Effects of EEG Preprocessing Methods on Modified Diffusion Entropy Analysis and Complexity Synchronization

Author(s):

Sabrina Sullivan, Scott E. Kerick, and J. Cortney Bradford

Abstract:

Electroencephalogram (EEG)-derived complexity metrics offer insight into the brain’s capacity for adaptive information processing; however, the influence of artifact removal on these estimates remains poorly understood. This study investigated how EEG preprocessing and artifact removal affect estimates of neural complexity and complexity synchronization (CS), as quantified by modified diffusion entropy analysis (MDEA), a method grounded in crucial event theory. Two EEG datasets were analyzed: one dominated by persistent artifacts from treadmill walking with exoskeleton assistance, and one dominated by transient artifacts from a seated high- and low-stress shooting task. MDEA was applied to the datasets after each of the four preprocessing stages: 1) robust re-referencing with high-pass filtering, 2) artifact subspace reconstruction, 3) canonical correlation analysis, and 4) adaptive mixture independent component analysis (ICA) with IC pruning. In both persistent and transient artifact datasets, mean and standard deviation of complexity estimates decreased from minimally to maximally processed data; however, temporal fluctuations in complexity were largely preserved across all preprocessing stages. The relationship between conditions remained intact throughout the pipeline for the persistent-artifact dataset (pre- vs. post-adaptation). The CS between EEG and electrocardiogram was consistently significant during low- and high-stress conditions but it was consistently greater during low stress. These results indicate that although artifact removal systematically lowers the magnitude of complexity estimates, the temporal structure underlying neural complexity is largely robust to artifact contamination, suggesting that fewer preprocessing steps may suffice for some applications and offer practical benefits for computationally efficient, real-time neural monitoring.