Statistical Analysis of M/EEG Sensor- and Source-Level Data Jason Taylor

Statistical Analysis of M/EEG Sensor- and Source-Level Data

Jason Taylor MRC Cognition and Brain Sciences Unit (CBU)

Cambridge Centre for Ageing and Neuroscience (CamCAN)

Cambridge, UK

[email protected] | http://imaging.mrc-cbu.cam.ac.uk (wiki)

09 May 2011 | London | Thanks to Rik Henson and Dan Wakeman @ CBU

Practical aspects of…

The SPM approach to M/EEG Statistics:

A mass-univariate statistical approach to inference regarding effects in space/time/frequency (using replications across trials or subjects).

• In Sensor Space:

2D Time-Frequency

3D Topography-by-Time

• In Source Space:

3D Time-Frequency contrast images

SPM vs SnPM vs PPM vs FDR

Other issues & Future directions

Overview

The Multiple Comparison Problem

Random Field Theory (RFT) is a method for correcting for multiple statistical comparisons with N-dimensional spaces (for parametric statistics, e.g., Z-, T-, F- statistics).

Note these comparisons are often made implicitly, e.g., by viewing (“eye-balling”) data before selecting a time-window or set of channels for statistical analysis! RFT correction depends on the search volume.

-> When is there an effect in time, e.g., GFP (1D)?

-> When/What frequency is there an effect in time and frequency (2D)?

-> When/Where is there an effect in sensor topography space and time (3D)?

-> When/Where is there an effect in source space and time (4-ish D)?

Worsley Et Al (1996). Human Brain Mapping, 4:58-73

CTF Multimodal Faces Dataset | SPM8 Manual (Rik Henson)

CTF MEG & EEG Data:128 EEG275 MEG3T fMRI (with nulls)1mm3 sMRI

2 sessions

~160 face trials and ~160 scrambled trials per session

Group: N=12 subjects, as in Henson et al, 2009 a,b,c

Chapter 33 SPM8 Manual

Neuromag Faces Dataset | Dan Wakeman & Rik Henson @ CBU

Biomag 2010 Award-Winning Dataset!

Neuromag MEG & EEG Data:70 EEG102 Magnetometers204 Planar GradiometersH&VEOGECG1mm3 sMRI

6 sessions

faces & scrambled-face images (480 trials)

Group: N=18

Neuromag Lexical Decision Dataset | Taylor & Henson @ CBU

Taylor & Henson, submitted

Neuromag MEG Data:102 Magnetometers204 Planar GradiometersH&VEOGECG1mm3 sMRI

4 sessions

words & pseudowords presented visually (480 trials)

Group: N=18

Fac

esS

cram

ble

d

Faces > Scrambled

Where is an effect in time-frequency space?

Kilner et al., 2005, Nsci LettersCTF Multimodal Faces

Single MEG channelMean over trials of Morlet wavelet projection (i.e., induced + evoked)-Write t x f image per trial-Compute SPM, RFT correct

Fac

es >

Scr

amb

led

Where is an effect in time-frequency space?

Neuromag Faces

EEG MEG (Mag)

100-220ms8-18Hz

Fre

q (H

z)

6

9

0

-500 +1000 Time (ms)

Fre

q (H

z)

6

9

0

-500 +1000 Time (ms)

Single channel-Effect over subjects

Where is an effect in sensor-time space?

Topography-x-Time Image (EEG)

Stats (F): Faces vs Scrambled(single subject, over trials)

CTF Multimodal Faces

GLM: Removing variance due to confounds

Each trial

Each trial-type (6)

beta_00* images reflect mean (adjusted) 3D topo-time volume for each condition

Confounds (4)

Henson et al., 2008, NImage

W/in Subject(1st-level) model

Without transformation to Device Space

With transformation to Device Space

Where is an effect in sensor-time space?

Analysis over subjects (2nd Level)NOTE for MEG: Complicated by variability in head-positionSOLUTION: Virtual transformation to same position in sessions, subjects

Taylor & Henson (2008) BiomagNeuromag Lexical Decision

Stats over 18 subjects, RMS of planar gradiometers

*Improved* (i.e., more blobs, larger T values) w/ transform

Where is an effect in source space (3D)?

