Lecture 2 - CNS Flashcards
modern EEG device components
A/D converter battery USB relayer acquisition computer stimulus computer
why is EEG with battery
main would create electrical noise
to exclude risk of electrical shock
why two computers for EEG measurment
to ensure both have enough resources
less of a problem with modern computers
Nyquist rate
you sampling rate has to be at least twice as high as the highest expected frequency
in practice it’s usually 3 to 5 times as high
Nyquist frequeny
the maximal frequency you can reliably extract with your sampling rate
i. e. sampling rate / 2
aliasing
when sampling rate is too low
high frequencies get recorded as low frequencies
sample connects peaks to minimums
CD sampling rate
44 kHz
because lowest distinguishable frequency is 22 kHz
greatest sources of noise in EEG
muscle movements, e. g. chewing
blinks
lateral eye movements
drifts
noise from mains
50 hz in Europe, 60 in us
slow voltage drifts
due to electrode polarisation, sweat etc.
pre-processing pipeline
- re-referencing
- filtering
- ocular correction with ICA
- artifact rejection
re-referencing
changing the location of the reference electrode
the artificial zero, that is never actually zero
common reference electrode locations
most common is mastoids
also earlobes
or tip of the nose
common mode noise
when impedance differs between active, ground and reference electrodes
the higher the impedance the more common mode noise
solution: high impedance input amplifiers
getting rid of common mode noise
(A-G) - (R - G) = A - R
= active electrode - reference
pattern decompostion
location of reference electrode does not matter
high-pass filer
lets only high frequencies through
low-pass filter
lets only low frequencies through
notch filters
filter out specific frequencies
band-pass
removes a range of frequencies
between two defined points
band-stop filters
attenuated frequencies between x and y
disadvantage filters
always change the signal
throw out data
can change shape of the ERP
ocular correction
with ICA
or Gratton & Coles method
VEOG and HEOG
VEOG
vertical electric oculogram
one electrode above, one below the eye
HEOG
horizontal electro oculogram
one electrode lateral of each eye
Gratton & Coles method
regressing out blinks
with linear regression
problem: might remove actual brain activity from VEOG
ICA
independent component analysis
for approximating solution to superposition problem
blinks and eye movements are biggest contributors
therefore ICA works good on them
number of component = number of electrodes
assumes normal distribution of random activity
everything non-normal is something important
rotates distribution to best fit the observation
after ocular correction
components can be removed manually
ICA can be repeated and components can be thrown out as much as you want
artifact rejection
removing whole trials containing artifacts
throwing data out instead of correcting it
can be done automatically or by visual inspection
