Pre-processing: Bad Channels Through Baseline Correction

Return to Video Tutorials

Video Navigator

  1. Bad Channels (27s)
  2. Dealing with Bad Channels (1m 36s)
  3. Re-referencing (3m 07s)
  4. Why the Reference Matters (3m 37s)
  5. Bipolar Channels (5m 10s)
  6. Epoching (6m 19s)
  7. Baseline Correction (7m 38s)
  8. Why to Perform Baseline Correction (9m 12s)
  9. Process in Matlab (10m 15s)

Bad Channels

  1. When an electrode is not collecting data or is collecting very noisy data
    1. Flat lines
    2. Channel that is moving drastically up or drastically down (data is going out of range)
    3. Very noisy data ("hairy catterpillar)
  2. Recommendation:  If more than 10% of your data is from bad channels, you ought not to use that data

Dealing with Bad Channels

  1. Option 1 - Remove the Channel
    1. Best when a small set of electrodes (32 or under)
    2. Cannot use that participant IF the channel you are dropping is one of interest in the study
  2. Option 2 - Interpolate the Channel
    1. Best when a larger set of electrodes (64 or more)
    2. Uses information from other electrodes to make the best prediction of what would have been seen at that electrode
      1. Electrode signals are NOT independent of each other; the closer they are on scalp the more similar their signal tends to be

Re-referencing

  1. Re-referencing allows you to change the reference type from whatever was used for the recording
    1. Sometimes recordings are done with an unknown reference, no reference, or you want to change it
  2. Multiple ways of referencing
    1. Mastoid, Average, etc

Why the Reference Matters

  1. Electrodes measure neural activity across the scalp but activity closest to the electrode is strongest
  2. If the reference is too close to the electrode of interest, can subtract out critical neural activity
    1. Example: A mastoid reference is too close to the critical electrodes for N170 (a component related to face processing, generated in the the temporal cortex). A mastoid reference would attenuate the N10 activity, making an average reference more desirable to use, as the amplitude of the N170 will be much larger.
    2. Be aware of the reference typically used for your component of interest
  3. References can also change the polarity of the signal

Bipolar Channels

  1. Bipolar channels can be created to make eye artifacts easier to detect.
  2. They are created using simple math like subtraction because eye channel pairs are opposite in polarity.
  3. A Bipolar VEOG channel creates a channel that doubles the amplitude of vertical eye movements & blinks
    1. Math: VEOG Lower – VEOG Upper (or whatever electrode is above the eye).
  4. Can you guess what the equation would be for a bipolar HEOG channel?
  5. A Bipolar HEOG channel creates a channel that doubles the amplitude of horizontal eye movements.
    1. Math: HEOG Right – HEOG Left

Epoching

  1. EEG recorded as continuous signal for the duration of the experiment
  2. Event codes/triggers/markers indicate occurrence of stimuli and responses
    1. Useful for breaking the data into segments
  3. Epoching extracts fixed-length segments of data from continuous EEG
  4. Epochs are time-locked to the event codes of interest, which is necessary for ERP averaging
  5. Epoch includes
    1. Baseline period: Typically 100-200ms before the event code
    2. Trial period: Time after the event code (e.g., 500-1500ms)

Baseline Correction

  1. Baseline Correction helps adjust our data, because there is stuff going on before our event
  2. Recall that prior to averaging we extract segments (epochs) of the EEG surrounding target events (all stimulus types)
  3. Estimate noise in baseline period and remove these "uninteresting" signals from waveform
  4. Subtract Mean pre-stimulus voltage from each point in the waveform.
    1. Adjusts the waveform so that we look at the data as if it started at 0 (lets us examine the data segments as if they started from the same point)
  5. Baseline period can also be another chosen period designated as the baseline
    1. E.g., for response-locked ERPs, we might choose a baseline time period pre-response
  6. Baseline Correction is typically performed after epochs are extracted from EEG data

Why to Perform Baseline Correction

  1. Because anything that influences or creates noise in the pre-stimulus (baseline) period will also exert influence on our post-stimulus amplitude measurements.
  2. Baseline Correction addresses drift caused by skin hydration, skin potentials and static charges
  3. Works well to minimize voltage offsets and gradual drifts
  4. Information prior to the target influences voltage but is unrelated to the condition being studied

Process in Matlab

  1. Open Matlab, Set Paththe eeglab plugin, and open eeglab
    1. More in depth instructions are provided in the Data Processing Stream and Pre-processing: Filtering videos
  2. Import raw data using the brain vision plugin, and selecting the .vhdr file
  3. Filter the data with a Butterworth filter from .1 to 30 Hz, and save the dataset
  4. Plot the data and look for any bad channels
    1. Make sure to adjust the scale, and scroll through the various timepoints
  5. Re-reference the data (PURSUE uses option 2)
    1. Option 1: automated process in eeglab
      1. EEGlab > Tools > Re-reference the data > Choose the reference you want (average or specific channels) and make sure to retain the reference if necessary
    2. Option 2: Use mathematical equations
      1. ERPlab > EEG Channel Operations > Load list > select the file of equations > Run > save the dataset to the program and somewhere on your computer
        1. Either create the mathematical equations, or load a file that has them
        2. The equations generate the mastoid reference and subtract that from each channel, as well as generate the eye channels
        3. The save tag _chop stands for channel operations
  6. Epoch the data
    1. Create an event list to see what codes are within the data
      1. ERPlab > Eventlist > Create EEG EVENTLIST > Export the list and save it on your computer
        1. Add code for boundary events
        2. If a warning comes up, select "Overwrite them"
    2. Binning: Tell the system what the events mean
      1. ERPlab > Assign bins (BINLISTER) > select BDF file (Bin Descriptor File) specific to that experiment
        1. Read EVENTLIST from the Current Dataset
        2. Write resulting EVENTLIST to Text file, and name the file
        3. Run
        4. Save the dataset
    3. Check the bin information
      1. ERPlab >BDF Visualizer
      2. Load eventlist (which you just created in the previous step)
      3. Load BDF file
        1. Loads information on how your codes are related to your conditions
      4. Analyze BDF
        1. Shows how many total event codes are present, as well as how many codes are in each bin
        2. This is a check step
    4. Epoch the Data
      1. ERPlab > Extract bin based epochs > Run
      2. Will extract the epochs and do baseline correction at the same time
      3. Make sure to adjust the window to your epoch, and select the baseline correction option
      4. Save the dataset
  7. Look at the epoched data
    1. Tools > Channel Data (scroll)
      1. We now see dotted lines where there are breaks within the trials

Additional Information

PURSUE teaching modules provide instructors with everything they need to add EEG/ERP content in existing courses, teach a full semester course, or train research assistants in the lab. Follow this link to Lab Training Modules that can be used with tutorial videos.

Associated Teaching Modules: Lab Training Modules, Bad Channels Through Epoching

Download Training Labs