EEG Flashcards
Why is it important to shield n form electrical noise in the environment?
• The signal we get with EEG is already tiny
• Noise from the environment is much bigger and would
significantly distort the signal
• shields n from this noise
How does oscillatory activity outside of n affect the signal picked up by the electrode
- Any oscillatory activity produces a magnetic field
- This magnetic field is picked up by the electrode
- Signal electrode picks up form this > than signal we are interested in
Can you describe the layout of the cage?
Cage – describe the layout?
- two screens one is inside the shield (digitization) and one is outside (computer sending stimuli)
- The outside screen sends stimuli into the inside screen
- Key = both screens are linked, stimulus screen sends pulse to digitization screen saying NOW! And links data to the sitimuli
- There is also the electrodes and amplifier in the cage and perhaps a response button pad for n
Why do we need multiple electrodes?
- EEG picks up on voltage, voltage is the electrical potential of current moving from A to B.
- Can’t be measured from one place
How many electrodes are needed in EEG? What are they
• EEG systems are differential amplifiers
• Need a minimum of 3 electrodes to work
> Active electrode – placed at a region of interest
> Ground electrode – placed somewhere convenient, it is linked to the ground circuitry of the amplifier
> Reference electrode – anywhere on the scalp, not region of interest
I though we just need to measure current moving from A to B? Why are 2 electrodes not enough?
- The ground electrode is linked to the ground circuitry of the amplifier
- Every electrical device has a ground circuitry – very noisy!
- We could look at the voltage between A and G and yes this would tell us something about how the current moves from A to B
- But it would be infiltrated by the noise of the ground circuit – we don’t want that
- Adding a third electrode allows us to subtract the noise
I though we just need to measure current moving from A to B? Why are 2 electrodes not enough?
- The ground electrode is linked to the ground circuitry of the amplifier
- Every electrical device has a ground circuitry – very noisy!
- We could look at the voltage between A and G and yes this would tell us something about how the current moves from A to B
- But it would be infiltrated by the noise of the ground circuit – we don’t want that
- Adding a third electrode allows us to subtract the noise
How does adding a reference electrode subtract the noise?
- Amplifier calculates the difference between: A-G and R-G
- Effectively gives us the difference between A and R, removing the noise from the ground circuitry
Describe a channel
- The three electrodes are combined into a single channel
- The waveforms we see on the screen are the potential from A-G and R-G
- So we look at the electrical flow of current in the brain using three electrodes
Why shouldn’t we have the reference site on the body?
- Must be on the scalp
- If on the body – picks up a signal produced by the heartbeat called EKG
- Dipoles we want to measure are tiny, signal from the heartbeat is massive
- Would massively interfere with the signal
Where should we place the reference electrode?
- Placed at neutral site ideally to allow us to see the flow of current towards the region of interest
- This is because the amplifier looks at the difference between the active and reference site, taking activity from both areas in to consideration
- ERP reflects activity from both sites
- We want a site with the least noise – otherwise will influence the results
what are some common reference sites used?
- tip of the nose (annoying for n)
- chin (annoying)
- mastoids (most common)
- earlobes
things to consider when choosing a site for the reference electrode
- Electrically neural as possible
- Convenient for n
- Not biased towards a particular hemisphere
- Are you comparing your findings with another study? Important to use the same reference site
- Where you place it in relation to the active electrode
When chosing a refernce site why do you need to consider how far it is from the active electrode?
- The location of reference affects the size of the thing we are measuring
- if both nearby, on the same hemisphere then the result (the voltage between both areas) wont be so big
- thus the absolute size of the ERP will differ based on the reference site used
Average mastoid reference – why can’t you just use one of the mastoids?
- Left mastoid or right mastoid
- Reference is biased towards one hemisphere
- Would lead to an imbalance In between the active electrodes in the left and right hemisphere
How do we calculate the average mastoid?
- How is this done – during recording you couple the active electrode and one of the mastoids (e.g., left) and reference this with the other (right) – after recording EEG is pretty cool – allows you to now play around with stuff. Thus after the recording is complete we calculate the average mastoid
- Subtract half the difference between the LM and RM from the active electrodes.
- = average mastoid reference
What does digitization do
- converts continuous signal (waveform deflections) into a series of numbers
- the more numbers the better
- the numerical range depends on the resolution of the digitizer
What is a digitizer government name
analogue-to-digital converter (ADC)
how does the digitiser convert a deflection into a number
• Record different voltage values
• Resolution of 12 bites = records 2 to the power of 12 or 4096 voltage values
• waveform goes up and down on the y axis with say 4096 diff points on the y axis (for 12 bite resolution)
• Amplifier is set to measure the difference between -5 and +5 microvolts – divided into 4096 points.
