you had it coming... precursors of performance errors tom eichele, md phd department of biological...
TRANSCRIPT
You had it coming Precursors of Performance Errors
Tom Eichele MD PhDDepartment of Biological and Medical Psychology
University of Bergen
Source New York Times
Tom Eichele MD PhD University of Bergen
Pre-Error Speedingeg Gehring amp Fencsik JN 2001
ACCERNeg Debener et al JN 2005
Stimulus Timing Convolution Matrix Pseudoinverse IC timecourse Estimated HRF
=
Stimulus TimingEstimated HRF Design Matrix IC timecourse
X y β1n==
LS
Single Trial Weights
Deconvolution
Single Trial Analysis
SubjM sM(vM)
1
N
A1
A2
AM
12
K
12
K
12
K
B1
B2
BM
12
K
12
K
12
K
T1
T2
TM
12
12
K
12
K
1
L
1
L
1
L
1
N
1
N
F1-1
F2-1
FM-1
G-1 Acirc-1 Ĉ-1
xM(j)
x(j) ŝ(j)x2(j)
x1(j)
Subj 2 s2(v2)
Subj 1 s1(v1) u1(v1)
u2(v2)
uM(vM)
1
N
1
N
Ky1(j)
y2(j)
yM(j)
y1(i1)
y2(i2)
yM(iM)
Even
ts1 Data Generation 2 Acquisition 3 Reduction 4 Decomposition 5 Component Selection amp Inference
1 Replicability 2 Physiology
3 Population
4 Event-Related Response DeCon
5 Functional Modulation STA
Map-based criteria
Timecourse-based criteria
Events in a stimulus paradigm evoke neural responses in task-related sources in the presence of background activity in a number of subjects 1M Sources s at locations v are spatio-temporally mixed in A and hemodynamically convolved
Mixed signals u are recorded by the MRI scanner in BThe raw data aretransformed to T bypre-processing (motion correction normalization smoothing filtering)
Preprocessed signals yare compressed to a setnumber of factors in F with PCA to reducecomputational loadIndividual PCs are concatenated to an aggregate set G
Spatial ICA estimates the inverse of A and the aggregate components C Back-Reconstructionof individual data spatial (Gi
-1Acirc)FiYi
temporal FiGi Acirc
From the initial set of components keep those that 1 Replicate across runs andhellip 2 Represent grey matter andhellip3 Generalize to the population4 For the timecourses of remaining ICs
deconvolve the hemodynamic response5 If there is a HRF estimate single trial
amplitudes
RCZ
33 24 8
52 -30 -42
-10 28 38
24 -60 8
PC
oIFG pMFC
SMA
Cent
SMASFG
Ins
IC1
IC2
IC3
IC4
2 4 6 8 10 12 14 16 18 20
-05
0
05
Latency (sec)
Ampl
itude
(au
)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Ampl
itude
(β)
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Ampl
itude
(β)
Ampl
itude
(β)
Ampl
itude
(β)
fdr 001 t12=446 puncorr=3910-4
fdr 001 t12=483 puncorr=2110-4
fdr 001 t12=405 puncorr=8110-4
fdr 001 t12=386 puncorr=1110-3(a)
(d)
(g)
(j) (k) (l)
(h) (i)
(e) (f)
(b) (c)
left right
Different task Similar FindingsLi et al NIMG 2007
Contrast Trial preceding Error vs preceding Correct
XX
KX
XX
K
80Go 20 No Go
3 Go stimuli every 6 seconds
ISI of 1 2 or 3 secs
No Go stimuli every 10-15 secs
TR = 15 secs
Another Different Task same Precursors GoNoGo
Preprocessing amp Component Selection100 healthy right-handed participants 50m50f 18-55 years 3T scanner OLIN SPM2 realignment
spatial normalization resampling to 3x3x3 mm voxels 8mm FWHM smoothing
GIFT group ICA with 64 estimated components 50 ICASSO re-runs map-based selection of ICs
HRF deconvolution from remaining components HRF-based selection single trial estimation from event related ICs second level temporal ICA on concatenated single-trial modulations precursor estimation
