Transcript Lecture3x - U of L Class Index

Electroencephalography
• The field generated by a patch of cortex can be
modeled as a single equivalent dipolar current source
with some orientation (assumed to be perpendicular
to cortical surface)
Electroencephalography
• Electrical potential is
usually measured at
many sites on the head
surface
• More is sometimes
better
Magnetoencephalography
• For any electric current, there
is an associated magnetic field
Electric
Current
Magnetic
Field
Magnetoencephalography
• For any electric current, there
is an associated magnetic field
Electric
Current
• magnetic sensors called
“SQuID”s can measure very
small fields associated with
current flowing through
extracellular space
Magnetic
Field
SQuID
Amplifier
Magnetoencephalography
• MEG systems use many sensors
to accomplish source analysis
• MEG and EEG are
complementary because they
are sensitive to orthogonal
current flows
• MEG is very expensive
EEG/MEG
• EEG changes with various
states and in response to
stimuli
Two ways to approach EEG data
• The Event-Related
Potential
– Phase-locked or
“evoked”
– High inter-trial phase
consistency
– Retains polarity
information at scalp
– Rejects time-locked
but not phase-locked
changes
• Time/Spectral
Analysis
– Includes Non-phaselocked or “induced”
plus “evoked” signal
– Ignores inter-trial
phase consistency
(measured
differently)
– Rejects polarity at
scalp
Time-Frequency Analysis of EEG/MEG
• Any complex waveform can be decomposed into
component frequencies
– E.g.
• White light decomposes into the visible spectrum
• Musical chords decompose into individual notes
Time-Frequency Analysis of
EEG/MEG
• EEG is characterized by various
patterns of oscillations
• These oscillations
superpose in the raw
data
4 Hz
8 Hz
15 Hz
21 Hz
4 Hz + 8 Hz + 15 Hz + 21 Hz =
Time-Frequency Analysis of EEG/MEG
• The amount of energy at any frequency is expressed as
% power change relative to pre-stimulus baseline
• Power can change over time
Frequency
48 Hz
% change
From
Pre-stimulus
24 Hz
16 Hz
8 Hz
4 Hz
0
(onset)
+200
+400
Time
+600
Time-Frequency Analysis of EEG/MEG
• We can select and collapse any
time/frequency window and plot relative
power across all sensors
Win
Lose
The Event-Related Potential
(ERP)
• Embedded in the EEG signal is the small electrical response due to
specific events such as stimulus or task onsets, motor actions, etc.
The Event-Related Potential
(ERP)
•
Embedded in the EEG signal is the small electrical response due to specific
events such as stimulus or task onsets, motor actions, etc.
•
Averaging all such events together isolates this event-related potential
The Event-Related Potential
(ERP)
• We have an ERP waveform for every electrode
The Event-Related Potential
(ERP)
• We have an ERP waveform for every electrode
The Event-Related Potential
(ERP)
• We have an ERP waveform for every electrode
• Sometimes that isn’t very useful
The Event-Related Potential
(ERP)
• We have an ERP waveform for
every electrode
• Sometimes that isn’t very
useful
• Sometimes we want to know
the overall pattern of potentials
across the head surface
– isopotential map
The Event-Related Potential
(ERP)
• We have an ERP waveform for
every electrode
• Sometimes that isn’t very
useful
• Sometimes we want to know
the overall pattern of potentials
across the head surface
– isopotential map
Sometimes that isn’t very useful - we want to know the
generator source in 3D
Brain Electrical Source Analysis
• Given this pattern on the scalp,
can you guess where the
current generator was?
• Source Imaging in EEG/MEG
attempts to model the
intracranial space and “back
out” the configuration of
electrical generators that gave
rise to a particular pattern of
EEG on the scalp
Brain Electrical Source Analysis
• EEG data can be coregistered with highresolution MRI image
Source
Imaging
Result
Structural
MRI with EEG
electrodes
coregistered
CCBN Dense-Array EEG
Event Triggers
Data Files
Stimuli
Raw EEG
.raw
MatLab
Fieldtrip
BrainVoyager
SPSS
-EEG spectral analysis
- MRI coregistration
Netstation –
records EEG and
event triggers
.sfp
Digamize –records
electrode locations
BESA
-post-processing
-ERP averaging
-voltage maps
-source imaging
MANUSCRIPT
Basic Elements of ERP Design
• EEG, therefore ERP, doesn’t provide interpretable absolute
voltage
• The voltage is always relative to something else
• That something else may be:
– The pre-stimulus baseline
– A control condition
Basic Elements of ERP Design
• Thus a fundamental aspect of ERP design is not to plan to
report voltages but rather a difference in voltage between
two or more conditions
• What are some examples of conditions you might want to
compare?
