Ch 6: Practicalities of EEG Measurement and Experiment Design

This chapter goes over some more practical advice when designing an experiment that will use EEG headsets to collect data

6.1 Designing Experiments: Discuss, Pilot, Discuss, Pilot

It is much more important to have a well designed experiment over having perfectly formatted/low noise data. It is best to discuss experiment ideas with colleagues and do a pilot test run with 1-2 subjects, doing your intended complete set of analysis on the collected pilot data. By doing this, you can find issues with experiment design early, and correct them before you have spent the effort of going through many subjects and trials.

6.2 Event Markers

Event markets are square wave pulses that are sent to the EEG amplifiers on separate channel to denote events. properties of the pulse  (such as the amplitude) is used to denote the ID of the event. Event markers are critical to experiments as they time-lock specific event IDs to the EEG output data. It is trivial to ignore / remove excess event markers, but not trivial to "add" them back in after an experiment, so its better to err on having too many. However, many systems do not allow multiple markers to be made simultaneously, so be wary of this during the experiment design phase. Be wary of marker length (as they use square waves).

6.3 Intra- and Intertrial Timing

Due to analysis artifacts and also simply to allow the brain to return to its previous state, it is important that there is enough time between trial events (author recommends several hundred ms). 

The timing from one trial to the next trial (not events, called intertrial interval), consider the time needed for baseline normalization, including the time needed due to analysis artifacting. for ERPs, the baseline period ends at the t=0 event, but with the period of time that comes with time-frequency decomposition, you may need 200-500 ms before the event time for normalization, since temporal filtering can cause activity to leak into the time before the actual t-0 event. Generally, the author recommends an intertrial period of 1 second.

These intertrial intervals can be constant, or semi-random, with each having pros and cons. With constant intertrial intervals, the subject may start trying to "guess" the next event and mentally prepare themselves, which can impact the EEG data. While randomized intertrial periods can avoid this, users may try guessing anyways, and be 'surprised' which can again impact the data.

6.4 How Many Trials You Will Need

This depends on the signal to noise characteristics of the data collected, the size of the effect, and the type of analysis. The author recommends a minimum of 50 trials in general but it depends on various factors.

6.5 How Many Electrodes You Will Need

The number of EEG / MEG electrodes again depends on various factors, like the type of analysis. You will need more (over 100) particularly if you intend to do brain localization studies. In general, the author suggests having at least 64 electrodes if possible. There are challenges with incrementing electrode count. compute time increases with more data (but generally this isn't a massive concern). The big issues is that prep time increases too, especially with gel electrodes as each electrode needs to be prepped, thus a study with 256 electrodes vs 64 will be painfully slower to prep in between subjects for, so as a rule of practicality, go with what is needed for the experiment.

6.6 Which Sampling Rate to Use When Recording Data

The sampling rate defines the temporal resolution of the source data. Nyquists theorem states that technically one would need to sample 2 twice the desired frequency in order to be able to reconstruct it for analysis. Realistically you need to sample more to improve the signal to noise ratio. Technically however, past a few thousand Hz there is little gained from sampling further. Furthermore, sampling at higher rates needs more disk space and needs more compute time.

Its always possible to down sample data, to its better to sample on the too high end of the spectrum rather than the too low. 

The author recommends choosing easy numbers for sampling rate (divisible by 10) and recommends the general rate of 1000 Hz for most experiments (but again this varies depending on what you are looking for)

6.7 Other Optional Equipment to Consider

Other general tips and suggestions for experiments:

• Using EMG for user input allows you to see when a user was about to press the wrong button then the correct one (partial user error). Force grips are useful for consistent simple input
• Eye tracking is useful to ensure the user was looking at the correct place when awaiting an event (IE factoring out trials where users were looking out elsewhere). It can also aid in removing noise from ocular motor activity. In addition, it allows you also to look at saccades, which can be used to compare results with other studies
• EEG electrode placement is usually done with templates, which can have an error of a few cm between individuals.  While this is usually OK with many studies, for brain localization experiments you may want to use extra tools to ensure the EEGs electrodes are placed as accurately as possible.
• The user should be comfortable during the experiment (IE an comfy chair) because if they are shifting around, the motor artifacts can ruin the data
• A good response device to match the experiment is important. It should be as low latency as possible, but not be confusing with many buttons that may give user pause when responding. Buttons should be easy to press but give some sort of confirmation to the user that they were pressed (like a click)