Description and analysis plan for the EEGManyPipelines project
detect bad channels by RANSAC
, interpolate bads. note down how many channels were interpolated per participant.
Re-referencing to average electrode (reasoning: to approximat electrode reference that is most generalizable to all electrodes)
- Band: 0.1-100 Hz (FIR). Before/after epoching is not relevant, as jitter is only +-4ms
- downsampling to 250 Hz (as for ERP no relevant frequencies are higher than that, and it's questionable if higher frequencies can even be measured reliably in EEG)
- For ERPs: 35 LP + Notch 50 Hz
- For Open questions: leave original LP
- ICA, find Eyeblink components from 64 components
- blinks will be easy to detect
- horizontal eye movements will be more difficult
- check manually!
- Only apply ICA correction to time spans where eye blinks / heog are actually observed
- Apply to all channels
- Are eyeblinks corrected? Visual random sample check
-
Epoch data based on trial trigger -200 to 500 ms
-
Apply artefact criteria per trial
- same as Bublatzky et al 2020)
- maximal allowed voltage step of 50 mV/ms;
- maximal allowed difference of values in 200 msec intervals of 200 mV;
- minimal/maximal allowed amplitude of ± 100 mV
-
good/bad trials measurement, percentage bad per participant
-
calculate good/bad channels percentage per trial type / hypothesis interest
Data at this point:
epochs_freq
- from -200ms to +500ms after trigger onset
- baseline corrected from -200ms to 0ms
- bad epochs marked by
autoreject
- Original sampling frequency of 512
epochs_erp
- same as above
- additional LP 35 Hz
- downsampled to 250 Hz
There is an effect of scene category (i.e., a difference between images showing
man-made vs. natural environments) on the amplitude of the N1 component, i.e. the
first major negative EEG voltage deflection.
Using epochs_erp
- Don't use peak amplitude, because not reliable, as peak cannot always be determined accurately
Find best electrode to detect the N1
- make avg of trials per subj per electrode
- Find first negative peak 50-120ms, see which electrode has lowest amplitude
- use this electrode for further calculations
calculate component
- calculate voltage per participant for chosen electrode
- time windows for component min to max of peak +-5ms
- AUC +-5ms of peak, take one time window for all participants
- repeated measures ttest
Results:
- Negative peak between
0.08
and0.15
on grand average is at125ms @ Fz
. - voltage +-5ms:
man-made = -3.09+-1.86 uV
natural = -2.76+-1.87 uV
- repeated measures ttest (
ttest_rel
)p = 0.001187815
- Cluster analysis showed significant cluster on occipital electrodes at a later point
There are effects of image novelty (i.e., between images shown for the first time/new
vs. repeated/old images) within the time-range from 300–500 ms ...
a. ... on EEG voltage at fronto-central channels.
b. ... on theta power at fronto-central channels.
c. ... on alpha power at posterior channels.
Using epochs_erp
H2a Calculate
- Take Fronto-Central channels for calculation ('FCx')
- AUC 300ms-500ms of peak
- repeated measures ttest
Results
- Negative peak between
0.3
and0.5
on grand average is atFCz
- Taking electrodes
['FCz', 'FC1', 'FC2']
- voltage from
300-500ms
on these electrodes:new = -6.40+-3.50 uV
old = -5.99+-3.52 uV
- repeated measures ttest (
ttest_rel
)p = 0.000001946596
H2b Calculate
- calculate spectra for theta (4-7 Hz) for the time range using Wavelet
methods:
For each participant:
For each epoch of participant:
1. calculate wavelet with
- freq = array([4. , 4.5 , 5. , 5.5, 6., 6.5, 7.0 ])
- use half cycles
2. take segment between 300-500ms for fronto-central electrodes
3. calculate mean over result
Result
- For electrodes
[FCz, FC1, FC2]
- Mean Power between 300-500 ms
new 7.06e-10+-2.59-10
old 7.27e-10+-2.53e-10
p = 0.000802
Problems:
- How to normalize histograms? -> doesn't matter bc within-subject and same sessions
- What to do with outliers?
methods:
For each participant:
For each epoch of participant:
1. calculate wavelet with
- freq = array([8. , 8.5, 9. , 9.5, 10. , 10.5, 11. , 11.5, 12. , 12.5, 13. , 13.5, 14.0])
- use half cycles
2. take segment between 300-500ms for posterior electrodes
3. calculate mean over result
Result
- For electrodes
[PO7', 'PO3', 'POz', 'PO8', 'PO4', 'O1', 'O2']
- Mean Power between 300-500 ms
new 5.33e-10+-2.09e-10
old 5.32e-10+-2.04e-10
p = 0.85
There are effects of successful recognition of old images (i.e., a difference between
old images correctly recognized as old [hits] vs. old images incorrectly judged as new
[misses]) ...
a. ... on EEG voltage at any channels, at any time.
b. ... on spectral power, at any frequencies, at any channels, at any time.
- plot grand avrg and differences between categories
- for later: divide in time windows, calculate statistics, use cluster analysis / stats corrections to check where differences are significant
- mass univat ansatz
- cluster correction from oostenfeld
- Take repeated measures F test
- Report all channels and all timepoints that are in cluster
Clusters based on TFCE
Clusters based on 0.05, two clusters found
Cluster analysis using 1 sample ttest reports three clusters that stretch across time and frequency and channels
4. There are effects of subsequent memory (i.e., a difference between images that will
be successfully remembered vs. forgotten on a subsequent repetition) ...
a. ... on EEG voltage at any channels, at any time.
b. ... on spectral power, at any frequencies, at any channels, at any time.
same as above
Cluster analysis found no cluster that was significant with p<0.05.
pvalue map. darkest blue=0.2
No significant clusters.
-
Should timepoint/electrode be decided on ALL epochs or only on epochs relevant for the current H1?
-
man-made
contains all different kind of conditions for H1, combine?
'man-made/new/correct-rejection/N/A': 158
'man-made/new/correct-rejection/forgotten': 94
'man-made/new/correct-rejection/remembered': 24
'man-made/new/false-alarm/N/A': 15
'man-made/new/false-alarm/forgotten': 18
'man-made/new/false-alarm/remembered': 5
'man-made/old/hit/N/A': 131
'man-made/old/hit/forgotten': 102
'man-made/old/hit/remembered': 6
'man-made/old/miss/N/A': 10
'man-made/old/miss/forgotten': 25
'man-made/old/miss/remembered': 3
- spectogram: averaging before spectogram creation or after?
- Equalize event counts before averaging?
- Resampling later?
make difference topo plot
- create difference on subj level
- sanity check for f test: calcualte on previous results, sohuld give same p val
EMP20 weird oscillations EMP21 noisy