IC1
IC8
IC15
IC22
IC2
IC9
IC16
IC23
IC3
IC10
IC17
IC24
IC4
IC11
IC18
IC25
IC5
IC12
IC19
IC26
IC6
IC13
IC20
IC27
IC7
IC14
IC21
IC28
Selected maps from spatial ICA
HRFs from spatial ICA time courses
Trial-by-trial sequences from spatial ICA
Trial-by-trial sequences from 2nd level temporal ICA
Error signal vs Precursor weights
Axial view
Coronal view
L R
Top 3 Error signals (red ) and Precursors (blue)
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
Pre-Error Speedingeg Gehring amp Fencsik JN 2001
ACCERNeg Debener et al JN 2005
Stimulus Timing Convolution Matrix Pseudoinverse IC timecourse Estimated HRF
=
Stimulus TimingEstimated HRF Design Matrix IC timecourse
X y β1n==
LS
Single Trial Weights
Deconvolution
Single Trial Analysis
SubjM sM(vM)
1
N
A1
A2
AM
12
K
12
K
12
K
B1
B2
BM
12
K
12
K
12
K
T1
T2
TM
12
12
K
12
K
1
L
1
L
1
L
1
N
1
N
F1-1
F2-1
FM-1
G-1 Acirc-1 Ĉ-1
xM(j)
x(j) ŝ(j)x2(j)
x1(j)
Subj 2 s2(v2)
Subj 1 s1(v1) u1(v1)
u2(v2)
uM(vM)
1
N
1
N
Ky1(j)
y2(j)
yM(j)
y1(i1)
y2(i2)
yM(iM)
Even
ts1 Data Generation 2 Acquisition 3 Reduction 4 Decomposition 5 Component Selection amp Inference
1 Replicability 2 Physiology
3 Population
4 Event-Related Response DeCon
5 Functional Modulation STA
Map-based criteria
Timecourse-based criteria
Events in a stimulus paradigm evoke neural responses in task-related sources in the presence of background activity in a number of subjects 1M Sources s at locations v are spatio-temporally mixed in A and hemodynamically convolved
Mixed signals u are recorded by the MRI scanner in BThe raw data aretransformed to T bypre-processing (motion correction normalization smoothing filtering)
Preprocessed signals yare compressed to a setnumber of factors in F with PCA to reducecomputational loadIndividual PCs are concatenated to an aggregate set G
Spatial ICA estimates the inverse of A and the aggregate components C Back-Reconstructionof individual data spatial (Gi
-1Acirc)FiYi
temporal FiGi Acirc
From the initial set of components keep those that 1 Replicate across runs andhellip 2 Represent grey matter andhellip3 Generalize to the population4 For the timecourses of remaining ICs
deconvolve the hemodynamic response5 If there is a HRF estimate single trial
amplitudes
RCZ
33 24 8
52 -30 -42
-10 28 38
24 -60 8
PC
oIFG pMFC
SMA
Cent
SMASFG
Ins
IC1
IC2
IC3
IC4
2 4 6 8 10 12 14 16 18 20
-05
0
05
Latency (sec)
Ampl
itude
(au
)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Ampl
itude
(β)
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Ampl
itude
(β)
Ampl
itude
(β)
Ampl
itude
(β)
fdr 001 t12=446 puncorr=3910-4
fdr 001 t12=483 puncorr=2110-4
fdr 001 t12=405 puncorr=8110-4
fdr 001 t12=386 puncorr=1110-3(a)
(d)
(g)
(j) (k) (l)
(h) (i)
(e) (f)
(b) (c)
left right
Different task Similar FindingsLi et al NIMG 2007
Contrast Trial preceding Error vs preceding Correct
XX
KX
XX
K
80Go 20 No Go
3 Go stimuli every 6 seconds
ISI of 1 2 or 3 secs
No Go stimuli every 10-15 secs
TR = 15 secs
Another Different Task same Precursors GoNoGo
Preprocessing amp Component Selection100 healthy right-handed participants 50m50f 18-55 years 3T scanner OLIN SPM2 realignment
spatial normalization resampling to 3x3x3 mm voxels 8mm FWHM smoothing
