Preprocessing for EEG & MEG Tom Schofield & Ed Roberts.

Preprocessing for EEG & MEG

Tom Schofield & Ed Roberts

Data acquisition

Data acquisition

Using Cogent to a generate marker pulse..

drawpict(2);

outportb(888,2); tport=time;

waituntil(tport+100); outportb(888,0);

logstring( [‘displayed ‘O’ at time ' num2str(time) ]);

Two crucial steps Activity caused by your stimulus (ERP) is

‘hidden’ within continuous EEG stream ERP is your ‘signal’, all else in EEG is

‘noise’ Event-related activity should not be

random, we assume all else is Epoching – cutting the data into chunks

referenced to stimulus presentation Averaging – calculating the mean value

for each time-point across all epochs

Extracting ERP from EEG

ERPs emerge from EEG as you average trials together

Overview

Preprocessing steps Preprocessing with SPM What to be careful about What you need to know about

filtering

mydata.mat

Epoching

Epoching - SPM

Creates: e_mydata.mat

Downsampling

Nyquist Theory – minimum digital sampling frequency must be > twice the maximum frequency in analogue signal

Select ‘Downsample’ from the ‘Other’ menu

Downsample

Creates: de_mydata.mat

Artefact rejection

BlinksEye-movementsMuscle activityEKGSkin potentialsAlpha waves

Artefact rejection

BlinksEye-movementsMuscle activityEKGSkin potentialsAlpha waves

Artefact rejection - SPM

Creates: ade_mydata.mat

Artefact correction Rejecting ‘artefact’ epochs costs you

data Using a simple artefact detection

method will lead to a high level of false-positive artefact detection

Rejecting only trials in which artefact occurs might bias your data

High levels of artefact associated with some populations

Alternative methods of ‘Artefact Correction’ exist

Artefact correction - SPM SPM uses a

robust average procedure to weight each value according to how far away it is from the median value for that timepoint

WeightingValue

Outliers are given

less weight

Points close to median

weighted ‘1’

Artefact correction - SPM

Normal average

Robust Weighted Average

Robust averaging - SPM

Creates: ade_mydata.mat

Artefact Correction

ICA Linear trend detection Electro-oculogram ‘No-stim’ trials to correct for

overlapping waveforms

Artefact avoidance

Blinking Avoid contact lenses Build ‘blink breaks’ into your paradigm If subject is blinking too much – tell them

EMG Ask subjects to relax, shift position, open mouth slightly

Alpha waves Ask subject to get a decent night’s sleep beforehand Have more runs of shorter length – talk to subject in between Jitter ISI – alpha waves can become entrained to stimulus

Averaging

R = Noise on single trialN = Number of trials

Noise in avg of N trials (1/√N) x R

More trials = less noiseDouble S/N need 4 trialsQuadruple need 16 trials

Averaging

Creates: made_mydata.mat

Averaging

Assumes that only the EEG noise varies from trial to trial

But – amplitude will vary But – latency will vary Variable latency is usually a bigger

problem than variable amplitude

Averaging: effects of variance

Latency variation can be a significant problem

Latency variation solutions

Don’t use a peak amplitude measure

Time Locked Spectral Averaging

Other stuff you can do – all under ‘Other’ in GUI

Merge data sessions together Calculate a ‘grand mean’ across

subjects Rereference to a different

electrode FILTER

Filtering

Why would you want to filter?

Potential Artefacts

Before Averaging… Remove non-neural voltages Sweating, fidgeting Patients, Children Avoid saturating the amplifier Filter at 0.01Hz

Potential Artefacts

After Averaging…

Filter Specific frequency bands Remove persistent artefacts Smooth data

Types of Filter

1. Low-pass – attenuate high frequencies

2. High-pass – attenuate low frequencies

3. Band-pass – attenuate both

4. Notch – attenuate a narrow band

Properties of Filters

“Transfer function”1. Effect on amplitude at each frequency2. Effect on phase at each frequency

“Half Amp. Cutoff”1. Frequency at which amp is reduced by

50%

High-pass

Low-pass

Band-pass and Notch

Problems with Filters

Original waveform, band pass of .01 – 80Hz

Low-pass filtered, half-amp cutofff = ~40Hz

Low-pass filtered, half-amp cutofff = ~20Hz

Low-pass filtered, half-amp cutofff = ~10Hz

Filtering Artefacts “Precision in the time domain is inversely related to

precision in the frequency domain.”

Filtering in the Frequency Domain

AB C

D E

Filtering in the Time Domain

Filtering in the time domain is analogous to smoothing

At a given point an average is calculated in relation to two nearest neighbours or more

X+1

X-1

X

Filtering in the Time Domain

Waveform progressively filtered by averaging the surrounding time points.

Here x = ((x-1)+x+(x+1))/3

Recipe for Preprocessing

1. Band-pass filter e.g.0.1 – 40Hz

2. Epoch

3. Check/View

4. Merge

5. Downsample?

6. Artefacts; Correction/Rejection

7. Filter

8. Average

Recommendations

1. Prevention is better than the cure

2. During amplification and digitization minimize filtering

3. Keep offline filtering minimal, use a low-pass

4. Avoid high-pass filtering

Summary

1. No substitute for good data2. The recipe is only a guideline3. Calibrate4. Filter sparingly5. Be prepared to get your hands

dirty

References

An Introduction to the Event-related Potential Technique, S. J. Luck

SPM Manual