You are on page 1of 16

Characterization of evoked and induced

activity in EEG and assessment of intertrial
variability
Kostas Michalopoulos, Vasiliki Iordanidou, Giorgos A. Giannakakis,
Konstantina S Nikita and Michalis Zervakis
Outline
• ERP activity.
• ERP activity measures.
• Intertrial Entropy.
• Demonstration in Real Data.

Event related potentials
• Events can induce time-locked changes in the activity
of neuronal population.
• Basic assumptions:
• Signal of interest has fixed time-delay to the
stimulus.
• Ongoing EEG acts as additive noise.
• Averaging across trials enhances this kind of activity
and suppress noise.


Induced activity
• Represent frequency specific changes of the ongoing
EEG.
• Activity time-locked to event, but not phase locked.
• Suppressed by averaging.
Event Related potentials
• Evoked responses
• Brain response to stimulus analyzed
through averaging across trials.
• Averaging enhances time and phase
locked activity to stimulus event.
• Major features are amplitude and latency
of peaks and valleys of the ERP waveform.


• Induced responses
• Oscillatory activity time-locked to stimulus
but not phase locked interact with the ERP
response.
• Such activity is eliminated in the average
ERP.
• Manifested as decrease or increase of
band power to stimulus.
• Widely known as ERS/ ERD activity.
Image from G. Pfurtscheller,F.H.Lopes da Silva, 1999
Time frequency representation
• Time frequency provides a mean to localize, in time
and frequency, ERP activations.
• For multiple trials, we need a metric to summarize
similar activity in time-after-event.
Evoked activity measure
• Phase inter-trial coherence


• Equality holds if and only if the trials involve the
same basic signal with the same phase.
• This metric is expanded to the time-frequency
representation of a signal , with k and n
indicating the frequency and time ticks, respectively.

1
] [
] [
] [ s =
¿
¿
i
i
i
i
PIC
k X
k X
k c
] , [ n k X
i
Induced activity measure
• Phase-shift inter-trial coherence




• Equality holds if and only if |Xi[k,n]| is constant for all trials i.
• Eliminates the complex phase effects and compares the
intertrial content only based on its energy in specific
frequency bands.
• Quantifies and highlights frequency bands of increased energy
in all trials.
• Does not discriminate phase and not phase-locked activity.

2
2
1
[ , ]
[ , ] 1,
1
[ , ]
i
i
i
i
X k n
T
c k n
X k n
T
(
(
¸ ¸
= s
¿
¿
Event Related Synchronization /
Desynchronization(ERS/ ERD)
• ERS/ ERD represent a mean increase or decrease in
event-related power.
• Defined as the percentage of increase or decrease in
mean trial power from a mean power baseline.
• First measure that quantified induced activity.


Evoked - Induced and Global Energy
• Previous measures use the mean power across trials.
• Take under consideration the mixed activity of
evoked and induced activations.
• Difficult to discriminate between power changes due
to induced and evoked activity.
• A method to extract the induced activity out of the
global energy:
• Subtract the average ERP out of each trial.
• Removes only phase locked activity.

Intertrial Power Entropy
• The Shannon entropy gives a useful criterion for
analyzing and comparing probability distributions.
• Inter - trial power entropy:



• Reflects the order/ disorder of the power levels for
each frequency among trials.
Relative intertrial entropy
• Different power levels for each trial


• q[k,n] is the baseline power distribution
• positive - zero only when p[k,n] ≡ q[k,n]
• intertrial entropy change indicates increase or
decrease of entropy

Application in real EEG data.
• Dataset was provided by the Ecological University of
Bucharest, Romania
• Signals were digitally sampled at 1024Hz
• A stimulator provided 40 2kHz target tones (20%) and 160
1kHz non-target tones(80%)
• Inter-stimulus interval was 1.29s
• Auditory oddball experiment
• Expected activations:
• phase-locked oscillations, especially in the theta and delta
bands related to P300 activity
• and non phase-locked (induced) oscillatory activity, mainly
in alpha band.
Application in real EEG data.
• Example results from a healthy subject

Phase Intertrial
Coherence
Phase-shift
Intertrial
Coherence
ERD/ ERS Intertrial Entropy
Change
Relative intertrial
Entropy
Global Channel
Power
Induced
Channel Power
Global
Power
Alpha ERD Delta Theta ERS
80% Entropy
Decrease
40% Entropy Increase
Mean Relative
Entropy0.47
Induced
Power
Alpha ERD Delta Theta ERS
80% Entropy
Decrease
19% Entropy Increase
Mean relative Entropy
0.16
Discussion
• The measures for identifying and separating phase
from non-phase locked activity facilitate the rejection
of task irrelevant brain activations.
• Separating induced from global activity makes
interpretation of the measures easier.
• Method presented needs optimization. Subtraction of the
mean leaves evoked activity residuals.
• Intertrial entropy can be used to investigate brain
activations based on the power distribution across
trials.