GIFT group ICA with 64 estimated components 50 ICASSO re-runs map-based selection of ICs
HRF deconvolution from remaining components HRF-based selection single trial estimation from event related ICs second level temporal ICA on concatenated single-trial modulations precursor estimation
IC1
IC8
IC15
IC22
IC2
IC9
IC16
IC23
IC3
IC10
IC17
IC24
IC4
IC11
IC18
IC25
IC5
IC12
IC19
IC26
IC6
IC13
IC20
IC27
IC7
IC14
IC21
IC28
Selected maps from spatial ICA
HRFs from spatial ICA time courses
Trial-by-trial sequences from spatial ICA
Trial-by-trial sequences from 2nd level temporal ICA
Error signal vs Precursor weights
Axial view
Coronal view
L R
Top 3 Error signals (red ) and Precursors (blue)
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
ACCERNeg Debener et al JN 2005
Stimulus Timing Convolution Matrix Pseudoinverse IC timecourse Estimated HRF
=
Stimulus TimingEstimated HRF Design Matrix IC timecourse
X y β1n==
LS
Single Trial Weights
Deconvolution
Single Trial Analysis
SubjM sM(vM)
1
N
A1
A2
AM
12
K
12
K
12
K
B1
B2
BM
12
K
12
K
12
K
T1
T2
TM
12
12
K
12
K
1
L
1
L
1
L
1
N
1
N
F1-1
F2-1
FM-1
G-1 Acirc-1 Ĉ-1
xM(j)
x(j) ŝ(j)x2(j)
x1(j)
Subj 2 s2(v2)
Subj 1 s1(v1) u1(v1)
u2(v2)
uM(vM)
1
N
1
N
Ky1(j)
y2(j)
yM(j)
y1(i1)
y2(i2)
yM(iM)
Even
ts1 Data Generation 2 Acquisition 3 Reduction 4 Decomposition 5 Component Selection amp Inference
1 Replicability 2 Physiology
3 Population
4 Event-Related Response DeCon
5 Functional Modulation STA
Map-based criteria
Timecourse-based criteria
Events in a stimulus paradigm evoke neural responses in task-related sources in the presence of background activity in a number of subjects 1M Sources s at locations v are spatio-temporally mixed in A and hemodynamically convolved
Mixed signals u are recorded by the MRI scanner in BThe raw data aretransformed to T bypre-processing (motion correction normalization smoothing filtering)
Preprocessed signals yare compressed to a setnumber of factors in F with PCA to reducecomputational loadIndividual PCs are concatenated to an aggregate set G
Spatial ICA estimates the inverse of A and the aggregate components C Back-Reconstructionof individual data spatial (Gi
-1Acirc)FiYi
temporal FiGi Acirc
From the initial set of components keep those that 1 Replicate across runs andhellip 2 Represent grey matter andhellip3 Generalize to the population4 For the timecourses of remaining ICs
deconvolve the hemodynamic response5 If there is a HRF estimate single trial
amplitudes
RCZ
33 24 8
52 -30 -42
-10 28 38
24 -60 8
PC
oIFG pMFC
SMA
Cent
SMASFG
Ins
IC1
IC2
IC3
IC4
2 4 6 8 10 12 14 16 18 20
-05
0
05
Latency (sec)
Ampl
itude
(au
)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Ampl
itude
(β)
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Ampl
itude
(β)
Ampl
itude
(β)
Ampl
itude
(β)
fdr 001 t12=446 puncorr=3910-4
fdr 001 t12=483 puncorr=2110-4
fdr 001 t12=405 puncorr=8110-4
fdr 001 t12=386 puncorr=1110-3(a)
(d)
(g)
(j) (k) (l)
(h) (i)
(e) (f)
(b) (c)
left right
Different task Similar FindingsLi et al NIMG 2007
Contrast Trial preceding Error vs preceding Correct
XX
KX
XX
K
80Go 20 No Go
3 Go stimuli every 6 seconds
ISI of 1 2 or 3 secs
No Go stimuli every 10-15 secs
TR = 15 secs
Another Different Task same Precursors GoNoGo
