[FieldTrip] scalp EEG resting-state analysis
gabriel.delara at med.uni-goettingen.de
Fri Jul 27 17:04:33 CEST 2018
Dear Field trippers,
My first e-mail to this list. Thanks for paying attention to this, whoever took time to read it.
I am currently doing my PhD in Göttingen, at the moment running a scalp-EEG (26-channels) analysis with Fieldtrip for a dataset of healthy participants.
I have a repeated-measures design, where each participant undergoes 4 sessions, with n = 20.
For each session, I performed two 5' eyes-open resting-state recordings per participant: one before and one after intervention (pre-, post-).
The goal is to simply detect power changes in the pre X post comparison.
Recordings were segmented into 2s trials with 50% overlap followed by other pre-processing steps.
I am having some doubts on the feasibility/correctness of the steps after the pre-processing. My main critical question is the procedures sequence to correctly run the Monte Carlo stats in the end (if and when to keep trials, to average trials, to average across participants, and so on).
Some options available:
1. First running TFA (mtmfft, dpss,) keeping the trials. Then, appending participants with ft_appendfreq. Afterwards, stats: parameters (Montecarlo, depsamplesT, cluster correction, 5.000 iterations). First problem arises in the design matrix: Since trials are kept in the FA, they are uneven for every session (appending leads to different trials between pre X post), and also the appended data does not keep the participant as UO, so design matrix has to be design as trials in the first row. To make it work, trials have to be shortened from one condition, so they are paired (which seems obvious for depsamplesT, since it does would not accept different values in the uvar). Statistics run after doing all this, but I am not sure, since having a uvar with ~10.000 points, that were the individual trials of every session (there are 20 participants) sounds strange (there were no 10.000 participants). This option was run for one condition, significant difference happened, clusters were formed, etc. Is this reasonable?
2. First running TFA (mtmfft, dpss,) keeping the trials. After, apply ft_freqdescriptives to every subject, in oder to be able to use ft_freq_grandaverage. Apply ft_freqgrandaverage across participants, but keeping individuals (for the design matrix).After ft_ freqgrandaverage, stats as described above, with design matrix with 1:20 in the first row (uvar) and the two conditions (pre X post) at the second. This works, but leads to no clusters found (ran it on more than one condition). Doubt here is: using ft_freqdescriptives and then ft_freq_grandaverage compromises the monte carlo procedure, having just one value for each channel (the grand-averaged trials power)?
Gabriel Amador de Lara
wissenschaftlicher Mitarbeiter / PhD candidate
Klinik für klinische Neurophysiologie / Department of Clinical Neurophysiology
Universitätsmedizin / Medical Center
Robert-Koch-Str. 40, 37075
+49 0551 3919265
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the fieldtrip