Validation of EEG Pre-processing Pipeline by Test-Retest Reliability.

Artifact removal and validation of pre-processing approaches remain as an open problem in EEG analysis. Cleaning data is a critical step in EEG analysis, per-formed to increase the signal-to-noise ratio and to eliminate unwanted artifacts. Methodologies commonly used for EEG pre-processing are: filtering, interpolation of bad channels, epoch segmentation, re-referencing, and elimination of physiological artifacts such as eye blinking or muscular activity. It is important to consider that the order and application of these steps affect signal quality for further analysis. In order to validate a pre-processing pipeline that can be considered in a clinical follow-up, this paper evaluated test-retest reliability of EEG recordings. EEG signals were acquired during eyes-closed resting state condition in two groups of healthy subjects with a follow-up of one and six months respectively. Signals were pre-processed with five different methodologies commonly used in literature. Test-retest reliability by intraclass correlation coefficient was calculated for power spectrum measures in each pre-processing approach and group. The results showed how test-retest reliability was significantly affected by pre-processing pipeline in both follow-ups. The pre-processing pipeline that com-bines robust reference to average and wavelet ICA improves the test-retest reliability.