Preprocessing amp Component Selection100 healthy right-handed participants 50m50f 18-55 years 3T scanner OLIN SPM2 realignment
spatial normalization resampling to 3x3x3 mm voxels 8mm FWHM smoothing
GIFT group ICA with 64 estimated components 50 ICASSO re-runs map-based selection of ICs
HRF deconvolution from remaining components HRF-based selection single trial estimation from event related ICs second level temporal ICA on concatenated single-trial modulations precursor estimation
IC1
IC8
IC15
IC22
IC2
IC9
IC16
IC23
IC3
IC10
IC17
IC24
IC4
IC11
IC18
IC25
IC5
IC12
IC19
IC26
IC6
IC13
IC20
IC27
IC7
IC14
IC21
IC28
Selected maps from spatial ICA
HRFs from spatial ICA time courses
Trial-by-trial sequences from spatial ICA
Trial-by-trial sequences from 2nd level temporal ICA
Error signal vs Precursor weights
Axial view
Coronal view
L R
Top 3 Error signals (red ) and Precursors (blue)
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
Stimulus Timing Convolution Matrix Pseudoinverse IC timecourse Estimated HRF
=
Stimulus TimingEstimated HRF Design Matrix IC timecourse
X y β1n==
LS
Single Trial Weights
Deconvolution
Single Trial Analysis
SubjM sM(vM)
1
N
A1
A2
AM
12
K
12
K
12
K
B1
B2
BM
12
K
12
K
12
K
T1
T2
TM
12
12
K
12
K
1
L
1
L
1
L
1
N
1
N
F1-1
F2-1
FM-1
G-1 Acirc-1 Ĉ-1
xM(j)
x(j) ŝ(j)x2(j)
x1(j)
Subj 2 s2(v2)
Subj 1 s1(v1) u1(v1)
u2(v2)
uM(vM)
1
N
1
N
Ky1(j)
y2(j)
yM(j)
y1(i1)
y2(i2)
yM(iM)
Even
ts1 Data Generation 2 Acquisition 3 Reduction 4 Decomposition 5 Component Selection amp Inference
1 Replicability 2 Physiology
3 Population
4 Event-Related Response DeCon
5 Functional Modulation STA
Map-based criteria
Timecourse-based criteria
Events in a stimulus paradigm evoke neural responses in task-related sources in the presence of background activity in a number of subjects 1M Sources s at locations v are spatio-temporally mixed in A and hemodynamically convolved
Mixed signals u are recorded by the MRI scanner in BThe raw data aretransformed to T bypre-processing (motion correction normalization smoothing filtering)
Preprocessed signals yare compressed to a setnumber of factors in F with PCA to reducecomputational loadIndividual PCs are concatenated to an aggregate set G
Spatial ICA estimates the inverse of A and the aggregate components C Back-Reconstructionof individual data spatial (Gi
-1Acirc)FiYi
temporal FiGi Acirc
From the initial set of components keep those that 1 Replicate across runs andhellip 2 Represent grey matter andhellip3 Generalize to the population4 For the timecourses of remaining ICs
deconvolve the hemodynamic response5 If there is a HRF estimate single trial
amplitudes
RCZ
33 24 8
52 -30 -42
-10 28 38
24 -60 8
PC
oIFG pMFC
SMA
Cent
SMASFG
Ins
IC1
IC2
IC3
IC4
2 4 6 8 10 12 14 16 18 20
-05
0
05
Latency (sec)
Ampl
itude
(au
)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Ampl
itude
(β)
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Ampl
itude
(β)
Ampl
itude
(β)
Ampl
itude
(β)
fdr 001 t12=446 puncorr=3910-4
fdr 001 t12=483 puncorr=2110-4
fdr 001 t12=405 puncorr=8110-4
fdr 001 t12=386 puncorr=1110-3(a)
(d)
(g)
(j) (k) (l)
(h) (i)
(e) (f)
(b) (c)
left right
Different task Similar FindingsLi et al NIMG 2007
Contrast Trial preceding Error vs preceding Correct
XX
KX
XX
K
80Go 20 No Go
3 Go stimuli every 6 seconds
ISI of 1 2 or 3 secs
