Mining Eventrelated Brain Dynamics for Brainwave control

1
Mining Event-related Brain Dynamics for Brainwave
control
University of Edinburgh, January 11, 2005

• Arnaud Delorme
• Swartz Center for Computational Neuroscience
• Institute for Neural Computation
• University of California San Diego
• La Jolla CA

2
YES
Recording
Electrodes
Time (ms)
3
I realized that
It struck me that
Brain Dynamic Events
I wondered if
All of a sudden ...
The feeling hit me like
I looked to see if
I noticed that
I looked again at .
I decided that
It occurred to me that
I imagined
I searched my memory for
4
YES
SVM

• Brain dynamics
• brain source activity
• coherence

Electrodes
Time (ms)
5
Single Scalp Electrode
Single Neuron
6
Relative Independence
7
Mixture of Brain source activity
8
1
A
-2
1.57
B
-0.33
Linear Combination
YAB
ICA
XYW
Infomax ICA


YW-1X
A
B
9
Example Speech Separation
10
Historical Remarks

• Herault Jutten ("Space or time adaptive signal
  processing by neural network models, Neural Nets
  for Computing Meeting, Snowbird, Utah, 1986)
  Seminal paper, neural network
• Bell Sejnowski (1995) Information Maximization
• Amari et al. (1996) Natural Gradient Learning
• Cardoso (1996) JADE

• Applications of ICA to biomedical signals
• EEG/ERP analysis (Makeig, Bell, Jung Sejnowski,
  1996).
• fMRI analysis (McKeown et al. 1998)

11
ICA Training Process

• Remove the mean
• x x - ltxgt

• Sphere the data by diagonalizing its covariance
  matrix,
• x ltxxTgt-1/2(x-ltxgt).

12
(No Transcript)
13
Largest 30 Independent Components (single subject)
Onton, Delorme and Makeig, in prep.
14
Overlapping Maps and Spectra
15
Two Neck Muscle Processes
16
Some Independent EMG Components
17
Two Central Alpha Processes
18
Software
EEGLAB Import data and ICA decomposition
DIPFIT Dipole localization for ICA comp.
19
ICA applied to EEG data
ICA component scalp maps
Electrodes
ICA

0
-
Components
Localization
Time (ms)
20
ICA Reliability

• Data
• 14 subjects performing a memory task
• 71 electrodes
• more than 300,000 data points
• Decomposition
• 23 ICA algorithms plus PCA and Promax
• Analysis
• Localization of all components with a single
  dipole
• Plotting the cumulative number of component
  against
• the residual variance

21
Decomposition Quality Across Algorithms
Metric Dipolarity
Number of Components (of 72)
Number of Components (of 71)
Residual Variance (comp. map vs. dipole) ()
Residual Variance ()
22
(No Transcript)
23
(No Transcript)
24
(No Transcript)
25
(No Transcript)
26
(No Transcript)
27
assessing brain dynamics in single trials
28
(No Transcript)
29
(No Transcript)
30
Clustering
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
(No Transcript)
35
(No Transcript)
36
(No Transcript)
37
(No Transcript)
38
Phase ITC
Normalized (no amplitude information)
39
TWO SIMULATED THETA PROCESSES
40
Phase coherence
41
Subjects
Clustering
EEG
ICA
Spectral analysis ITC Coherence
Brain dynamic movie
Localize
42
5 Hz
Lµ
CM
FM
Rµ
La
P3b
Ca
P3f
Ra
43
(No Transcript)
44
(No Transcript)
45
feedback
Recording
Electrodes
Thought events
Time (ms)
46
A Forced-Choice Speeded Response Task
response

!
CUE
FEEDBACK
TARGET
-800 ms
1005 ms
deadline
HHSHH
Targets
RH !
SSSSS
HHHHH
LH !
SSHSS
Luu
,
Fraiche
Tucker,
2000.
47
Erickson Flanker Task
Stimulus-Locked Theta Dynamics (4.9 Hz)
Movie Delorme Makeig, 2002
LOSE!
WIN!
LOSE!
Data Luu, Fraiche Tucker, 2000
48
Erickson Flanker Task
Stimulus-Locked Alpha Dynamics (10.7 Hz)
Movie Delorme Makeig, 2002
49
Suport Vector machine usingSpiking neural network
ON center
OFF center
Level 1
Level 2
50
Suport Vector machine usingSpiking neural network
ON center
OFF center
51
Emergence to object selectivity

• Supervised learning
• Fixed by the expérimentalist

• 8 (x40) views for the learning phase
• 2 (x40) views for the testing phase

52
Emergence to object selectivity
Learning base 100
Testing base 97,5
53
Emergence to object selectivity
54
(No Transcript)
55
Application to neurofeedback

• Clinical (epilepsy, ADHD, )
• Though reading
• For control of external objects
• For therapy

56
(No Transcript)