ICA decomposition and component analysis · Independent Component Analysis ... ICA decomposition...
Transcript of ICA decomposition and component analysis · Independent Component Analysis ... ICA decomposition...
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 1
ICA decomposition and component analysis
Task 1Run ICA
Exercise...Task 2
Plot componentsIdentify components
Task 3Plot component powerPlot component ERP & erpimagesPlot ERSP/Cross coherence
Exercise...
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 2
ICA decomposition and component analysis
Task 1Run ICA
Exercise...Task 2
Plot componentsIdentify components
Task 3Plot component powerPlot component ERP & erpimagesPlot ERSP/Cross coherence
Exercise...
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 3
Reject continuous data
To prepare data for ICA, reject ‘strange‘ artifacts but keep stereotyped artifacts!
Stereotyped
eye blinks
Strange
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 4
Independent Component Analysis
x = scalp EEG W = unmixing matrix
ICA
W-1 (scalp projections)
W*x = u
u = sources
*x = W-1*u
Cha
nnel
s
Time
u = sources
Com
pone
nts
Time
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 5
‘extended’,1
Option Default Comments
‘extended’ 0 1 is recommended to find sub-gaussians
‘stop’ 1e-7 final weight change stop
‘lrate’ determined too small too long…from data too large wts blow up
‘maxsteps’ 512 Should not need more?
‘pca’ 0 or Decompose only a EEG.nbchan principal data subspace
‘stop’,1e-7‘lrate’,1e-3‘maxsteps’,750‘pca’,50
Runica/binica options
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 6
Runica Progress…
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 7
Exercise
• Load dataset 'faces_3.set' from '…/data/' directory
• Reject noise from continuous or epoched data
• Run ICA
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 8
ICA decomposition and component analysis
Task 1Run ICA
Exercise...Task 2
Plot componentsIdentify components
Task 3Plot component powerPlot component ERP & erpimagesPlot ERSP/Cross coherence
Exercise...
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 9
Plot ICA scalp maps
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 10
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 11
Plot ICA scalp maps
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 12
Compare 'good' and 'bad' scalp maps
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 13
Scroll component activities
Note: Activity like this, not separated by ICA, should be removed
and ICA run again for better decomposition
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 14
Com
pone
nts
Cha
nnel
s
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 15
Plot ICA component properties
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 16
Reviewing component properties
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 17
Component scalp maps/properties
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 18
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 19
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 20
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 21
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 22
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 23
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 24
Export ICA weights
How can I apply these weights to other datasets?
Only apply weights to same subject, same session data!!
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 25
Importing ICA weights
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 26
Imported weights
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 27
ICA decomposition and component analysis
Task 1Run ICA
Exercise...Task 2
Plot componentsIdentify components
Task 3Plot component powerPlot component ERP & erpimagesPlot ERSP/Cross coherence
Exercise...
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 28
Plot component power
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 29
Component ERPs
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 30
Data (all channels)
Data Envelope (max min)
Definition: The data envelope
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 31
Component contributions to the dataset ERP
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 32
Component contribution to the dataset ERP
ArtifactComponents!
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 33
ERP Image basics
Trial 1
Trial 2
Trial 3
Trial 4
ERP Image
by default, sorted bytime-on-task
(1st trial, 2nd trial, ...)
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 34
ERP Image basicsTrial 1:Trial 2:
.
.
.
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 35
ERP Images: smoothing across trials
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 36
Component ERP Images
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 37
Plot IC ERSP
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 38
Plot IC cross coherence
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 39
Channel/IC statistics
EEGLAB Workshop III, Nov 15-18, 2006, Singapore: Julie Onton – Data Decomposition with ICA 40
Exercise
• Look at your component maps/activations
• Find components that contribute most to:
• Data power spectrum at 6 Hz
• The ERP between 100 and 500 ms-Remove noise components when plotting
• Plot and study different ERP images for these
components
• Plot ERSP and/or IC cross coherence for selected ICs