User's Guide‎ > ‎Analysis‎ > ‎

A - Spontaneous EEG

Spontaneous EEG with Cartool 

Original contribution from Verena Brodbeck & Lucas Spierer


Single trial / individual subject and/or group level Cartool analysis of spontaneous thinking associated with mental imagery, maintenance in short term memory, etc.


Experimental design used in this tutorial: Delayed match to sample task between sequences of sounds separated by a 4 second interval. Epochs of interest are the 4 second time periods over which subjects have to maintain in memory a given dimension (i.e. the conditions) of the sound sequence to resolve the same/different delayed match to sample task (for another example of spontaneous eeg design and for theoretical background, see e.g. Lehmann et al; 1998; Brain electric microstates and momentary conscious as building blocks of spontaneous thinking: I. Visual imagery and abstract thoughts; Int J Psychophys; Jun;29(1):1-11).


1.         Take your raw EEG data!


2.         Extract triggered epochs of interest (via Averaging tool, output one file/epoch (here 60X4s long epochs/condition) and filter (0.1-40Hz + eventually 50Hz Notch).


3.         Down-sample (here: from 1024 to 102.4Hz, via Export tracks) each extracted epoch of interest / condition, and eventually interpol bad electrodes.


4.         Scan each down-sampled/interpolated epoch for GFP peak (Marker->scanning track to generate marker, 1TF@100Hz minimum gap (gives ~80 peaks/4s period).


5.         For each condition, glue together each GFP peak TFs of each epoch to get one long epoch/condition (Averaging, 1TF@100Hz post MaxGFP marker, output = All epoch in one file, gives ~5000TF long epoch (=~80*60) per condition).


6.         Use the file done in step 5. as input for the segmentation (ignore map polarity;  as AAHC will crash cartool buffer due to very long epochs, use K-mean). 


7.         Do the fitting on each extracted epochs/condition (i.e. output of step 3) and/or on each extracted epoch/condition with only GFP peak (export step 3 output, all tracks, and MaxGFP markers in the trigger slot of the Input time period).


8.         Do your statistic on the output of the fitting (on #TFs or GEVs);

Q: Are there different maps better representing each conditions?

  • Individual level analysis: Univariate ANOVA / multiple independent t-test (i.e. intra-subject variability). For t-tests: Is e.g. mapN more present/ explain more variance in Condition 1 than in condition 2?
  • Group level analysis: Group level analysis can also be done by using the step 5 output of all condition and all subjects in the same segmentation (then repeated measure ANOVA and paired t-test on the fitting output, inter-subject variability).

9.         Do your inverses on topographies (maps) of interest / each condition.


10.      Average and/or do your stat on each inverse/condition/subject (i.e. possibly go to group level analysis for source analysis).



-       Down-sampling of step 3 was necessary here to reduce the size of the data



-       Alpha oscillations: oscillation will dramatically influence GFP peaks marking, particularly in design where the subject are eye-closed during the period of interest. Although microstates occurrence is independent of oscillatory activity, you should check that alpha power is similar in each condition over your period of interest.

-     the choice of 10 ms minimum gap in the step 4., for the GFP marking, may have a strong influence on the number of marked GFP peaks as well as on when they will be detected and thus have a strong influence on subsequent analysis. The choice of this minimum gap is driven by studies indicating that periods of stable topograph below 10ms probably don't have physiological significance (Guthrie and Buchwald, 1991; see also literature on functionnal microstate for this question).

-       The same microstate/topography re-occurring two or more time over a given time period not necessarily reflect the activity of the same underlying neural network (where two different topographies necessarily signify two different underlying networks).

-       The frequency band used in the filtering may also influence your results

-       You should check whether the activity recorded at the beginning or at the end of your period of interest relates to the same process (potential differences may for example be reflected by more GFP peaks found -and marked- during e.g. the first or second half of your time period of interest).




Step 2: 

Step 4:


 Step 5: 



Step 6:

 Step 7: