methods for dummies m/eeg analysis: contrasts, inferences and source localisation

27
Methods for Dummies M/EEG Analysis: Contrasts, Inferences and Source Localisation Diana Omigie Stjepana Kovac

Upload: december

Post on 05-Jan-2016

44 views

Category:

Documents


1 download

DESCRIPTION

Methods for Dummies M/EEG Analysis: Contrasts, Inferences and Source Localisation. Diana Omigie Stjepana Kovac. Last week revisited. What do we measure with EEG and MEG? Why use these techniques? What do we do we do with the raw data we record? Downsampling Montage Mapping Epoching - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

Methods for Dummies

M/EEG Analysis:Contrasts, Inferences and

Source Localisation

Diana OmigieStjepana Kovac

Page 2: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

Last week revisited

What do we measure with EEG and MEG?

Why use these techniques?

What do we do we do with the raw data we record?• Downsampling• Montage Mapping• Epoching• Filtering• Artefact Removal• Averaging

Page 3: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

What Next?Event related potentials (ERPs) are signal-averaged epochs of EEG that are time-locked to the onset of a stimulus

A waveform is a time series that plots scalp voltage (µV, T) over time (ms)

However,We might also want

• to carry out statistics comparing between conditions, subjects ..

• to localise the generators of the electrical activity

0100200300400500600-300-200-1000100200300msfemto T MRT15 (235)trial 1 1 trial 2 2

0100200300400500600-300-200-1000100200300msfemto T MRT15 (235)trial 1 1 trial 2 2

0100200300400500600-300-200-1000100200300

msfemto T MRT15 (235)trial 1 1 trial 2 2

Page 4: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

This week

Contrast and Inferences Source Reconstruction

Page 5: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

Contrasts and Inferences using SPM 8

……Which buttons do we need to press?

Page 6: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

EEG data acquired on 128 channel ActiveTwo system sampled at 2048Hz

Randomised presentation of 86 faces and 86 scrambled faces

Experimental Paradigm

Page 7: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

Aim:Identify at what point in time andover what sensor area the greatest

difference lies in the responses to faces and non faces.

Steps

Page 8: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

2D Interpolation

Transformation of discreet channels into a continuous 2D interpolated image of M/EEG signals

Sensor Space Scalp Space

Page 9: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

MULTI DIMENSIONALSCALP SPACE

create a 2D space by flattening the sensor locations and interpolating between them to create an image of M*M pixels ( where M=number of channels)

or

Create a 3 D space with time as added dimension. M*M*S (where S= number of samples)

Page 10: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

MULTI DIMENSIONALSCALP SPACE

2Dcreate a 2D space by flattening the sensor locations and interpolating between them to create an image of M*M pixels ( where M=number of channels)

or 3DCreate a 3 D space with time as added dimension. M*M*S (where S= number of samples)

Page 11: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

MULTI DIMENSIONALSCALP SPACE

2Dcreate a 2D space by flattening the sensor locations and interpolating between them to create an image of M*M pixels ( where M=number of channels)

or 3DCreate a 3 D space with time as added dimension. M*M*S (where S= number of samples)

Page 12: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

MULTI DIMENSIONALSCALP SPACE

2Dcreate a 2D space by flattening the sensor locations and interpolating between them to create an image of M*M pixels ( where M=number of channels)

or 3DCreate a 3 D space with time as added dimension. M*M*S where S= number of samples

Page 13: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

MULTI DIMENSIONALSCALP SPACE

2Dcreate a 2D space by flattening the sensor locations and interpolating between them to create an image of M*M pixels ( where M=number of channels)

or 3DCreate a 3 D space with time as added dimension. M*M*S where S= number of samples

Time

Page 14: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

Background

Random Field theory allows us to: • make N dimensional spaces from sensor locations. • take into account the spatial correlation across pixels.

• correct for multiple statistical comparisons.

Page 15: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

SPM 8Steps

1st

L

E

V

E

L

Page 16: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

New directory

Faces & Scrambled faces

3D image file for each trial with dimension

32x 32x 161

1st

L

E

V

E

L

Page 17: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

Sections through X and Y expressed over time

2D x-y space interpolated from the flattened electrode

locations at one point in time

1st

L

E

V

E

L

3D IMAGE

Page 18: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

1st level analysis of EEG data is• not about modeling the data ( as in fMRI)• the transformation of data from

filename.mat and filename.dat format to image files (N1fT1 format)

• a necessary step to create the images which we carry out 2nd level analysis on

1st

L

E

V

E

L

Page 19: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

1st level analysis button

• Used only when you know in advance the time window that you are interested in.

• The Specify 1st level button results in a 2D image with just spatial dimensions.

Page 20: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

2nd

L

E

V

E

L

Page 21: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

Smoothing

• Important step to take before 2nd level analysis (In SPM, use smooth images function in the drop down other menu)

• Used to adjust images so that they better conform to the assumptions of random field theory

• Necessary for taking into consideration spatial and temporal variability between subjects

• General guiding principle: Let smoothing kernel match the data feature you need to enhance. Try to smooth the images with different kernels and see what looks best.

Page 22: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

2nd

L

E

V

E

L

Which Buttons Do weNeed to Press?

Page 23: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

2nd

L

E

V

E

L

Create a new directory Then

To produce a batch windowSelect directory created

Select two sample t test as designMake group 1 contain 1st type of trials

Make group 2 contain other type.

Save batch descriptionRun batch window

Page 24: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

2 sample t-testDesign Matrix

click

click

2nd

L

E

V

E

L

Page 25: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation
Page 26: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

Result showing regions within epochs where faces and non faces differ reliably

Maxima [-13 -78 180] & [21 -68 180]

Coordinates correspond to the left and right posterior sites at 180ms

Page 27: Methods for Dummies M/EEG Analysis: Contrasts, Inferences and   Source Localisation

Time-frequency analysisTransform data into frequency spectrum

-100 -50 0 50 100 150 200 250 300 350 400

-60

-40

-20

0

20

40

60

ms

fem

to T

MRO33 (200)

-100 0 100 200 300 400

5

10

15

20

25

30

35

40

45

ms

Hz

MRO33 (200)

-1500

-1000

-500

0

500

1000

1500

Ideal for induced responses i.e. responses not phase locked to the stimulus onsetDifferent methods but SPM uses the Morlet Waveform Transform ( mathematical functions which breaks a signal into different components)Trade off between time resolution and frequency resolution