320 likes | 503 Views
Feb 28, 2010. NEMO data meta-analysis: Application of NEMO analysis workflow to consortium datasets (redux). http://nemo.nic.uoregon.edu. Overview of NEMO Project Aims. Design and test procedures for automated & robust ERP pattern analysis and classification
E N D
Feb 28, 2010 NEMO data meta-analysis: Application of NEMO analysis workflow to consortium datasets (redux) http://nemo.nic.uoregon.edu
Overview of NEMO Project Aims • Design and test procedures for automated & robust ERP pattern analysis and classification • Capture rules, concepts in a formal ERP ontology • Develop ontology-based tools for ERP data markup • Apply ERP analysis tools to consortium datasets • Perform meta-analyses of consortium data • Build data storage & management system
The three pillars of NEMO Focus of this All-Hands Meeting • ERP Ontologies • ERP Data • ERP Database & portal
TUTORIAL #1:Viewing ERP Data in EEGLAB TUTORIAL #2:Decomposition with PCA TUTORIAL #3:Segmentation with Microstates TODAY
TUTORIAL #4:Extracting ontology-based attributes And exporting to text or RDF TOMORROW
Overview Steps in Meta-analysis • Collect ERP data sets with compatible functional attributes • Decompose / segment the ERP data into discrete spatio-temporal patterns for analysis & labeling • Mark-up patterns within each dataset: labeling of spatial & temporal characteristics (functional labels assigned in step 1) • Cluster patterns within data sets • Link labeled clusters across data sets • Label linked clusters (i.e., establish mappings across patterns from different dataset)
Focus of 1st Annual All-Hands Meeting • Collect ERP data sets with compatible functional attributes • Decompose / segment the ERP data into discrete spatio-temporal patterns for analysis & labeling • Mark-up patterns within each dataset: labeling of spatial & temporal characteristics (functional labels assigned in step 1) • Cluster patterns within data sets • Link labeled clusters across data sets • Label linked clusters (i.e., establish mappings across patterns from different dataset)
Overview Steps in Meta-analysis • Collect ERP data sets with compatible functional attributes • Decompose / segment the ERP data into discrete spatio-temporal patterns for analysis & labeling • Mark-up patterns within each dataset: labeling of spatial & temporal characteristics (functional labels assigned in step 1) • Cluster patterns within data sets • Link labeled clusters across data sets • Label linked clusters (i.e., establish mappings across patterns from different dataset)
Combining Top-down and Bottom-up Methods for ERP Pattern Classification Gwen Frishkoff University of Pittsburgh Robert Frank, Haishan Liu, & Dejing Dou University of Oregon
Human Brain Mapping • Current Challenges • Tracking what we know • Ontologies • Integrating knowledge to achieve high-level understanding of brain–functional mappings • Meta-analyses • Important Considerations • Stay true to data (bottom-up) • Achieve high-level understanding (top-down) “Understanding without data is empty. Data without understanding are blind”
ERP Patterns 1,000 ms SPACE(Scalp Topography) TIME (in 10s of ms)
What do we know? Observed Pattern = “P100” iff • Event type is visual stimulus AND • Peak latency is between 70 and 160 ms AND • Scalp region of interest (ROI) is occipital AND • Polarity over ROI is positive (>0) ? FUNCTION TIME SPACE
Why does it matter? • Robust pattern rules would provide a good foundation for– • Development of ERP ontologies • Labeling of ERP data based on pattern rules • Cross-experiment, cross-lab meta-analyses
A Case Study • Simulated ERP datasets • PCA & ICA methods for spatial & temporal pattern analysis • Spatial & temporal metrics for labeling of discrete patterns • Revision of pattern rules based on mining of labeled data
Simulated ERPs (n=80) P100 N100 N3 MFN P300 + NOISE
ERP pattern analysis ✔ • Temporal PCA (tPCA) • Gives invariant temporal patterns (new bases) • Spatial variability as input to data mining • Spatial ICA (sICA) • Gives invariant spatial patterns (new bases) • Temporal variability as input to data mining • Spatial PCA (sPCA) ✔ X Multiple measures used for evaluation (correlation + L1/L2 norms)
Measure Generation Vector attributes = Input to Data mining (clustering & classification) T1 T2 S1 S2 Input to data mining: 32 attribute vectors, defined over 80 “individual” ERPs (observations) CoN ROI ± Centroids CoP A B
Data mining • Vectors of spatial & temporal attributes as input • Clustering observations patterns (E-M accuracy >97%) • Attribute selection (“Information gain”) ± Centroids Peak Latency CoN ✔ CoP • Figure 3. Info gain results for spatial ICA.
Revised Rule for the “P100” Pattern = P100v iff • Event type is visual stimulus AND • Peak latency is between 76 and 155 ms AND • Positive centroid is right occipital AND • Negative centroid is left frontal SPACE TIME FUNCTION
What we’ve learned • Bottom-up methods result in validation & refinement of top-down pattern rules • Validation of expert selection of temporal concepts (peak latency) • Refinement of expert specification of spatial concepts (± centroids) • Alternative pattern analysis methods (e.g., tPCA & sICA) provide complementary input to bottom-up (data mining) procedures
Some Preliminary Conclusions • Factor Retention may still be an issue for us collectively to explore • Unrestricted rotation vs. data reduction prior to rotation • For unrestricted path, what number to retain at end (after rotation)? • Also for unrestricted path, how to order factors at end (after rotation) • We agreed to explore these issues, try to decide on final analysis pipeline by some date in near future (TBD…)