Table 1.
Evaluation of different methods for automated artifact rejection in the most popular open-source software packages for EEG data analysis (EEGLAB, FieldTrip, Brainstorm, and MNE). For all data rejection methods, the data is first high-pass filtered at 0.5 Hz (see “Methods” section). The percentage of significant channels is compared to raw data high-pass filtered at 0.5 Hz. For each method, we indicate the range of parameters tested. Then for each dataset, the first value indicates the optimal parameter, the second value (in parentheses) is the percentage of rejected trials (or interpolated channels for the EEGLAB clean_rawdata channel correlation method) followed by the p-value (see “Methods” section).
| Parameter values | Visual Go/No-go dataset | Face dataset | Auditory Oddball dataset | |
|---|---|---|---|---|
| EEGLAB clean_rawdata channel correlation threshold | 0.15, 0.3, 0.6, 0.7, 0.8, 0.85, 0.9, 0.95, 0.97 | – | 0.95 (7%; p < 0.0001) | 0.97 (29%; p = 0.002) |
| EEGLAB clean_rawdata ASR threshold | 5, 10, 20, 50, 100, 150, 200 | – | – | – |
| EEGLAB ICLabel probability of eye category threshold | 0.5, 0.6, 0.7, 0.8 , 0.9 | 0.5 ( p = 0.01) | – | – |
| EEGLAB ICLabel probability of muscle category threshold | 0.5, 0.6, 0.7, 0.8 , 0.9 | – | 0.6 ( p = 0.008) | |
| FieldTrip ft_artifact_zvalue LF z-score threshold | 1, 1.5, 2, 2.5, 3, 4, 5, 6 | – | – | – |
| FieldTrip ft_artifact_zvalue HF z-score threshold | 1, 1.5, 2, 2.5, 3, 4, 5, 6 | 5 (3%; p < 0.0001) | – | – |
| Brainstorm bad segments LF sensitivity threshold (1 to 5) | 1, 2, 3, 4, 5 | 5 (28%; p < 0.002) | – | – |
| Brainstorm bad segments HF sensitivity threshold (1 to 5) | 1, 2, 3, 4, 5 | – | – | – |
| Brainstorm trial thresholding method parameter | 200, 400, 600, 1000, 2000, 5000 | – | – | – |
| MNE Autoreject number of trials used for calibration | 20 trials (default) & all trials | all (12%; p < 0.005) | – | 20 (19%; p < 0.0001) |