High-pass filtering artifacts in multivariate classification of neural time series data

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Traditional EEG/MEG data processing involves high-pass filtering and baseline correction to remove slow drifts.
• High-pass filtering can introduce temporal displacement artifacts in time-series analyses.
• The impact of these preprocessing steps on time-resolved multivariate pattern classification analyses (MVPA) remains largely unexplored.

Purpose of the Study:

• To investigate the effects of traditional preprocessing methods on MVPA.
• To introduce and evaluate a novel detrending method, trial-masked robust detrending, for EEG/MEG data.
• To determine if trial-masked robust detrending can prevent artifactual pattern displacement.

Main Methods:

• Developed trial-masked robust detrending by masking cortical events within trials.
• Applied high-pass filtering, standard robust detrending, and trial-masked robust detrending to real and simulated EEG data from a working memory experiment.
• Utilized temporal generalization analyses to assess pattern displacement.

Main Results:

• High-pass filtering and standard robust detrending, especially with baseline correction, introduced artifacts causing pattern displacement into activity-silent periods.
• These displacements were particularly evident in temporal generalization analyses.
• Trial-masked robust detrending successfully eliminated these displacements, producing artifact-free decoding.

Conclusions:

• Traditional EEG/MEG preprocessing methods like high-pass filtering and robust detrending can introduce detrimental temporal artifacts in MVPA.
• Trial-masked robust detrending offers an effective solution, preventing pattern displacement and enhancing decoding accuracy.
• The findings suggest careful consideration of preprocessing steps for robust MVPA in neuroscience research.