Transcript BBMIapecs-Feb05 - Computer and Information Science

Neuroinformatics Research at UO
Experimental Methodology and Tool Integration
16x256
bits per
millisec
(30MB/m)
CT / MRI
segmented
tissues
EEG
NetStation
processed
EEG
Interpolator 3D
NeuroInformatics Center
BrainVoyager
mesh generation
source localization
constrained to
cortical surface
BESA
BBMI: Brain, Biology, Machine Initiative
EMSE
Feb 2005
NeuroInformatics Center (NIC) at UO

Application of computational science methods to
human neuroscience problems




Integration of neuroimaging methods and technology






Tools to help understand dynamic brain function
Tools to help diagnosis brain-related disorders
HPC simulation, large-scale data analysis, visualization
Need for coupled modeling (EEG/ERP, MR analysis)
Apply advanced statistical analysis (PCA, ICA)
Develop computational brain models (FDM, FEM)
Build source localization models (dipole, linear inverse)
Optimize temporal and spatial resolution
Internet-based capabilities for brain analysis services,
data archiving, and data mining
NeuroInformatics Center
BBMI: Brain, Biology, Machine Initiative
Feb 2005
Funding Support

BBMI federal appropriation



$40 million research attracted by BBMI
$10 million gift from Robert and Beverly Lewis family


Established Lewis Center for Neuroimaging (LCNI)
NSF Major Research Instrumentation


DoD Telemedicine Advanced Technology Research
Center (TATRC)
“Acquisition of the Oregon ICONIC Grid for Integrated
COgnitive Neuroscience Informatics and Computation”
New proposal


NIH Human Brain Project Neuroinformatics
“GENI: Grid-Enabled Neuroimaging Integration”
NeuroInformatics Center
BBMI: Brain, Biology, Machine Initiative
Feb 2005
Electrical Geodesics Inc. (EGI)


EGI Geodesics Sensor Net
Dense-array sensor technology


256-channel geodesics sensor net




AgCl plastic electrodes
Carbon fiber leads
Net Station


64/128/256 channels
Advanced EEG/ERP data analysis
Stereotactic EEG sensor registration
Research and medical services

Epilepsy diagnosis, pre-surgical planning
NeuroInformatics Center
BBMI: Brain, Biology, Machine Initiative
Feb 2005
NeuroInformatics for Brainwave Research

Electroencephalogram (EEG)

EEG time series analysis




Event-related potentials (ERP)
 Averaging to increase SNR
 Linking brain activity to sensory–motor, cognitive
functions (e.g., visual processing, response programming)
Signal cleaning (removal of noncephalic signal, “noise”)
Signal decomposition (PCA, ICA, etc.)
Neural Source localization
NeuroInformatics Center
BBMI: Brain, Biology, Machine Initiative
Feb 2005
EEG Dense-Array Methodology
NeuroInformatics Center
BBMI: Brain, Biology, Machine Initiative
Feb 2005
APECS: A new tool for EEG data decomposition


Automated Protocol for Electromagnetic Component Separation
Motivation




EEG data cleaning (increases SNR)
Separation of EEG components (addresses superposition)
Data preprocessing prior to source localization
Distinctive Features




Implements variety of algorithms (PCA, ICA, SOBI, etc.)
Uses multiple metrics for fast, automatic classification of
extracted components
Applies multiple criteria to evaluate success of decomposition (to
ensure that artifacts are cleanly separated from cortical activity)
Calls high-performance, parallel C++ implementations of
Infomax and FastICA algorithms
NeuroInformatics Center
BBMI: Brain, Biology, Machine Initiative
Feb 2005
APECS Evaluation: Qualitative Criteria
NeuroInformatics Center
BBMI: Brain, Biology, Machine Initiative
Feb 2005
APECS Evaluation: Quantitative Criteria
Covariance between “baseline”
(blink-free) and ICA-filtered data.
Yellow, Infomax; blue, FastICA.
Infomax gives consistently better
results. FastICA results are more
variable.
ICA decompositions most successful
when only one spatial projector is
strongly correlated with blink
“template” (spatial filter).
NeuroInformatics Center
BBMI: Brain, Biology, Machine Initiative
Feb 2005
High-Performance ICA
Parallel FastICA: Factor Increase In Performance
Over MATLAB fastica.m
 Over 130 times faster than
MATLAB fastica.m
 Greater than 8-fold increase in
200
180
160
Performance Scaling Factor
Parallel FastICA:
140
performance on 32 processors
120
100
80
60
40
Parallel Infomax: Factor Increase In Performance Over
MATLAB runica.m
Neuronic
pSeries 655
pSeries 690
20
4.0
0
1
2
4
6
8
10
12
14
16
18
20
22
24
26
28
30
32
Number of Processors
3.5
Parallel Infomax:
 Over 3 times faster than MATLAB
runica.m
 Greater than 3-fold increase in
performance on 4 processors
Performance Scaling Factor
3.0
2.5
2.0
1.5
1.0
Neuronic
pSeries 655
pSeries 690
0.5
0.0
1
2
4
6
Number of Threads (Processors)
NeuroInformatics Center
BBMI: Brain, Biology, Machine Initiative
Feb 2005
Brain, Machine, and Education

Pittsburgh Science of Learning Center (PSCL) Collaboration


LearnLab Research Facility (U. Pittsburgh, CMU)






http://pslc.hcii.cs.cmu.edu/tiki-index.php
Authoring tools for online courses, experiments, and integrated
computational learner models
Support for running in vivo learning experiments
Longitudinal microgenetic data from entire courses
Data analysis tools, including software for learning curve
analysis and semi-automated coding of verbal data
Parallel studies of learning using cognitive neuroscience (EEG,
fMRI) methods
Multidisciplinary Effort



Computer Science (e.g., Maxine Eskanazi, Jamie Callan — CMU)
Linguistics & ESL (e.g., Alan Juffs — U. Pittsburgh )
Psychology (Charles Perfetti — U. Pittsburgh)
NeuroInformatics Center
BBMI: Brain, Biology, Machine Initiative
Feb 2005