- 4096 is coded as +5 and 0 is coded as -5. The numbers are coded as somewhere between 0 and 4096.
What is “amplifier blocking”
- Something that prevents us from recording values above +5 (4096) or below -5 (0)
- If a voltage is out of this range it will be represented as either 4096 or 0
- Gain of the amplifier is set so these values it not exceeded – set to a range that makes sense. Otherwise, the values we get would not be meaningful. All it would tell us is if a value is smaller or higher than another
- Most EEG systems have 24 bites of resolution so the issue of amplifier blocking isn’t such a problem today
What does the Y axis of a waveform reflect?
- The size of the potential
* Measured in microvolts
What does the x axis of a waveform reflect?
time
How can we convert to a signal into a waveform
- So each time is associated with a value between 0 and 4096 (depending on the resolution of the amplifier), at another time point there is another value between 0 and 4096
- This is the waveform is constructed
Why do we have so much information with EEG waveform
- Because each time point is associated with a value between 0-4096
- And at each time point this isn’t just the value associated with one but 64 channels
- Very quickly becomes a lot of information
How can we reduce the information load of the continuous EEG signal
- Sampling
- continuous EEG signal is converted into s sequence of discrete-time points
- lower number of time points for each of the 64 channels
what is the sampling period?
• Time between consecutive samples
What is the sampling rate
- Number of samples per unit of time (seconds)
- Measured in Hz
- 500Hz = 500 samples per second
How do we determine the sampling rate is?
- Depends on the frequency content of the signal we are recording
- Must be at least 4 times faster than the highest frequency you have
How do we extract the ERP from the EEG signal?
- Activity is time-locked to a specific event
- We can then average together different signals all time-locked to the same event
- The background noise is random and would cancel each other out
- Would leave behind only the signal of interest
What things can cause noise in EEG signal?
> electrical activity that is not the signal of interest
Electrical signals form the environment
Non EEG biological signals
Voltage fluctuations measured by EEG are tiny, by how much do we amplify the signal so that it is detectable?
• 10-50k
What do we mean by electrical noise from the environment?
- Electrical sources produce electromagnetic fields
- this can be picked up by the electrodes on n’s head
- any electrical device induces quite a dramatic voltage change in the signal
- much bigger than the one where inrteresyed in
How exactly do electrical devices affect the EEG signal ? explain this using a cable as an example of the electrical device in the room.
- Cable = is a conductor with an oscillatory voltage inside
- Brings with it an electromagnetic field
- This produces voltage changes (electrical noise) in n, the electrical wires, and the electrodes
> Oscillating voltages influence voltages in nearby conductors – this electrical noise will be observed in the EEG.
What are the two major sources of electrical nosie
- AC current line - induces 50Hz – causes oscillations in the EEG of exactly 50Hz.
- Video monitor - these have a refresh rate, they will produce 120 new images every second. Here you have an oscillation of 120Hz
- Ac line noise is sinusoidal, noise from monitors = more spikey.
Name 2 ways to reduce background noise
• Passive shielding – conductive metal that surrounds to-be-shielded regions.
- this will cancel out electromagnetic fields.
- Rooms, cables – both can be shielded
• Electrodes with front end amplifiers
How do electrodes with front end amplifiers boost the signal-to-noise ratio?
- Normally, the active electrode sends a signal through a cable to an amplifier and the signal is boosted
- But here, a small amplifier is placed near the electrode, amplifying the signal before you send it somewhere else.
- if you have a tiny signal and noise interferes with it, the effect is strong. If you have a much bigger signal and that is then faced with noise, less interference
what are the disadvantages of front-end amplifiers
- Expensive
- Must fiddle with it a lot and they break easy
- imagine you have 64 electrodes and then 64 tiny amplifiers sitting with them, it’s a bit much
what are non EEG signals (biological noise), name 3 examples
Internal noise comes from the body and cannot be eliminated
• skin potential
• blinks
• muscle movements
why are post processing techniques not perfect
• because they always distort the original signal
how can we minimise biological noise
- getting n as comfortable as possible.
- Any muscle activity, stiff neck, clenched teeth will produce muscle activity, electromyography which is itself a large signal
- muscles you have are larger than the generators in the brain - produce much bigger electrical signal
how can we deal with artifacts in the data
- artifact correction – correcting data
* artifact rejection – just getting rid of contaminated trials
where could electrodes pick up EKG signals?
- on the body
- on the head
- E.g. by placing an electrode on the blood vessel which pulsates with the heartbeat
What waveforms might emerge that are NOT artifacts (e.g. caused by muscle movment)
- Those appearing when n becomes sleepy
- results in slow waves that are bigger in amplitude.
- Within a certain frequency range – alpha frequency – which is between 8 and 12 Hz.