No Go stimuli every 10-15 secs
TR = 15 secs
Another Different Task same Precursors GoNoGo
Preprocessing amp Component Selection100 healthy right-handed participants 50m50f 18-55 years 3T scanner OLIN SPM2 realignment
spatial normalization resampling to 3x3x3 mm voxels 8mm FWHM smoothing
GIFT group ICA with 64 estimated components 50 ICASSO re-runs map-based selection of ICs
HRF deconvolution from remaining components HRF-based selection single trial estimation from event related ICs second level temporal ICA on concatenated single-trial modulations precursor estimation
IC1
IC8
IC15
IC22
IC2
IC9
IC16
IC23
IC3
IC10
IC17
IC24
IC4
IC11
IC18
IC25
IC5
IC12
IC19
IC26
IC6
IC13
IC20
IC27
IC7
IC14
IC21
IC28
Selected maps from spatial ICA
HRFs from spatial ICA time courses
Trial-by-trial sequences from spatial ICA
Trial-by-trial sequences from 2nd level temporal ICA
Error signal vs Precursor weights
Axial view
Coronal view
L R
Top 3 Error signals (red ) and Precursors (blue)
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
RCZ
33 24 8
52 -30 -42
-10 28 38
24 -60 8
PC
oIFG pMFC
SMA
Cent
SMASFG
Ins
IC1
IC2
IC3
IC4
2 4 6 8 10 12 14 16 18 20
-05
0
05
Latency (sec)
Ampl
itude
(au
)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
-05
0
05
Ampl
itude
(au
)
2 4 6 8 10 12 14 16 18 20Latency (sec)
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Ampl
itude
(β)
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Trials-6 -5 -4 -3 -2 -1 Error +1 +2
-02
0
02
Ampl
itude
(β)
Ampl
itude
(β)
Ampl
itude
(β)
fdr 001 t12=446 puncorr=3910-4
fdr 001 t12=483 puncorr=2110-4
fdr 001 t12=405 puncorr=8110-4
fdr 001 t12=386 puncorr=1110-3(a)
(d)
(g)
(j) (k) (l)
(h) (i)
(e) (f)
(b) (c)
left right
Different task Similar FindingsLi et al NIMG 2007
Contrast Trial preceding Error vs preceding Correct
XX
KX
XX
K
80Go 20 No Go
3 Go stimuli every 6 seconds
ISI of 1 2 or 3 secs
No Go stimuli every 10-15 secs
TR = 15 secs
Another Different Task same Precursors GoNoGo
Preprocessing amp Component Selection100 healthy right-handed participants 50m50f 18-55 years 3T scanner OLIN SPM2 realignment
spatial normalization resampling to 3x3x3 mm voxels 8mm FWHM smoothing
GIFT group ICA with 64 estimated components 50 ICASSO re-runs map-based selection of ICs
HRF deconvolution from remaining components HRF-based selection single trial estimation from event related ICs second level temporal ICA on concatenated single-trial modulations precursor estimation
IC1
IC8
IC15
IC22
IC2
IC9
IC16
IC23
IC3
IC10
IC17
IC24
IC4
IC11
IC18
IC25
IC5
IC12
IC19
IC26
IC6
IC13
IC20
IC27
IC7
IC14
IC21
IC28
Selected maps from spatial ICA
HRFs from spatial ICA time courses
Trial-by-trial sequences from spatial ICA
Trial-by-trial sequences from 2nd level temporal ICA
Error signal vs Precursor weights
Axial view
Coronal view
L R
Top 3 Error signals (red ) and Precursors (blue)
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
Different task Similar FindingsLi et al NIMG 2007
Contrast Trial preceding Error vs preceding Correct
XX
KX
XX
K
80Go 20 No Go
3 Go stimuli every 6 seconds
ISI of 1 2 or 3 secs
No Go stimuli every 10-15 secs
TR = 15 secs
Another Different Task same Precursors GoNoGo
Preprocessing amp Component Selection100 healthy right-handed participants 50m50f 18-55 years 3T scanner OLIN SPM2 realignment
