Signal Processing: EEG to ERP

1
Signal Processing EEG to ERP

• A Darren Parker Presentation

2
EEG to ERP
To get an ERP signal out of the raw EEG, our goal
must be to reduce the signal to noise ratio.
3
How can we reduce noise?

• Averaging trials in the same condition together.
• Filter out frequencies that are not of any
  concern. (i.e. 60Hz noise)
• Rereference data so channels become more
  uniformly emphasized with relation to the
  reference channel.
• Reject bad trials.
• Correct for artifacts in the data, like blinks.

4
Filtering
5
Filtering

• Waveforms can be broken down into a series of
  sine waves at different frequencies.
• Our filters separate the signal into the
  individual sine components, then remove the
  components we arent interested in.

6
Rereferencing

• Need a reference for each electrode to be
  measured relative to.
• Electrodes near the reference will show little
  signal, since there is not a great deal of
  difference between them.
• Electrodes far away will be overrepresented since
  they are far away and much more different.

7
Rereferencing
Cz Reference
Rereferenced
8
Rejecting Bad Trials

• Sometimes parts of the waveform are just messy,
  due to subjects moving around in one way or
  another.
• Some artifacts are common and stereotyped, such
  as blinks, and sometimes eye movements.
• Potential for correction factors.
• Others are more random with no predictable
  pattern.

9
Rejection
Blink
Eye Movment
Garbage
10
Eye Blink Correction

• Blinks and (sometimes) eye movements are common,
  and an individuals blinks and eye movements may
  be fairly constant.
• We can try to build a prototype for a subject, so
  we can correct for blinks (and eye movements).

11
Averaging

• The EEG records all changes in potentials at all
  electrodes, not just the signals we are
  interested in. There will always be additional
  noise since our subjects arent dead.
• Assuming that other processes are random, and not
  time-locked to the stimulus, averaging trials
  together in any given condition will increase the
  signal and decrease any random noise.

12
Averaging
Both pictures taken from http//neurocog.psy.tuft
s.edu/images/erp.htm
13
The Plot Thickens

• By now you should have your perfect little ERP
  averages but is everything really that perfect?
• Bad channels affect averaging and rereferencing.
• Filtering is not perfect.
• What reference is best?
• Latency jitter.
• Artifact correction woes.