When Javier Lopez-Calderon and I designed the artifact rejection process in ERPLAB, Javier suggested that we should refer to the process of flagging epochs as artifact detection, because those epochs aren’t actually deleted from the EEG dataset. The rejection (or exclusion) of the flagged epochs actually occurs during the averaging process. I thought that was a great idea. So, I will use the phrase artifact detection to refer to the process of determining which epochs should be flagged and the phrase artifact rejection to refer to the process excluding flagged epochs from the averaged ERPs.
EEGLAB and ERPLAB also contain a separate set of routines that actually delete problematic segments of data from the continuous EEG. These routines are primarily used as a preprocessing step in the artifact correction process, as will be described in the next chapter. In the present chapter, the term artifact rejection will be used in the context of epoched EEG data.
Chapter 6 in Luck (2014) provides important theoretical background about a broad range of artifacts and about the nature of the artifact detection/rejection process. It will be helpful (but not absolutely necessary) for you to read that chapter before proceeding. The goal of the present chapter is to make this theoretical background more concrete and demonstrate the practical issues that arise in real data.
We’ll focus on data from a few example participants who I selected not because they had “good” data but because they were quite challenging. The data will come from two of the ERP CORE experiments (Kappenman et al., 2021), one looking at the mismatch negativity (MMN) and one looking at the N2pc component. The MMN paradigm was described in the previous chapter, and the N2pc paradigm will be described later in the present chapter. We’ll mainly consider blinks and eye movements, because they’re the most common large artifacts, but the exercises will also teach you general principles that you can use for other kinds of artifacts and other types of experiments.
Artifact correction has many advantages over artifact rejection, and it will be covered in the next chapter. In almost all cases, however, I recommend combining rejection and correction. Also, the problems created by artifacts are the same whether you’re using rejection or correction, so you’ll need to read at least the first part of this chapter even if you’re mainly planning to use correction instead of rejection.
Organization of the Chapter
Artifact detection is conceptually simple, but it requires a lot of decisions, and you need to know how to make the best decisions to achieve the best possible data. As a result, this chapter is pretty long. Here’s the overall structure:
- The first part of the chapter describes three main problems that are typically addressed by artifact rejection and provides an overview of the detection+rejection process.
- The second part of the chapter takes you through a series of exercises in which you’ll see how to detect and reject blinks, eye movements, and other miscellaneous artifacts in the context of the MMN experiment.
- The last part of the chapter takes you through exercises that teach you how to detect and reject small but consistent eye movements, which are especially problematic in experiments with lateralized target stimuli (especially N2pc and CDA experiments) or lateralized responses (mainly LRP experiments). If you don’t conduct experiments of this sort, you can skip this part of the chapter.
The exercises focus on data from only two participants in each experiment. I strongly recommend looking at the data from additional participants and repeating the artifact detection procedures with those participants. You can find the data from additional participants in the MMN_Data N2pc_Data folders inside the Chapter_8 folder.
Keep in mind that we did nothing to try to minimize blinking when designing and running these experiments because we knew we would use artifact correction rather than artifact rejection to deal with blinks. As a result, many participants blinked on a large proportion of trials. We would have needed to exclude many of these participants if we had rejected rather than corrected blinks.
