L9: MEG Event Related Analyses Flashcards
Event-related analyses in MEG happens when
we got data preprocessed and taking as much artefacts as we can
The most popular statistical analysis for EEG/MEG is
event-related analyses
The effects of averaging time course over multiple repetitions of condition, to increase SNR, depends on - (2)
kind of response - only work if responses are consistent and happen same time in different trials
can improve SNR but lose signal if responses are not consistent
What does this diagram show - evoked responses? - (3)
Perfect sample as responses on 3 different trials is identical and has same time course - (looking 100ms after stimulus onset) = time-locked
The averaging of 3 different trial’s response gets same strong peak (all peaks line up - phase-locked)
(evoked response)
What is evoked response?
Time locked and phase lock
What does time-locked mean?
Same amount of time after stimulus onset
What does phase lock means
At that time its the same phase
Diagram of cycle, amplitude phase and explain what phase is (2)
phase is where we are in oscillation
we have up and down measure of activity and phase is whether we are at baseline, peak or second hit to baseline or at bottom
Evoked responses are good to average across trials and thus use
event-related analyses
Induced responses is when responses are not
phase-locked
Diagram of evoked vs induced responses
Diagram of evoked vs induced
For induced - (3)
Responses are at slightly different times and because slightly different phases
Time locked as 100 milliseconds after stimulus onset, all responses are happening so all around 100 ms
Not phase locked as peaks are not same and when we average we get reduced signal and unlikely to miss effects so no longer improving SNR
Event-related analyses depend on averaging so
can used on evoked responses but not induced responses (can’t average)
We don’t average when - (2)
responses that are not time-locked (e.g., changes in attention or steady-state or resting state-analyses) as well as responses that are not phase-locked
bad to average across time course of trials
For responses that are not time-locked and phase-locked we instead calculate
power in a given frequency per trial and average those (lose phase info as it is varying)
When we have induced responses, what analysis we do?
frequency-based analyses
An event-related potential (ERP) is created In EEG when
averaging over many trials across their time courses gets a complex waveform
An ERP asks in EEG
when (and broadly where -topography spatially) is there a change in strength of the electric potential
What does event-related analyses give us in EEG
ERP
Diagram of ERP in EEG and what graph shows? - (4)
N = negative effects
P = positive effects
Label by number they happen in data (e.g., N1, N2, P1,P2) - label positive and negative deflections sequentially
Negative voltages are plotted upwards
In ERP in EEG, you are measuring the - (2)
amplitude (strength) of responses and response latency (how long after stimulus presentation [0ms] it occurs
ERPs can be compared across conditions e.g., clinical groups etc..
Diagram of latency and amplitude what it shows EEG ERP - (2)
- Amplitude = how high peak is from baseline
- latency - was response after 100 or 200 ms
ERP components can be given standard labels
by approximate time they occur in milliseconds
Why is it useful to give ERP components standard labels by the approximate time they occur in milliseconds?
Able to discuss EEG findings across studies
ERP components can be given standard labels by the (approximate) time they occur in milliseconds - (6)
P100 – basic visual response (occipital) - positive effect after 100 ms
N100 – basic auditory response (more anterior) - negative effect after 100 ms
N170 – to face stimuli - negative effect after 170 ms
P300 – decision making
Lateralized readiness potential (LRP) ( an indicator of motor planning)
N400 – semantic processing (e.g., differences in response at N400 when given sentence ‘John ate democry at dinner’ vs ‘John ate broccili’ at dinner’)
Event-related analyses is most widely used approach in
MEG
In event-related analyses in MEG it pools over
lots of trials and average out the noise to get clearer signals
In event-related analyses in MEG, it assesses
the time course of changes in magnetic field after stimulus presentation
In MEG event-related analyses we record a
magnetic evoked potential (MEP) - MEG version of ERP
The MEP asks
when (and broadly where) is there a change in the magnetic field strength
Note that MEP can also stand for
Motor Evoked Potential - a paradigm where electrical or magnetic stimulation causes a motor response – make sure you don’t get confused!
Diagram of butterfly plot in MEG shows
Different sensors and magnetic field strength shown across time and black is average
In components of MEP, the sign of deflections is less stereotyped than ERPs in EEG as…
we usually prefix them with M for magnetic rather than P for positive and N for negative
In components of MEP, the sign of deflections is less stereotyped than ERPs in EEG
for example.. - (2)
For example, the M100 is an early visual response, the M170 is associated with stimuli such as faces
The M170 can appear as a positive or negative change in the magnetic strength
Diagram example of
MEP components less stereotyped - (3)
Around 170ms we get negative effect for faces and scrambled faces on left
In another graph to right around 170ms we get positive effect for faces and scrambled faces
For incidental reasons (e.g., cortical folding)
Typical in MEG to use senor topographies of magnetic field strength at a single time point (e.g., 170 ms) or averaged across time window to show
spatial distribution of MEPs
Diagram of sensor topography in MEG
The sensor topography in MEG can differ between participants because of
individual differences in cortical folding and overall brain anatomy
The sensor topography in MEG can look different from EEG topographies for identical experimental condition because
magnetic fields are at right angles to electrical potential differences
Diagram of how we make comparisons between conditions in MEG using butterfly plots - explain
take butterfly plot of faces subtracted from scrambled faces to give our difference
In MEG we can compare its response in two conditions - (2) butterfly plots
We can subtract the time courses for each condition and look at the difference in activity
Can test for significant differences at the group level e.g., with t-test per time point (details in lecture 11) - see if its different from 0 and difference greaer than 0 and outside CI -sig
Asks – when do the responses differ between two conditions?
In MEG we can compare its response in two conditions - (3) topographies of head spatially -where is the difference?
Also possible to subtract topographies across the whole head
In general, we will do group-level statistics such as t-tests on data like this (more details in lecture 8)
Asks – (broadly) where do the responses differ between two conditions? (until we go into source space – lecture 11)
In MEG calculate global power by…
summarising the MEG activity across all sensors -quick check of seeing if something is different in responses in conditions
Global field power can be thought of
average amount of activity detected by scanner -summing everything across
Global field power sometimes can show… but can not… - (2)
coarse differences between conditions and when they occur
but can not tell us where
What does power mean in global field power?
amplitude squares so values are always positive
Diagram of global field power shown in the bottom of all the time coruse we run
