Single-trial normalization for event-related spectral decomposition reduces sensitivity to noisy trials

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Classical event-related potential (ERP) models have limitations in capturing complex brain dynamics.
• Spectral techniques like event-related spectral perturbation (ERSP) are widely used but lack consensus on baseline correction.
• Current ERSP methods often average spectral estimates before baseline removal, potentially skewing results.

Purpose of the Study:

• To address limitations in current ERSP baseline correction methods.
• To introduce and evaluate novel single-trial-based ERSP baseline correction techniques.
• To improve the accuracy and robustness of spectral analysis in electroencephalography.

Main Methods:

• Comparison of classical ERSP baseline correction with single-trial correction methods.
• Development of new methods involving trial normalization or centering before baseline correction.
• Characterization of time-frequency responses and outlier robustness of proposed methods.

Main Results:

• Classical single-trial baseline correction can lead to positively skewed post-stimulus ERSP values.
• Proposed single-trial correction methods minimize artifactual data and are robust to outliers.
• New methods demonstrate improved performance and reliability in spectral analysis.

Conclusions:

• Single-trial based ERSP baseline correction offers significant advantages over classical averaging methods.
• The developed methods enhance the reliability of EEG spectral analysis for studying brain dynamics.
• These advancements provide more accurate tools for neuroscience research.