Why might sometimes we not be able to do artifact rejection?
- Because the artifact might be produced in a systemic manner
- EKG – happens every second – cant throw away the whole data set
What is the most common artifact ?
- Eye blinks
- induce massive change in amplitude most pronounced over electrodes over the front of the head
- this is bc the eye is a small dipole. Each of our eyes has an electrical gradient.
- This has a positive end at the front and a negative at the back. Each time we move our eyelids – changes modulation of the electrical potential of eyes to surrounding areas.
- (if we close our eyelids, it changes this electrical field).
How to know if you have a blink artifact?
you will have a characteristic scalp distribution – helps to discern them
how can we correct artifacts? E.g. blinks
- fourier transform
* filtering
What is Fourier transform?
- waveform dispelled into sine waves of different frequencies
- then Fourier Transform is a technique that plots each individual sine wave
- then can eliminate specific frequencies. E.g. I don’t want that low frequency (eye blink) and I cut that frequency from the data.
- then go back to see the full dataset without it.
What is filtering?
- Remove frequencies we are not interested in
- if devices produce 50 Hz artifacts and the signal that we are interested = between 8-30 HZ - just filter out anything over 30Hz.
What is low or high pass filtering?
- a type of filtering
* anything below X hz we keep, and everything above it – we delete. Vice versa
After you have got the data and eliminated the noise as much as possible whats next?
- Signal averaging
- EEG data in each trial contains both the ERP waveform and random noise.
- Averaging enough of these trials together will considerably reduce this background noise.
What is signal averaging?
- The extraction of the signal specific to the event from the EEG signal
- Time locked manner
- then average together the waveforms generated at that time in a point-by-point manner
- begins to emerge a kind of waveform
- because the background noise is eliminated
is it the more trials the better the signal to noise ratio? Would 1 mil trials then be a dream?
- No – yes the more trials averaged together boost the signal to noise raito
- But after a certain point – addition of more trials becomes less and less effective
What would a variable latency affect in the waveform?
- The peak
- Jitters in latency between trials will reduce the amplitude of a peak in the average
- Results in different ERP amplitude!
Name a scenario in which varied latencies in activity across trials would be problematic (limitation)
- comparing the effects of two groups, and one has more jittered latencies than the other e.g. in a study of cognitive ageing
- Older groups might have more latency variability.
- Simply by averaging trials together you cannot see the difference between two groups ERPs
- need to analyse individual frequencies - other analysis techniques will help
How can we mitigate the problem of jitters in latency across trials affecting the averaged peak (counter to the limitation)
• Measure the area of the waveform as opposed to peak
why is it difficult to capture the underlying neural response with an average? (big limitation)
• Latency variability makes it hard to see a given neural response in an averaged waveform.
–> Should use time-frequency-analysis techniques to better see induced neural activity.
• sometimes latent components overlap with one another such that the movement of one would affect movement of the other
could be a shift in the a or b component that produced the overall waveform
What are ways to reduce confounds in EEG
- Consider physical properties of stimuli
- Use as identical stimuli as you can in different experimental conditions
- Bare in mind both the preparation and execution of a motor response will elicit ERP activity. Don’t mistake button presses for stimulus induced activity
- Whenever you can, vary experimental conditions within blocks and not between them.
Ways to reduce confounds: physical properties
- Make sure physical properties of stimulus don’t evoke electrical activity in itself
- some ERP components are sensitive to things like luminance/contrast
- imagine study on race – claiming to find ERP component that detects ethnic identity but really it’s just picking up on light or dark colours
- P1 sensitive to physical properties of stimuli e.g., colour – might find higher p1 amplitude - cant then conclude P1 computes race information
Ways to reduce confounds: using identical stimuli In diff conditions
- if you are comparing faces that are familiar and not familiar to n.
- you can use a familiar face for one for the unfamiliar face for the other!
Ways to reduce confounds: motor responses
- motor activity in one body region activates the contralateral hemisphere
- button press with the right hand would evoke activity in the left hemisphere
- n press right botton with stimulus A; and left button with stimulus B
- Don’t mistake button presses for stimulus induced activity
- To deal with this: have one group use 1 button, other group use a diff one ; easy to check because activity should switch if left button press is switched to the right while stimuli stay fixed
Ways to reduce confounds: vary conditions within blocks
- Try and have all conditions in a single block.
- Maybe n perform differently because they are more tired in blocks nearing the end of the trial.
- Alternatively, they might get more practice as the study goes on and perform better in tasks near the end.
- Changes in brain activity can reflect either of these.
Let’s say we have a neuron that is selective to a line at 45 degree orientation to the right
will this respond to all lines of 45 degrees in an image?
No each cell in V1 has a small receptive field and responds to stimuli in that receptive field