Transcript Data Analysis & Metric Generation

Meta Analysis First Steps
Data Analysis, Metric Generation and
Extracted Pattern Annotation
Project Goals for Meta-Analysis
• Subgoal #1: Complete first statistical meta-analysis of
ERP patterns from NEMO consortium datasets
– Target first paper submission by May 2009
• Subgoal #2: Compare pattern mappings from
different meta-analyses & establish functionally
relevant links between patterns
– Lexical, semantic, & memory-related ERP
– Establish meaningful pattern classes, hierarchies based on
meta analyses results
Meta Summary of ERP Data Meta Analysis
Analyze
Mark-up
Label
Cluster
Link
Label Linked Clusters
Publish
(rinse and repeat)
Meta Analysis Steps
• Obtain ERP data sets with compatible functional constraints
– NEMO consortium data
• Decompose / segment the ERP data into discrete spatio-temporal patterns
– PCA / ICA / Microstate Segmentation
• Mark-up patterns with their categorical, functional and spatio-temporal
characteristics
– NEMOautolabel
• Label patterns
• Cluster patterns within data sets
• Link labeled clusters across data sets
• Label linked clusters
• Publish
Datasets for Meta-analysis #1
Datasets for Meta-analysis #2 & #3
Techniques for Decomposing / Segmenting ERP
Data Into Discrete Spatio-Temporal Patterns
• Component Separation
– PCA: Principal Components Analysis
• Established protocol with supporting literature
– Dien, Frishkoff, Kayser & Tenke
• Applied to 9 consortium data sets from 3 separate labs
– ICA: Independent Components Analysis
• Established protocol with supporting literature, though less extensive
than PCA ERP research
– Makeig et al
• Mixed results / interpretation difficulties w.r.t. consortium data
• Automated Windowing / Microstate Segmentation
• Established protocol with supporting literature
– Lehman, Koenig, Murray
• In progress: currently adding to NEMOautolabel
Microstate Segmentation Overview
Simulated Microstates
• Simulated data set of 4 distinct, finite duration, topographies
(microstates). Note topographies are partially overlapping
Microstate Segmentation Overview
Microstate Boundaries
• Microstate border probability function (MSBPF): Quantifies
probability of topographic change as a function of time
Overview of PCA
• Temporal PCA
– Variables: Time samples
– Observations: Channel waveforms across conditons + subjects
– Relationship matrix quantifies temporal correlations
Basis of approach for decomposing and statistically quantifying NEMO
consortium data
• Spatial PCA
–
–
–
–
Variables: Channel locations
Observations: Spatial topographies across conditons + subjects
Relationship matrix quantifies spatial correlations
Problematic due to high spatial overlap of patterns from volume conduction
Not used due to concerns of misallocation of variance / factor splitting
PCA Decomposition Protocol
for Analysis of Consortium Data
• Dien PCA Toolbox / ERP Toolkit
• ReadSegRaw / PCAtoRaw to import from and export to
EGI segmented simple-binary files
• Temporal PCA algorithms
–
–
–
–
–
Covariance relationship matrix
Kaiser factor loading normalization
Retain all factors prior to any and all rotations
Varimax rotation followed by Promax relaxation
Statistically analyze 25 post-rotation factors, sorted in order of
decreasing projected variance (based on FacVar)
PCAtoRaw / ERP PCA Toolbox
• PCAtoRaw run-time parameters
• PCAtoRaw invokes Dien ERP PCA Toolbox
PCAtoRaw / ERP PCA Toolbox
• Npraw data PCA decomposition summary
PCA Decomposition Protocol
Factor Retention – Part I
• Pre-rotation factor retention
– “The problem of the number
of components”
• Scree test
– Linear scale may underestimate factor
retention
• Parallel test
– Compare scree of experimental data to
scree of random data of equal dimensions
• Full pre-rotation retention
– Kayser & Tenke proposal
– Factor retention, pre-rotation, affects both explained variance and
rotation outcome
– Full pre-rotation retention eliminates effect of retention subjectivity on
