NeuroAnalyzer tutorials: Introduction to EEG experiments

using NeuroAnalyzer;

Workflow:

1. review literature: has it been done before / in the light of the literature, should it be done another way
2. plan, plan, plan
3. consider EEG-specific characteristics (e.g. if experiment requires movements \(\rightarrow\) artifacts) - e.g. consider chin rest
4. multiple repetition of the test condition (> 30, preferably > 100 times)
5. plan simple experiments (1 hypothesis at a time), simpler is better
6. do power calculation
7. consider study pre-registration
8. check stimulus presentation (e.g. monitor refresh rate 60 Hz gives 1 frame every 16.7 ms)
9. do at least 1 pilot participant
10. keep the experiment room cooled (higher temperature degrades signal)
11. make notes when collecting data
12. organize data (BIDS)
13. do not modify the experiment after it starts

Approaches

Approaches:

1. exploratory
2. hypothesis driven

Exploratory:

• permutation testing, apply multiple correction
• don’t worry about assumptions
• lower sensitivity to small effects
• mostly for 2-tests and correlations

Hypothesis driven:

• amenable to factorial designs
• multiple correction not required
• hypothesis driven
• can miss important, but unexpected results

Designs

Designs:

1. event-related: series of short duration responses occurring at a fixed latency relative to task-related events such as the onset or offset of sensory stimuli or behavioral responses
2. block-based (boxcar design)

Event-related experiment is structured around a series of experimental events (such as the presentation of stimuli).

Group- vs subject-effects

Subject level:

• variation over trials
• no population generalization
• requires large effect size

Group level:

• consistency of effects across groups
• generalize to population
• sensitive to small effects sizes

Event-related experiments

Time periods:

1. a fixation period (to be used as a baseline; see below)
2. the presentation of a stimulus (here represented by two conditions corresponding to two different facial emotions)
3. response period (where the subject is required to make a behavioral response to each stimulus)

Brain signals which are variable in their timing to experimental events usually cannot be measured using event-related designs.

Button presses and other motor responses will generate their own event-related responses, not only at the time of the behavioral response itself but also beforehand due to preparatory processes.

Block-based experiments

Alternatively known as a boxcar design.

Suitable for brain responses sustained continuously in response to a continuous experimental task (for instance to continuous sensory stimulation or continuous performance of a cognitive or behavioral task).

Event triggers are used to signal the onset of each block, so that data can be averaged across blocks.

Different conditions are separated into blocks, and brain signals are quantified by averaging data across blocks of the same condition rather than across a series of individual events (see event-related design).

Data can be averaged across blocks; BUT collapsing data across time within a block removes information about the timing of brain responses.

Measurement of brain responses in block designs is often based on comparison with a baseline time period during which no task-related responses are present (e.g. rest blocks or an inter-block interval at the start of each block that can then be used as a baseline for the corresponding block).

Event-related response can then be measured as a sustained response at the frequency of repetition (or at integer multiples of that frequency); e.g. presenting a sensory stimulus four times a second should generate four event-related responses a second which would be equivalent to a 4 Hz continuous response = steady-state evoked response.

Frequency tagging: different stimuli (or parts of a single stimulus) are modulated at different frequencies, and the brain responses to each stimulus can separated be measuring the signal at each corresponding frequency.

Triggers

Triggers are transmitted using transistor-transistor logic (TTL) which allows the transmission of different integer values; 0: no trigger, other values: trigger value.

Triggers can also be transmitted by assigning values to certain states of an analog input signal from a peripheral device or based on a physiological measure.

Triggers are usually transmitted as short duration pulses, so that the timing of the pulse can represent the timing of the corresponding event.

Triggers may be delayed relating to stimuli (transmission, registering, software processing) and may need to be time-shifted.

Estimating number of trials

Low frequencies have higher SNR = less trials required.

How to estimate number of trials:

• get one trial
• get one time-frequency band power of this trial
• correlate it with the mean of all time-frequency bands power of all trials (use Spearman correlation as power data is not normally distributed, no negative power values, highly skewed)
• do the same for the mean of two trials, then three, etc. up to all trials
• do the same for all frequencies
• draw a plot: