HC 3 Flashcards
(33 cards)
What are the steps for data-analysis ERP?
- Preprocessing
- Main signal processing
- Statistical testing
What do we do when preprocessing?
-(Re-)reference of the EEG signals
-Preprocessing reduces noise signals and formats EEG data for the main signal processing
-filtering
-correction of eye movements
artefacts
-artefact rejection
-segmentation
What are the two ways to do re-referencing (offline)?
- Mastoid reference
Cz - ((Lm+Rm)/2) - Average reference
Cz - ((Fz+F3+F4…O1+Oz+O2)/nr of electrodes)
Why do we have to take account for re-referencing?
ERPs can look very different with different references. This has to do with the distance between sensors. It affects ERP morphology, but not topography. Different references sensors are:
-Nose
-Average
-Earlobes
-Non-cephalic
-Mastoids
What is noise?
Electric signals not from the brain.
What are the common types of noise?
EMG= electromyogram for face movement
EOG= electrooculogram for eye movement. Blinking is the most occuring noise.
ECG= electrocardiogram for movement
What is noise reduction?
It is a visual inspection and artefact detection/rejection. We have to inspect the data first, is it okay to use?
We use topographic interpolation to try to reduce noise. Then we use independent component analysis (ICA) for removing eye movement artefacts and then use filtering. So, we get rid of the noise.
You can either use manual or automatic rejection.
What is the difference between automatic and manual rejection?
Manual: delete a segment from the data based on visual inspection.
Automatic: delete segment from the data if some criterion is met, for example if the amplitude from of a signal > 200uV
What is artefact?
Another word for noise.
What is the difference between liberal artifact- and conservative artifact rejection?
Liberal artifact rejection= many trials in the average, but contaminated with artifacts
Conservative artifact rejection= fewer trials in the average, though less artifacts
The more trials, the smoother the data.
What is topographic interpolation?
When one or a few electrodes are very noise throughout the experiment, replace the electrode with the mean of the surrounding electrodes.
What is ocular rejection?
If the participant keeps blinking before the stimulus, throw the data away.
What is ocular correction?
Independent component analysis gets rid of the blinks without deleting data.
ICA can be used to remove eye movement and other artefacts.
What is filtering?
It is removing high or low frequencies from the EEG signal. Why should we filter raw data? To remove non-neural voltages, such as the 50Hz AC noise(computer screen omits this noise regurarly).
What are the different types of filtering?
-High-cut/low-pass filter= remove high-frequency noise
-Low-cut/high-pass filer= remove low-frequency noise
-Band-reject/Notch filter= filters out noise between the lower and upper thresholds
Different filtering can result in different ERPs. This is important to note in the method section of a paper.
What is segmentation?
Cut the EEG into segments or epochs separately for both markers. Set a start and begin time epoch.
Average all epochs per condition.
What is a problem occuring during segmentation? And what is the solution?
Problem: the baseline is different per epoch. Cannot compare ERPs, because the baseline starts high or low.
Solution: baseline correction. Separately per electrode: compute the average baseline activity. Then, subtract the average baseline activity from each sample in the epoch.
What is a way to detect a peak?
Take the mean activity over some period. There are three results:
-When there is large lateny variation in peaks.
-When there is no peak, but more sustained activity
-When the difference between conditions lies between peaks
How do we do statistical testing?
We take an average window and export it to SPSS. We then take the peak difference between two sensors and run some analysis with it.
We can do a t-test comparing GO with NoGo at certain electrodes. But a better way is to do a Repeated Measures ANOVA with the variables Condition (Go vs NoGo) and Electrode.
What is multivariate pattern analysis (MVPA)?
It is a technique in neuro-imaging that can detect differences between conditions.
It is a machine learning technique that needs an algorythm. It needs to be fed data to make these.
We predict the type of stimuli it is, based on the EEG data we feed.
Imagine we present blue circles and red squares, how does the brain code for these two stimuli? Can we differentiate the neural responses between these two stimuli using EEG?
Solution 1: Compare ERPs of blue circles with red squares
Solution 2: MVPA, predict the labels of stimuli, based on the EEG data
How do we find the best decision boundary for seperate stimuli?
- We train the data by feeding it EEG data
- 80% of the data is getting used for data learning
- Then we test it on 20% new fresh data
What is the benefit of using MVPA compared to ERP analysis?
MVPA can extract any pattern from the data that ERP cannot and it does not have to be lateralized. It can identify differences, especially if the locus of the effect is unknown.
What is the Fourier Transform?
It is a mathematical technique used to analyze brain activity data, typically collected through methods such as EEG or fMRI.
It allows researchers to decompose complex brain signals into simpler frequency components, revealing the underlying rhythmic patterns of neural activity. This can be useful for studying cognitive processes that are associated with specific frequency bands.
By applying Fourier transformation to brain activity data, researchers can identify oscillatory patterns or brain waves that are associated with different cognitive functions. This can help understand how neural networks are organized and how they function during various cognitive tasks or states.
