SnPM: Statistical nonParametric Mapping A Permutation Test for PET & Second Level fMRI Thomas...
-
Upload
ethelbert-wright -
Category
Documents
-
view
219 -
download
0
Transcript of SnPM: Statistical nonParametric Mapping A Permutation Test for PET & Second Level fMRI Thomas...
SnPM:SnPM:Statistical nonParametric MappingStatistical nonParametric Mapping
A Permutation Test for PET &A Permutation Test for PET &Second Level fMRISecond Level fMRI
Thomas Nichols, University of MichiganThomas Nichols, University of MichiganAndrew Holmes, University of GlasgowAndrew Holmes, University of Glasgow
SnPM:SnPM:Statistical nonParametric MappingStatistical nonParametric Mapping
A Permutation Test for PET &A Permutation Test for PET &Second Level fMRISecond Level fMRI
Thomas Nichols, University of MichiganThomas Nichols, University of MichiganAndrew Holmes, University of GlasgowAndrew Holmes, University of Glasgow
B B B B B BA A A A A A
B B B B B BA A A A AA
B B B B B BA A A A A AB B B B B BA A A A A AB B B B B BA A A A A AB B B B B BA A A A A AB B B B B BA A A A A A
B B B B B BA A A A AAB B B B B BA A A A AAB B B B B BA A A A AAB B B B B BA A A A AAB B B B B BA A A A AA
t-statisticvariancemean difference
=
ran
dom
ise
12 subjects
6 B
A…
6 A
B…
differenceV5 PET activation experiment…
i
123456
789101112
……exampleexample……exampleexample
HH00:: scan would have been samescan would have been same
whatever the conditionwhatever the condition
– labelling as labelling as activeactive or or baselinebaseline arbitrary arbitrary
– re-label scans re-label scans equally likely statistic equally likely statistic imageimage• consider all possible relabellings consider all possible relabellings (exchangability)(exchangability)
permutation distributionpermutation distribution• of of eacheach voxel statistic voxel statistic ??
• of maximal voxel statisticof maximal voxel statistic
mean difference
smoothed variance
t-statistic“pseudo” t-statistic
variance
mean difference
SnPM with “pseudo” t-statistic
SPM with standard t-statisic
permutation distribution
SnPM with standard t-statisic? – similar!
SnPMSnPMSnPMSnPM
• SnPM:SnPM:
+ minimal assumptionsminimal assumptions• guaranteed validguaranteed valid
+ intuitive, flexible, powerfulintuitive, flexible, powerful
+ any statistic: voxel / summaryany statistic: voxel / summary
+ any summary statisticany summary statistic• maximum maximum pseudo tpseudo t – restricted – restricted
volume – cluster size / height / volume – cluster size / height / mass – omnibus testsmass – omnibus tests
– computational burdencomputational burden
– need sufficient relabellingsneed sufficient relabellings
• UsesUses• low dflow df• dodgy parametricdodgy parametric• no parametric resultsno parametric results
• SnPM:SnPM:
+ minimal assumptionsminimal assumptions• guaranteed validguaranteed valid
+ intuitive, flexible, powerfulintuitive, flexible, powerful
+ any statistic: voxel / summaryany statistic: voxel / summary
+ any summary statisticany summary statistic• maximum maximum pseudo tpseudo t – restricted – restricted
volume – cluster size / height / volume – cluster size / height / mass – omnibus testsmass – omnibus tests
– computational burdencomputational burden
– need sufficient relabellingsneed sufficient relabellings
• UsesUses• low dflow df• dodgy parametricdodgy parametric• no parametric resultsno parametric results
Non-parametric tests in Non-parametric tests in ffNI…NI…Non-parametric tests in Non-parametric tests in ffNI…NI…
• Classic testsClassic tests• Wilcoxon rank sum testWilcoxon rank sum test
• Kolmogorov-Smirnov testKolmogorov-Smirnov test
• Permutation testsPermutation tests• Holmes, Arndt Holmes, Arndt (PET)(PET)
• Bullmore, Locascio Bullmore, Locascio ((ffMRI)MRI)noise whitening, permutationnoise whitening, permutation
• Nichols & Holmes (Nichols & Holmes (ffMRI)MRI)label (re)-randomisationlabel (re)-randomisation
weak distributional assumptions
don’t assume normality replace data by ranks
lose information exchangeability
independence – fMRI minimal assumptions
exchangeabilityvalid often exactmultiple comparisons
via maximal statisticsflexible computational burden sufficient permutations additional power at low d.f.