Note sparseness of MSP inversions….

Analysis Mask

Source Analysis over N=12 subjects, MEG: 102 mags, MSP, evoked, RMS of source energy, smoothed 12-mm kernel

STEPS:

Estimate evoked/induced energy (RMS) at each dipole for a certain time-frequency window of interest

(e.g., 100-220ms, 8-18 Hz)

For each condition (Faces, Scrambled)

Smooth along 2D surface

Write data to 3D image (in MNI space)

Smooth by 3D Gaussian

Henson et al., 2007, NImage


MEG E+MEGA EEG

Neuromag Faces


Sensors

Over Participants

Over Voxels

Mags Grads

Classical SPM approach

Caveats:

-Inverse operator induces long-range error correlations (e.g., similar gain vectors from non-adjacent dipoles with similar orientation), making RFT conservative

-Need a cortical mask, else activity ‘smoothed’ outside

-Distributions over subjects may not be Gaussian (e.g., if sparse, as in MSP)

(Taylor & Henson, submitted; Biomag 2010)

Neuromag Lexical Decision


Sources (fused sensors)

Over Participants

Over Voxels

MSP IID

Classical SPM approach

Caveats:

-Inverse operator induces long-range error correlations (e.g., similar gain vectors from non-adjacent dipoles with similar orientation), making RFT conservative

-Need a cortical mask, else activity ‘smoothed’ outside

-Distributions over subjects may not be Gaussian (e.g., if sparse, as in MSP)


Neuromag Lexical Decision


p<.05 FWENon-Parametric Approach (SnPM)

Robust to non-Gaussian distributions

Less conservative than RFT when dfs<20

Caveats:

No idea of effect size (e.g., for power, future expts)

Exchangeability difficult for more complex designs

(Taylor & Henson, Biomag 2010)

SnPM Toolbox by Holmes & Nichols:http://go.warwick.ac.uk/tenichols/software/snpm/



p>.95 (γ>1SD)Posterior Probability Maps (PPMs)

Bayesian Inference

No need for RFT (no MCP)

Threshold on posterior probabilty of an effect greater than some size

Can show effect size after thresholding

Caveats:

Assume Gaussian distribution (e.g., of mean over voxels), sometimes not met (ok for IID…?)




Topological FDR correction

Choose an uncorrected threshold (e.g., p<.001) to define topologial features (e.g., peak and cluster size)

Topological FDR actually produces higher corrected p-values (i.e., fewer suprathreshol voxels) than FWE in the data used here

Caveat:

If sources are constrained to a gray matter cortical surface, are topological features as meaningful?

(Taylor & Henson, Biomag 2010)

SPM p<.001 unc


Taylor & Henson, submittedNeuromag Lexical Decision

Condition x Time-windowInteractions

Factorising time allows you to infer (rather than simply describe) when effects emerge or disappear.

We've used a data-driven method (hierarchical cluster analysis) in sensor space to define time-windows of interest for source-space analyses.

* high-dimensional distance between topography vectors

* cut the cluster tree where the solution is stable over longest range of distances

Where and When so effects emerge/disappear in source space (3D)?





* estimate sources in each sub-time-window

* submit to GLM with conditions & time-windows as factors (here: Cond effects per TWin)






* estimate sources in each sub-time-window

* submit to GLM with conditions & time-windows as factors (here: Cond x TW interactions)


Future Directions

• Extend RFT to 2D cortical surfaces (+1D time)? (e.g., Pantazis et al., 2005, NImage)

• Novel approaches to the multiple comparison problem (e.g., 'unique extrema' in sensors: Barnes et al., 2011, NImage)

• Go multivariate? To identify linear combinations of spatial (sensor or

source) effects in time and space(e.g., Carbonnell, 2004, NImage)

To detect spatiotemporal patterns in 3D images(e.g., Duzel, 2004, NImage; Kherif, 2003, NImage)

- The End -

Thanks!

Statistical Analysis of M/EEG Sensor- and Source-Level Data Jason Taylor

Documents

Transcript of Statistical Analysis of M/EEG Sensor- and Source-Level Data Jason Taylor