Discriminating Multiple Emotional States from EEG Using a Data-Adaptive, Multiscale Information-Theoretic Approach.

Area of Science:

• Neuroscience
• Signal Processing
• Affective Computing

Background:

• Electroencephalography (EEG) is a key tool for studying brain activity.
• Accurately classifying emotional states from EEG remains challenging.
• Traditional methods often rely on predefined frequency bands, potentially missing complex signal dynamics.

Purpose of the Study:

• To investigate the efficacy of multivariate sample entropy applied to EEG data for discriminating between multiple self-reported emotional states.
• To explore a data-driven, multiscale approach for analyzing EEG complexity.
• To compare this novel method against traditional EEG analysis techniques.

Main Methods:

• EEG data were collected from 30 participants viewing emotion-inducing video clips.
• Multivariate sample entropy was estimated using multivariate empirical mode decomposition (MEMD) across multiple data-driven scales.
• EEG data were also analyzed using traditional power spectral densities and hemispheric asymmetries across standard frequency bands (theta, alpha, beta, gamma).

Main Results:

• Multivariate, multiscale sample entropy significantly discriminated between five self-reported emotions (p < 0.05).
• This discrimination was not achieved using arousal scores or traditional EEG frequency band analyses.
• The complexity metric derived from MEMD showed superior performance in emotion classification.

Conclusions:

• Multivariate, multiscale sample entropy is a powerful and promising technique for objective emotion recognition from EEG.
• This data-driven approach overcomes limitations of predefined frequency bands in EEG analysis.
• The findings suggest a new avenue for developing more accurate brain-computer interfaces for affective states.