
Loes Koelewijn
PhD Candidate
Macquarie Centre for Cognitive Science (MACCS)
Macquarie University
Sydney NSW 2109
Australia

Ph:   +61 2 9850 4135
Fax: +61 2 9850 6059
email: loes.koelewijn at mq.edu.au
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20110330/b5542357/attachment-0001.html>