First Demo
• Contralaterality in Visual
System
– Hemifields project to
contralateral cortex
– Unrelated to which eye is
stimulated!
• Occular Albinism
– Eyes project
contralaterally,
irrespective of hemifield
Basic Elements of ERP Design
• The theory is that human visual cortex is organized
contralaterally
• The prediction is that right hemifield stimuli will drive
electrical activity in the left visual cortex and left hemifield
stimuli will drive electrical activity in right visual cortex
• How do we test that prediction?
Basic Elements of ERP Design
• Experimental approach:
• Choices:
– 1. you could compare ipsi to contra ERP waveforms with a trial
• E.g. O3 with O4
• What’s the problem?
O4
O3
Basic Elements of ERP Design
• Experimental approach:
• Choices:
– 1. you could compare ipsi to contra ERP waveforms with a trial
•
•
•
•
E.g. O3 with O4
What’s the problem?
You would be comparing ERPs from different parts of the brain!
How could you improve on that design?
Basic Elements of ERP Design
• Experimental approach:
• Choices:
– 2. you could compare electrodes ipsi to stimulus on one side with
electrodes contra to stimulus on the other side
• Notice those are the same electrode!
Measure contralateral
ERP magnitude
O3
Basic Elements of ERP Design
• Experimental approach:
• Choices:
– 2. you could compare electrodes ipsi to stimulus on one side with
electrodes contra to stimulus on the other side
• Notice those are the same electrode!
Measure ipsilateral ERP
magnitude
O3
• Hands on agenda today:
– Orientation to the EEG lab
– Build your dipole models
Principals of Digital Signal
Recording
How do we represent a continuously
variable signal digitally?
• Sampling
– Sampling rate – number of measurements per unit
time
– Sampling depth or quantization – number of
gradations by which the measurement can be
recorded
How do we represent a continuously
variable signal digitally?
• Sampling
– What would be the advantage to higher sampling
rates?
How do we represent a continuously
variable signal digitally?
• Sampling
– What would be the advantage to higher sampling
rates?
• Nyquist limit
How do we represent a continuously
variable signal digitally?
• Sampling
– What would be the advantage to higher sampling
rates?
• Nyquist limit
• Aliasing
– What would be the disadvantage?
• Data size
• Compute time
How do we represent a continuously
variable signal digitally?
• Sampling
– What would be the advantage to greater sampling
depth?
• Finer resolution
– What would be the disadvantage?
• Data size
• Possibly compute time
How do we represent a continuously
variable signal digitally?
• Sampling
– A note about data size and compute time:
• New data size = increase in quantization x number of samples x number of electrodes!
Filters used in EEG
What is a filter?
What is a filter?
• Filters let some “stuff” through and keep
other “stuff” from getting through
– What do we want to let through?
– What do we want to filter out?
What is a filter?
• The goal of filtering is to improve the signal to
noise ratio
– Can the filter add signal?
Different Kinds of Filters
•
•
•
•
Low-Pass (High-Cut-Off)
High-Pass (Low-Cut-Off)
Band-Pass
Notch
• Each of these will have a certain “slope”
How do Filters Work?
• Notionally:
– Transform to frequency domain
– Mask some parts of the spectrum
– Transform back to time domain
Are There Any Drawbacks?
• Yes
• Filters necessarily distort data
– Amplitude distortion
– Latency distortion
• Forward/backward/zero-phase
Recommendations
• Should you filter?
– Yes, when necessary to reveal a real signal
• Problem: how do you know it’s “real”
– No, always look at the unfiltered data first
• What filters should you use?
– Depends on your situation (e.g. what EEG band
are you interested in? Do you have 60Hz line
noise?)
– General rule: less aggressive filters are less
distorting