spatial normalization resampling to 3x3x3 mm voxels 8mm FWHM smoothing
GIFT group ICA with 64 estimated components 50 ICASSO re-runs map-based selection of ICs
HRF deconvolution from remaining components HRF-based selection single trial estimation from event related ICs second level temporal ICA on concatenated single-trial modulations precursor estimation
IC1
IC8
IC15
IC22
IC2
IC9
IC16
IC23
IC3
IC10
IC17
IC24
IC4
IC11
IC18
IC25
IC5
IC12
IC19
IC26
IC6
IC13
IC20
IC27
IC7
IC14
IC21
IC28
Selected maps from spatial ICA
HRFs from spatial ICA time courses
Trial-by-trial sequences from spatial ICA
Trial-by-trial sequences from 2nd level temporal ICA
Error signal vs Precursor weights
Axial view
Coronal view
L R
Top 3 Error signals (red ) and Precursors (blue)
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
XX
KX
XX
K
80Go 20 No Go
3 Go stimuli every 6 seconds
ISI of 1 2 or 3 secs
No Go stimuli every 10-15 secs
TR = 15 secs
Another Different Task same Precursors GoNoGo
Preprocessing amp Component Selection100 healthy right-handed participants 50m50f 18-55 years 3T scanner OLIN SPM2 realignment
spatial normalization resampling to 3x3x3 mm voxels 8mm FWHM smoothing
GIFT group ICA with 64 estimated components 50 ICASSO re-runs map-based selection of ICs
HRF deconvolution from remaining components HRF-based selection single trial estimation from event related ICs second level temporal ICA on concatenated single-trial modulations precursor estimation
IC1
IC8
IC15
IC22
IC2
IC9
IC16
IC23
IC3
IC10
IC17
IC24
IC4
IC11
IC18
IC25
IC5
IC12
IC19
IC26
IC6
IC13
IC20
IC27
IC7
IC14
IC21
IC28
Selected maps from spatial ICA
HRFs from spatial ICA time courses
Trial-by-trial sequences from spatial ICA
Trial-by-trial sequences from 2nd level temporal ICA
Error signal vs Precursor weights
Axial view
Coronal view
L R
Top 3 Error signals (red ) and Precursors (blue)
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
Preprocessing amp Component Selection100 healthy right-handed participants 50m50f 18-55 years 3T scanner OLIN SPM2 realignment
spatial normalization resampling to 3x3x3 mm voxels 8mm FWHM smoothing
GIFT group ICA with 64 estimated components 50 ICASSO re-runs map-based selection of ICs
HRF deconvolution from remaining components HRF-based selection single trial estimation from event related ICs second level temporal ICA on concatenated single-trial modulations precursor estimation
IC1
IC8
IC15
IC22
IC2
IC9
IC16
IC23
IC3
IC10
IC17
IC24
IC4
IC11
IC18
IC25
IC5
IC12
IC19
IC26
IC6
IC13
IC20
IC27
IC7
IC14
IC21
IC28
Selected maps from spatial ICA
HRFs from spatial ICA time courses
Trial-by-trial sequences from spatial ICA
Trial-by-trial sequences from 2nd level temporal ICA
Error signal vs Precursor weights
Axial view
Coronal view
L R
Top 3 Error signals (red ) and Precursors (blue)
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
IC1
IC8
IC15
IC22
IC2
IC9
IC16
IC23
IC3
IC10
IC17
IC24
IC4
IC11
IC18
IC25
IC5
IC12
IC19
IC26
IC6
IC13
IC20
IC27
IC7
IC14
IC21
IC28
Selected maps from spatial ICA
HRFs from spatial ICA time courses
Trial-by-trial sequences from spatial ICA
Trial-by-trial sequences from 2nd level temporal ICA
Error signal vs Precursor weights
Axial view
Coronal view
L R
Top 3 Error signals (red ) and Precursors (blue)
