[FieldTrip] Attempt at cluster analysis on seed to whole brain coherency values

[Eelke Spaak]

Dear Paul,

That all sounds good to me. Note that you could also have specified
cfg.parameter = 'cohspctrm' in the call to ft_freqanalysis, so the
renaming step to powspctrm was not strictly necessary.

One thing you'll want to keep in mind when doing this is that
imaginary part of coherency (ImC) is a biased quantity and depends on
the number of trials. So comparing two groups statistically will not
give you meaningful results if the number of data points for the two
groups is unequal. It's worth having a look at this paper which goes
into more details on this and related issues:

Vinck, Oostenveld, Van Wingerden, Battaglia, & Pennartz. An Improved
Index of Phase-Synchronization for Electrophysiological Data in the
Presence of Volume-Conduction, Noise and Sample-Size Bias. NeuroImage
55, no. 4 (April 15, 2011): 1548–65.
https://doi.org/10.1016/j.neuroimage.2011.01.055.

Cheers,
Eelke
On Mon, 15 Oct 2018 at 04:37, Paul Dhami <pdhami06 at gmail.com> wrote:
>
> Dear Fieldtrip,
>
> I have a dataset in which I'd like to compare, between controls and patients, their imaginary coherency values in a seed to whole brain manner. In other words, I would like to calculate the imaginary coherency between a single seed electrode and the rest of the remaining electrodes, and ultimately use ft_freqstatistics' cluster analysis to test for a significance difference between the groups' seed-to-whole-scalp coherency maps.
>
> From my understanding, this would require a bit of hacking to implement this. I attempted to do so and describe what I did below, with the goal to eventually create something that would work with freqstatistics :
>
> ran freq analysis with output as 'powandcsd' and method as 'mtmconvolv'
> on the results of freqanalysis, ran ft_connectivityanalysis with cfg.method = 'coh' and cfg.complex = 'absimag'
> with each subject's resulting structure from ft_connectivityanalysis, I first chose a seed of interest
> I then found the indices of the 59 channel combinations in relation to the seed of interest in labelcmb
> Using the indices, I then pruned/removed the remaining channel combinations of no interest from both the labelcmb and cohspctrm, reducing cohspctrm to a channel of interest (59) x frequency x time matrix (as in it only included values for the channel combinations of interest)
> I rename the dimord as 'chan_freq_time'
> create 'label' field with standard electrode labels
> I then create powspctrm in my structure, which holds the exact same data as the cohspctrm (created powspctrm strictly for freqstatistics)
> removed the fields of 'labelcmb' and 'cohspctrm'
> because my powspctrm is missing the seed channel, I then inserted a matrix of ones with the appropriate dimensions into where it should be (e.g. if my seed of interest was channel FCZ, I would then insert into my matrix of ones into the 19th position of the powspctrm, thus shifting the latter matrices so now it becomes a matrix with 60 channels in the appropriate order)
>
> From my understanding, the resulting structure of each subject should contain now the coherency values between the seed of interest and the rest of the electrodes in 'powspctrm'.
>
> I then used freq_statistics (in the standard way) with cluster correction to compare the coherency between groups, and from what I can tell with no errors popping up, it worked. I then interpreted the resulting clusters in a similar fashion as you would do for a typical frequency chan-freq-time analysis (instead of power, looking at coherency clusters now).
>
> My questions are:
> 1) In regards to implementation (assuming something like this can even be appropriately implemented), do things look okay?
> 2) Am I wrong in thinking that the cluster results of freq_statistics can be interpreted in a similar fashion as to a typical frequency chan-freq-time analsyis (just replacing power with coherency)?
>
> Sorry for the long-winded email, but any help would be greatly appreciated.
>
> Thank you,
> Paul
> _______________________________________________
> fieldtrip mailing list
> https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
> https://doi.org/10.1371/journal.pcbi.1002202