id_1025. IMPACT OF BLINK ARTIFACT REMOVAL METHODS ON DIRECTED TRANSFER FUNCTION ESTIMATION IN EEG
Weronika Bakun1,2, Joanna Duda-Goławska2,3, Maciej Kamiński1, Jarosław Żygierewicz1
1 University of Warsaw, Faculty of Physics, Biomedical Physics Division, 5 Pasteur St., Warsaw, Poland
2 University of Warsaw, Faculty of Psychology, Cognitive Psychology and Neurocognitive Science Division, 2D Banacha St., Warsaw, Poland
3 Polish Academy of Sciences, Institute of Psychology, Neurocognitive Development Lab, 1 Jaracza St., Warsaw, Poland
INTRODUCTION: Directed Transfer Function (DTF) is a measure, used in EEG analysis, that allows the identification of causal relationships between signals in the frequency domain. It is derived from the concept of Granger causality and is based on the analysis of a multivariate autoregressive (MVAR) model fitted to the data.
AIM(S): It is crucial to use the optimal preprocessing method before calculating DTF. In particular, Independent Component Analysis (ICA), the most commonly used method for removing ocular artifacts, reconstructs EEG signals as linear combinations of independent sources. While effective for artifact suppression, such linear recomposition may alter dependency patterns relevant for DTF estimation. The present study investigates how blink-removal methods affect DTF outcomes using a controlled experimental paradigm.
METHOD(S): Participants were instructed to blink in response to auditory cues across three 15-second blocks: (1) blinking every second, (2) blinking every five seconds, and (3) no blinking. Each block was repeated ten times. Four signal conditions were analysed: (i) original no-blink signal (assumed ground truth), (ii) blink-contaminated signal without correction, (iii) blink-contaminated signal cleaned using ICA, and (iv) blink-contaminated signal cleaned using analytical modelling of the artifact. DTF matrices were computed for each condition using identical MVAR modeling hyperparameters.
RESULTS: We hypothesise that ICA-based correction may introduce artificial dependencies detectable in DTF due to its linear-mixing framework, whereas parametric modelling—being temporally localised and non-mixing — may better preserve original information flow structure.
CONCLUSIONS: Preprocessing choices are not neutral with respect to effective connectivity analysis. For studies aiming to interpret DTF in terms of neurophysiological information flow, artifact-removal methods that minimize global signal recomposition should be preferred.
FINANCIAL SUPPORT: This work was supported by the European Union under the Horizon-Widera Europe program (grant agreement No. 101159414)