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
HRFs from spatial ICA time courses
Trial-by-trial sequences from spatial ICA
Trial-by-trial sequences from 2nd level temporal ICA
Error signal vs Precursor weights
Axial view
Coronal view
L R
Top 3 Error signals (red ) and Precursors (blue)
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
Trial-by-trial sequences from spatial ICA
Trial-by-trial sequences from 2nd level temporal ICA
Error signal vs Precursor weights
Axial view
Coronal view
L R
Top 3 Error signals (red ) and Precursors (blue)
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
Trial-by-trial sequences from 2nd level temporal ICA
Error signal vs Precursor weights
Axial view
Coronal view
L R
Top 3 Error signals (red ) and Precursors (blue)
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
Error signal vs Precursor weights
Axial view
Coronal view
L R
Top 3 Error signals (red ) and Precursors (blue)
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
Axial view
Coronal view
L R
Top 3 Error signals (red ) and Precursors (blue)
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
Given that there are precursors in the fMRIhellip
are they in the EEG as well
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
50 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
40050 100 150 200 250
100
200
300
400
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
-600 -400 -200 0 200 400-4
-2
0
2
4
6
8
Response-locked EEG decomposition
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
RT-sorted single trial images
Condition averages
R ErrorG IncompatibleB Compatible
Time (ms)
Pote
ntial
(microV)
Tria
ls
Component scalp maps
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
-6 -4 -2 0 2
-02
0
02
Trials
Am
plit
ud
e (z
)
Average +- SEM
PrecursorError Signal
120-180 ms post-stimulus
Trial-to-trial EEG dynamics
120-180 ms post-response
20-80 ms post-response
tIC1-S tIC2-S tIC3-S tIC4-S tIC5-S
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
tIC1-R tIC2-R tIC3-R tIC4-R tIC5-R
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
Summary
bull Error precursors existndash In different tasks Flanker GoNoGo Simonndash In different modalities Behavior fMRI EEGndash with similar trends spanning tens of secondsndash with similar spatial patterns involving
bull Increases in default mode regionsbull Decreases in executiveeffort regions
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
What now
Now that we know what the precursors are bad for we need to figure out what they are good for
Follow up EEG and EEG-fMRI follow-up
computational modellingbehavior prediction possible
bull Paper Eichele et al PNAS 2008bull Software GIFTEEGIFT icatbsourceforgenetbull Data httpportalmindunmedudconbull tomeichelepsybpuibno
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-
Markus UllspergerCologne
Vince D CalhounAlbuquerque
Stefan DebenerJena
Karsten SpechtBergen
Thanks
- Slide 1
- Slide 2
- Slide 3
- Slide 4
- Slide 5
- Different task Similar Findings Li et al NIMG 2007
- Another Different Task same Precursors GoNoGo
- Preprocessing amp Component Selection
- Slide 9
- Slide 10
- Slide 11
- Slide 12
- Slide 13
- Slide 14
- Slide 15
- Slide 16
- Slide 17
- Summary
- What now
- Slide 20
-