EEG-ERP Preprocessing#
Learning Objectives#
Be able to describe the standard steps in preprocessing EEG data for ERP analysis, including filtering, visual inspection, automated artifact detection and removal using independent components analysis, bad channel removal and interpolation, re-referencing, and averaging
Be able to describe the motivation for each of these preprocessing steps
Be able to perform each preprocessing step using MNE
Introduction#
Preprocessing is a series of signal processing steps that are performed on data prior to analysis (EDA and/or statistical analysis) and interpretation. In virtually all forms of neuroimaging data, including EEG and MEG, preprocessing is necessary in order to remove noise and obtain a clean signal of interest. In the case of EEG, the data recorded from the scalp are inevitably a mixture of signals generated from the brain (which we likely care about), and other signals generated by sources other than the brain (which we generally don’t wish to interpret).
When they first learn about preprocessing, some people think it sounds a lot like cheating — “doctoring” or “cooking” the data to engineer a specific pattern of desired results. This is not the case, however. Preprocessing steps are carefully designed and implemented based on our understanding of the source and nature of particular sources of noise, as well as our understanding about the nature of the electrical activity generated by the brain. Because the skull is such a poor conductor of electricity, the amplitude of EEG signals recorded from the scalp is very small. In contrast, the amplitude of signals from noise sources — especially from the muscles and eyes — are often an order of magnitude or more larger. Thus if we fail to remove these known sources of noise from the data, it would be unlikely that we would detect the brain activity at all, and at best our ability to do so and make valid interpretations would be greatly diminished. In other words, preprocessing is an essential step in EEG analysis.
In the series of lessons that follow, we will describe each standard step in preprocessing EEG data for ERP analysis, including why it is done, and how, using the MNE package. A script that runs a series of preprocessing steps on data is often called a pipeline; often this term is used to also include the entire process from transferring data from the EEG system to a server or other long-term storage, preprocessing, and statistical analysis. The present script is an example of a preprocessing pieline.
Sources of noise in EEG#
In EEG, sources of noise can be physiological and non-physiological.
Physiological sources of noise include muscles (especially face and neck muscles, which generate electrical potentials when contracting and relaxing), the eyes (which create distinctive signals both during blinks, and when the eyeballs move), and sometimes heartbeats (the heart produces strong, distinctive electrical signals as commonly measured with the electrocardiogram, or EKG) and breathing; the latter two sources typically are more of a problem for MEG than EEG data.
Non-physiological sources include artifacts generated when electrodes move relative to the scalp, when the electrode wires move, and ambient electromagnetic noise. Electromagnetic noise is created by virtually any device that uses electricity (except some battery-powered devices). In North America and some other parts of the world, line voltage — the alternating current that is provided through wired electrical outlets — oscillates at 60 Hz; in Europe and other parts of the world, line voltage alternates at 50 Hz. Thus depending on where the EEG data were recorded, they will likely contain a fairly distinct peak in the frequency spectrum at 50 or 60 Hz. While line noise is the most common source of electromagnetic interference, noise at other frequencies can also be produced by other equipment near the neuroimaging system (though sometimes this may be in another room or another floor of the building, and difficult to identify), or even stronger, transient sources (such as the 2-way radio of a passing truck).
We commonly refer to particular types of noise from known sources as artifacts in EEG. Thus you may see reference to eye blink artifact, line noise artifact, etc..
Best practices in EEG/MEG reporting
Especially for people who are new to any field of neuroimaging, there can be an overwhelming array of technical information relating to how the data were collected, preprocessed, and analyzed. This can make it challenging to know how, or what, to report. Fortunately, leaders in the field have reached a general consensus on this, and have published guidelines for reporting EEG and MEG study results. The most recent of these is from Pernet and colleagues (2020) and is endorsed by the Organization for Human Brain Mapping.