Questions?

Demonstration in Real Data. ERP activity measures. Intertrial Entropy. .Outline • • • • ERP activity.

• Basic assumptions: • Signal of interest has fixed time-delay to the stimulus. • Averaging across trials enhances this kind of activity and suppress noise.Event related potentials • Events can induce time-locked changes in the activity of neuronal population. • Ongoing EEG acts as additive noise. .

• Activity time-locked to event. • Suppressed by averaging. .Induced activity • Represent frequency specific changes of the ongoing EEG. but not phase locked.

1999 . Major features are amplitude and latency of peaks and valleys of the ERP waveform. Image from G. Manifested as decrease or increase of band power to stimulus. Such activity is eliminated in the average ERP. Widely known as ERS/ ERD activity.H.Event Related potentials • Evoked responses • • • Brain response to stimulus analyzed through averaging across trials. • Induced responses • • • • Oscillatory activity time-locked to stimulus but not phase locked interact with the ERP response. Pfurtscheller. Averaging enhances time and phase locked activity to stimulus event.F.Lopes da Silva.

in time and frequency. ERP activations. we need a metric to summarize similar activity in time-after-event. .Time frequency representation • Time frequency provides a mean to localize. • For multiple trials.

with k and n indicating the frequency and time ticks. • This metric is expanded to the time-frequency X representation of a signal [k . i  X [k ] [k ]  1  X [k ] i i i . respectively.Evoked activity measure • Phase inter-trial coherence c PIC i • Equality holds if and only if the trials involve the same basic signal with the same phase. n] .

n]  T i 2 • Equality holds if and only if |Xi[k. . • Eliminates the complex phase effects and compares the intertrial content only based on its energy in specific frequency bands.n]| is constant for all trials i. • Quantifies and highlights frequency bands of increased energy in all trials. n]    1. c[k . n]   i 1 2 X i [ k . • Does not discriminate phase and not phase-locked activity.Induced activity measure • Phase-shift inter-trial coherence 1   T  X i [ k .

. • Defined as the percentage of increase or decrease in mean trial power from a mean power baseline.Event Related Synchronization / Desynchronization(ERS/ ERD) • ERS/ ERD represent a mean increase or decrease in event-related power. • First measure that quantified induced activity.

• Removes only phase locked activity. • Take under consideration the mixed activity of evoked and induced activations. . • A method to extract the induced activity out of the global energy: • Subtract the average ERP out of each trial.Evoked .Induced and Global Energy • Previous measures use the mean power across trials. • Difficult to discriminate between power changes due to induced and evoked activity.

Intertrial Power Entropy • The Shannon entropy gives a useful criterion for analyzing and comparing probability distributions. . • Inter .trial power entropy: • • Reflects the order/ disorder of the power levels for each frequency among trials.

zero only when p[k.n] • intertrial entropy change indicates increase or decrease of entropy .Relative intertrial entropy • Different power levels for each trial • q[k.n] ≡ q[k.n] is the baseline power distribution • positive .

• Dataset was provided by the Ecological University of Bucharest. . mainly in alpha band.Application in real EEG data. especially in the theta and delta bands related to P300 activity • and non phase-locked (induced) oscillatory activity. Romania • Signals were digitally sampled at 1024Hz • A stimulator provided 40 2kHz target tones (20%) and 160 1kHz non-target tones(80%) • Inter-stimulus interval was 1.29s • Auditory oddball experiment • Expected activations: • phase-locked oscillations.

16 .47 Induced Power Alpha ERD 80% Entropy Decrease Delta Theta ERS 19% Entropy Increase Mean relative Entropy 0.Application in real EEG data. • Example results from a healthy subject Phase Intertrial Coherence Global Channel Power Phase-shift Intertrial Coherence ERD/ ERS Intertrial Entropy Change Relative intertrial Entropy Induced Channel Power Global Power Alpha ERD 80% Entropy Decrease Delta Theta ERS 40% Entropy Increase Mean Relative Entropy0.

Discussion • The measures for identifying and separating phase from non-phase locked activity facilitate the rejection of task irrelevant brain activations. • Method presented needs optimization. Subtraction of the mean leaves evoked activity residuals. • Intertrial entropy can be used to investigate brain activations based on the power distribution across trials. . • Separating induced from global activity makes interpretation of the measures easier.

Questions? .

 ½ ¾ ° ° ¾½¯f°  fn°€  ¯ f° f ¾  fn ¾ f¾ W ° f °½nf° ¾ ° ¾–f  f° fnf°¾ f¾ ° ½  ¾ °fn¾¾ f¾ .¾n¾¾° W @ ¯ f¾ ¾€ °€°–f° ¾ ½ff°–½f¾  €¯°° ½f¾ n fn€fnf   © n° €f¾  f° f°fnf°¾ W  ½ff°–° n €¯– ffn¯f ¾ ° ½ f°€ ¯ f¾ ¾ f¾  W .

. ¾°¾" .