rotation outcome
PCA Decomposition Protocol
Factor Retention – Part 2
• Post-rotation factor retention
– Determine number of retained
components for adequate
reconstruction of scalp recorded ERP
• Retained components represent
majority of ERP variance
– Factors are sorted on FacVar, the fraction
of data variance accounted for by each individual factor
– Default post-rotation sort order of ERP PCA Toolkit
– Statistical analysis, via NEMOautolabel, performed on retained factors
– Flag retained factors with “robust” variance or high relative Global Field Power
– Flag retained factors containing spatiotemporal characteristics of target patterns
PCAtoRaw / ERP PCA Toolbox
NPraw.raw Test Dataset Results
• PCAtoRaw output files:
– .log:
– .mat:
– .fig:
– .raw:
Summary run statistics
MATLAB workspace variables
Pre- and post-rotation factor
scree plots
Factor loadings projected back to
channel space
 One set for each conditon / cell
 Grand average (_G.raw) or subject-specific (_S##.raw)
PCAtoRaw / ERP PCA Toolbox
NPraw.raw Test Dataset Results
• Examine NPraw_G.raw factor waveforms, in scalp-surface space, at each
channel across conditions (TopoPlot Mode)
Factor 1 waveforms (0-900ms; 0.1uv/mm). Note condition effects / factor separation at centropariertal and anterior ventral sites
PCAtoRaw / ERP PCA Toolbox
NPraw.raw Test Dataset Results
• Examine NPraw_G.raw factor topographies, in scalp-surface space, at peak
intensity across conditions (TopoMap Mode)
Factor 1 scalp-surface topographies, 600ms post-stimulus, for the 4 NPraw
experimental conditions (L to R): ConFinal, ConMid, InconFinal, InconMid
NEMOautolabel
Marking up ERP Components / Microstates: NEMO_data
•
•
Mark-up observed patterns (components / microstates) with user-specified information
on the experimental procedure and subject group
Each mark-up element (NEMOautolabel label) has a unique NEMOautolabel ID and will
map to a corresponding element in the NEMO ontology
ERP_CompAnalysisMethod
Cond_Stan
ExptID
EEG_Montage
Event_Modality
SessID
Event_Type
CellNo
Stim_Type
CellLabel
ERP_ObsID
SubjID
Subject_Group
ERP_ObservedPattern
NEMOautolabel_Name
NEMOautolabel_Def
NEMOautolabel_ID
NEMOlex_Name
NEMOlex_ID
ExptID
ExptID represents
"experiment ID" and specifies
the experimental procedure
and subject group.
AL:0000003
experiment_id
NM:0000059
NEMOautolabel
Marking up ERP Components / Microstates: NEMO_data
•
•
Mark-up observed patterns (components / microstates) with their temporal
characteristics
MATLAB-based functions extract temporal metrics for each condition, subject and
component / microstate
–
–
Data driven
Harnesses expert-knowledge: Domain experts specify the temporal characteristics of interest
Ti_Max
Ti_Max_round
TI_Begin
TI_End
TI_Dur
TI_Dur_round
NEMOautolabel_Name
NEMOautolabel_Def
NEMOautolabel_ID
NEMOlex_Name
NEMOlex_ID
Ti_Max
Ti_Max specifies for each
temporal component the
time point of its peak
absolute intensity, in
milliseconds.
AL:0000019
ERP_pattern_peak_latency
NM:0000047
NEMOautolabel
Marking up ERP Components / Microstates: NEMO_data
•
•
Mark-up observed patterns (components / microstates) with their spatial characteristics
MATLAB-based functions extract spatial metrics for each condition, subject and
component / microstate
–
–
Data driven
Harnesses expert-knowledge: Domain experts specify the spatial characteristics of interest
COP_X2d
LatIndex_Threshold
EGICh_COP
COP_Y2d
Laterality_COP
ITTCh_COP
CON_X2d
LatIndex_COP
ROI_COP
CON_Y2d
ROInolat_COP
EGICh_CON
COP_X3d
Laterality_CON
ITTCh_CON
COP_Y3d
LatIndex _CON
ROI_CON
COP_Z3d
ROInolat_CON
CON_X3d
CON_Y3d
CON_Z3d
NEMOautolabel_Name
NEMOautolabel_Def
NEMOautolabel_ID
NEMOlex_Name
NEMOlex_ID
ITTCh_COP
International 10-10 electrode
location closest to the
component pair's center-ofpositivity xy-coordinate pair
(COP_X2d, COP_Y2d), in L2norm, on a montage-specific
2-D flat map of scalp-surface
electrode locations.
