Tackling EEG test-retest reliability with a pre-processing pipeline
based on ICA and wavelet-ICA
Abstract
The reliability of Electroencephalography (EEG) measurements on normal
human neurophysiology can be used to determine whether changes in brain
electrical activity in subjects with neurological diseases have
potential in the diagnosis or follow-up of the patients, being more
crucial in neurodegenerative diseases where reliable measures across
time might be needed. The objective of this study is to report the
reliability of relative band powers extracted from a two-year
four-session resting-EEG longitudinal study conditioned by an automated
pipeline that leverages state of the art EEG signal-processing
approaches involving ICA, wavelet-ICA, and normalization by a
recording-specific constant. The Intraclass Correlation Coefficient
(ICC) was used as a measure of reliability. Similarly, to assess the
association between age and relative performance. The results of the ICC
for EEG data acquisition and preprocessing process showed high
significant reliability, where an average ICC of 0.91 ± 0.04 was
obtained for neural related Independent Components (ICs) and 0.92 ± 0.03
for ROIs (p-value < 5% for all data). This study shows that
after performing four EEG recording sessions for 43-subject, the
recorded measurements were replicable, and the correlation of relative
power with the age of healthy subjects is consistent with the
literature. These results suggest that relative power measured from EEGs
preprocessed with the automated pipeline is a replicable metric across
sessions, and, consequently, is useful for the study of relative power
changes caused by the progression of neurodegenerative pathologies such
as Alzheimer’s disease.