via “pseudo” t-statistics
SnPM SnPM (standard (standard t)t)SnPM SnPM (standard (standard t)t)
• 12 scans12 scans
• 221212 permutations permutations
• All 2048/2 computedAll 2048/2 computed
• pp=1/2048=1/2048
• = 0.05 critical threshold:= 0.05 critical threshold: uu = 7.9248 = 7.9248
• Bonferoni critical threshold: Bonferoni critical threshold: uu = 9.0717 = 9.0717
• 30 min on Sparc Ultra 1030 min on Sparc Ultra 10
• 12 scans12 scans
• 221212 permutations permutations
• All 2048/2 computedAll 2048/2 computed
• pp=1/2048=1/2048
• = 0.05 critical threshold:= 0.05 critical threshold: uu = 7.9248 = 7.9248
• Bonferoni critical threshold: Bonferoni critical threshold: uu = 9.0717 = 9.0717
• 30 min on Sparc Ultra 1030 min on Sparc Ultra 10
SnPM SnPM (pseudo (pseudo tt))SnPM SnPM (pseudo (pseudo tt))
• 12 scans12 scans
• 221212 permutations permutations
• All 2048/2 computedAll 2048/2 computed
• p = p = 1/20481/2048
• = 0.05 critical threshold:= 0.05 critical threshold: uu = 5.120 = 5.120
• 40 min on a Sparc Ultra1040 min on a Sparc Ultra10
• 12 scans12 scans
• 221212 permutations permutations
• All 2048/2 computedAll 2048/2 computed
• p = p = 1/20481/2048
• = 0.05 critical threshold:= 0.05 critical threshold: uu = 5.120 = 5.120
• 40 min on a Sparc Ultra1040 min on a Sparc Ultra10
SnPM vs Parametric RF SnPM vs Parametric RF SnPM vs Parametric RF SnPM vs Parametric RF
Corrected Significance of ThresholdPermutation
RT Theory
Holmes AP, Blair RC, Watson JDG, Ford I (1996)“Non-Parametric Analysis of Statistic Images from Functional Mapping Experiments”Journal of Cerebral Blood Flow and Metabolism 16:7-22
Arndt S, Cizadlo T, Andreasen NC, Heckel D, Gold S, O'Leary DS (1996)“Tests for comparing images based on randomization and permutation methods”Journal of Cerebral Blood Flow and Metabolism 16:1271-1279
Nichols TE, Holmes AP (2002)“Nonparametric permutation tests for functional neuroimaging experiments: A primer with examples” Human Brain Mapping 15:1-25
Bullmore ET, Brammer M, Williams SCR, Rabe-Hesketh S, Janot N, David A, Mellers J, Howard R, Sham P (1995)“Statistical Methods of Estimation and Inference for Functional MR Image Analysis”Magnetic Resonance in Medicine 35:261-277
Locascio JJ, Jennings PJ, Moore CI, Corkin S (1997)“Time series analysis in the time domain and resampling methods for studies of functional magnetic resonance brain imaging” Human Brain Mapping 5:168-193
Raz J, Zheng H, Turetsky B (1999)“Statistical Tests for fMRI based on experimental randomisation” (ENAR Conference Proceedings)
Marchini JL, Ripley BD (2000)“A new statistical approach to detecting significant activation in functional MRI” NeuroImage
Holmes AP, Blair RC, Watson JDG, Ford I (1996)“Non-Parametric Analysis of Statistic Images from Functional Mapping Experiments”Journal of Cerebral Blood Flow and Metabolism 16:7-22
Arndt S, Cizadlo T, Andreasen NC, Heckel D, Gold S, O'Leary DS (1996)“Tests for comparing images based on randomization and permutation methods”Journal of Cerebral Blood Flow and Metabolism 16:1271-1279
Nichols TE, Holmes AP (2002)“Nonparametric permutation tests for functional neuroimaging experiments: A primer with examples” Human Brain Mapping 15:1-25
Bullmore ET, Brammer M, Williams SCR, Rabe-Hesketh S, Janot N, David A, Mellers J, Howard R, Sham P (1995)“Statistical Methods of Estimation and Inference for Functional MR Image Analysis”Magnetic Resonance in Medicine 35:261-277
Locascio JJ, Jennings PJ, Moore CI, Corkin S (1997)“Time series analysis in the time domain and resampling methods for studies of functional magnetic resonance brain imaging” Human Brain Mapping 5:168-193
Raz J, Zheng H, Turetsky B (1999)“Statistical Tests for fMRI based on experimental randomisation” (ENAR Conference Proceedings)
Marchini JL, Ripley BD (2000)“A new statistical approach to detecting significant activation in functional MRI” NeuroImage
Nonparametric approaches…Nonparametric approaches…Nonparametric approaches…Nonparametric approaches…