AL:0000036
TBA
TBA
References
PCA
Dien, J. (1998). Addressing misallocation of variance in principal components analysis of event-related potentials. Brain Topogr,
11(1), 43-55.
Dien, J., & Frishkoff, G. A. (2005). Introduction to principal components analysis of event-related potentials. In T. Handy (Ed.),
Event-Related Potentials: A Methods Handbook. (pp. 189-208). Cambridge, MA: MIT Press.
Dien, J., Beal, D. J., & Berg, P. (2005). Optimizing principal components analysis of event-related potentials: matrix type, factor
loading weighting, extraction, and rotations. Clin Neurophysiol, 116(8), 1808-1825.
Dien, J. (2006). Progressing towards a consensus on PCA of ERPs. Clin Neurophysiol, 117(3), 699-702; author reply 703-697.
Dien, J., Khoe, W., & Mangun, G. R. (2007). Evaluation of PCA and ICA of simulated ERPs: Promax vs. Infomax rotations. Hum Brain
Mapp, 28(8), 742-763.
Dien, J. (2009). Evaluating two-step PCA of ERP data with Geomin, Infomax, Oblimin, Promax, and Varimax rotations.
Psychophysiology.
Kayser, J., & Tenke, C. E. (2003). Optimizing PCA methodology for ERP component identification and measurement: theoretical
rationale and empirical evaluation. Clin Neurophysiol, 114(12), 2307-2325.
Kayser, J., & Tenke, C. E. (2005). Trusting in or breaking with convention: towards a renaissance of principal components analysis in
electrophysiology. Clin Neurophysiol, 116(8), 1747-1753.
References
ICA
Dien, J., Khoe, W., & Mangun, G. R. (2007). Evaluation of PCA and ICA of simulated ERPs: Promax vs. Infomax rotations.
Hum Brain Mapp, 28(8), 742-763.
Microstate Analysis
Michel, C. M., Murray, M. M., Lantz, G., Gonzalez, S., Spinelli, L., & Grave de Peralta, R. (2004). EEG source imaging. Clin
Neurophysiol, 115(10), 2195-2222.
Murray, M. M., Brunet, D., & Michel, C. M. (2008). Topographic ERP analyses: a step-by-step tutorial review. Brain
Topogr, 20(4), 249-264.
Koenig, T., Kochi, K., & Lehmann, D. (1998). Event-related electric microstates of the brain differ between words with
visual and abstract meaning. Electroencephalogr Clin Neurophysiol, 106(6), 535-546.
Koenig, T., & Lehmann, D. (1996). Microstates in language-related brain potential maps show noun-verb differences.
Brain Lang, 53(2), 169-182.
Lehman, D., & Skrandies, W. (1985). Spatial analysis of evoked potentials in man - A review. Progress in Neurobiology,
23, 227-250.
Pizzagalli, D., Lehmann, D., Koenig, T., Regard, M., & Pascual-Marqui, R. D. (2000). Face-elicited ERPs and affective
attitude: brain electric microstate and tomography analyses. Clin Neurophysiol, 111(3), 521-531.
References
Annotating functional attributes
Fox, P. T., Laird, A. R., Fox, S. P., Fox, P. M., Uecker, A. M., Crank, M., et al. (2005). BrainMap taxonomy of
experimental design: description and evaluation. Hum Brain Mapp, 25(1), 185-198.
Spatial & temporal metric generation
Handy, T. (2005). Basic Principles of ERP Quantification. In T. Handy (Ed.), Event-Related Potentials: A
Methods Handbook (pp. 33–56). Cambridge, MA: MIT Press.
Luck, S. (2005). An Introduction to the Event-Related Potential Technique Boston, MA: The MIT Press.
Otten, L. J., & Rugg, M. D. (2005). Interpreting Event-Related Brain Potentials. In T. Handy (Ed.), EventRelated Potentials: A Methods Handbook (pp. 3–16). Cambridge, MA: MIT Press.
Picton, T. W., Bentin, S., Berg, P., Donchin, E., Hillyard, S. A., Johnson, R., Jr., et al. (2000). Guidelines for
using human event-related potentials to study cognition: recording standards and publication
criteria. Psychophysiology, 37(2), 127-152.