Raghu Machiraju Firdaus Janoos Fellow Harvard Medical Istavan Pisti Morocz Instuctor Harvard Medical Premise Understanding the mind not only requires a comprehension of the workings of lowlevel neural networks but also demands a detailed map of the brains functi ID: 1048175
Download Presentation The PPT/PDF document "Spatio-Temporal Models for Mental Proces..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1. Spatio-Temporal Models for Mental Processes from fMRIRaghu MachirajuFirdaus Janoos, Fellow, Harvard MedicalIstavan (Pisti) Morocz, Instuctor, Harvard Medical
2. PremiseUnderstanding the mind not only requires a comprehension of the workings of low–level neural networks but also demands a detailed map of the brain’s functional architecture and a description of the large–scale connections between populations of neurons and insights into how relations between these simpler networks give rise to higher–level thought
3. GoalsUnderstanding the representation of mental processes in functional neuroimagingDistributed interactionsSpace and time !Comparing processes across subjectsNeurophysiologic interpretability
4. OutlinefMRI AnalysisRepresentationsSpatio-temporal ModelsConclusion
5. What Is fMRI ?fMRI is a non-invasive tool for studying brain activity Spatio-temporal data (4D)Spatial resolution – mm Temporal resolutions – secs Functional specializationClassical neuroscienceFunctional integration Functional and effective
6. The fMRI SignalThe BOLD Effect Measure of cerebral metabolismTask relatedDefault-state networksConfounds/NuisanceRandom – thermal + quantum mechanical Structured component Distortions, physiological, motion, reconstruction
7. The BOLD EffectMeasure of “oxygenated blood” in the brainVolume of deoxyhemoglobin T2* weighted EPI sequencesThe exact coupling between neuronal activity and the BOLD signal unknownLinked primarily to metabolic activity at synapsesDepends on rCBF, rBVO2, rCMRO2 The hemodynamic response function is highly variable
8. fMRI NoiseAcquisitionReconstructionMagnetic fieldInhomogeneities InstabilityPhysiologic functionsAliased onto signalHead motionCorrelated with the taskRegistration / Correction
9. Classical Pipeline
10. fMRI AnalysisFunctional LocalizationStatic Activity MapsGLM, PCA, ICA, PLS, Functional IntegrationFunctional ConnectivityCCA, ICA, PCA, DBNEffective ConnectivitySEM, DCM, DBN
11. Typical DCM
12. Benefits fMRI provides information about the activity of large neural assembliesStatic pictures of the foci of activity and the interconnectionsMental processes arise from dynamic relationships between the neural substratesSpatially distributed, temporally transient and occur at multiple scales of space and time.Time resolved analysis Ordering of information processing
13. Cascadic Recruitment
14. State-of-the-ArtJanoos et al., EuroVis2009
15. Need Decoding !VOXEL-wise Representations LimitedDynamic ProcessesDistributed Representations Needed Beyond functional localizationWhere vs. how Distributed activity and functional interactionsPattern Classifiers Atoms of Thought for Cracking Neural Code
16. Haxby, 2001
17. Mitchell, 2008
18. ChallengesVery controlled experiments with copious trainingGeneral results have not always been positiveApplications to arbitrary settings ?Temporal nature of mental processesNeurophysiologic interpretabilityMulti-subject analysis
19. InspirationLehmann, 1994
20. Preliminary Resultsvisuo–spatial working memory
21. 2 Patients
22. Functional Networks
23. Functional Connectivity Estimation
24. Functional DistanceZt – activation patternsf - transportation
25. Cost Metric
26. Functional Distancet1t2t3
27. Algorithm
28. Mental ArithmeticInvolves basic manipulation of number and quantities Magnitude based system – bilateral IPS Verbal based system – left AGAttentional system – ps Parietal LobuleOther systems – SMA, primary visual cortex, liPFC, insula, etc
29. Paradigm
30. Clustering in Functional Space
31. Spatial Maps10 same as 88 auditory cortices6 , judgment5 Frontal, parietal lobes3 visual size estimation1 Visual Cortex0+5.0-5.00s 4s 8s106187305492
32. CritiqueNo neurophysiologic modelPoint estimatesHemodynamic uncertainty Temporal structureFunctional distance - an optimization problemNo metric structureExpensive !
33. Functional Distance
34. Cost Metric
35. Cost MetricDistortion minimizing
36. Feature Space ΦOrthogonal Bases Graph PartitioningNormalized graph Laplacian of F
37. Feature Space Φ
38. Feature Selection
39. State-Space ModelJanoos et al., MICCAI 2010
40. (Reduced) State-Space Model
41. Model Size SelectionTypically strike a balance between model complexity and model fit Information theoretic or Bayesian criteriaNotion of model complexityCross-validationIID Assumption
42. Maximally Predictive CriteriaMultiple spatio-temporal patterns in fMRINeurophysiological task related vs. default networksExtraneousBreathing, pulsatile, scanner driftSelect a model that is maximally predictive with respect to taskPredictability of optimal state-sequence from stimulus, s
43. DyscalculiaDifficulty in learning arithmetic that cannot be explained by mental retardation, inappropriate schooling, or poor social environmentCore conceptual deficit dealing with numbersVery common : 3-6% of school-age childrenHeterogeneous
44. Paradigm
45. Results Self – same subjectCross – train on one subject and predict on another
46. Comparing ModelsSubject 1...Subject 42Subject 2fMRI DataΦ1Φ2Φ42
47. MDS Plot
48. MDS Plot
49. DrawbacksApproximations in the modelElimination of the activity pattern layerSpatially unvarying hemodynamicsUnsupervised approachNo explicit link to the experimentMay not necessarily learn relevant patterns
50. Semi-supervised ApproachLoose dependency between stimulus and signalNot preclude discovery of un-modeled effectsStabilize estimationGeneralizable to unconstrained designsFunctionally well-defined representation
51. The ModelJanoos et al., IPMI 2011Janoos et al., NeuroImage 2011
52. EM Algorithm
53. Mean Field Approximation
54. Estimation
55. DyslexiaSelective inability to build a visual representation of a word, used in subsequent language processing, in the absence of general visual impairment or speech disordersAffects 5-10% of the populationSpelling, phonological processing, word retrievalDisorder of the visual word form systemMultiple varietiesOccipital, temporal, frontal, cerebellum
56. Paradigm
57. Comparative Results
58. Overall Results
59. Overall ResultsGroupControlDyscalculicDyslexicK22.8 ± 3.3424.5 ± 5.2223.7± 3.91Error0.2879±0.0480.36414±0.0440.3199±0.047
60. Experimental Variables
61. 123Spatial Maps
62. Hemodynamic ResponsesMotor CortexIntra Parietal Sulcus
63. MDS Plots
64. MDS Plots
65. Phase 1Phase 2Phase 1: Product SizePhase 2: Problem DifficultyMDS Plots (2)
66. ConclusionProcess model for fMRI Spatial patterns and the temporal structureIdentification of internal mental processesNeurophysiologically plausibleTest for the effects of experimental variablesParameter interpretationComparison of mental processesAbstract representation of patterns
67. Thank You