Exploring EEG microstates for affective computing: decoding valence and arousal experiences during video watching^.

Area of Science:

• Neuroscience
• Affective Computing
• Computational Neuroscience

Background:

• Investigating electroencephalography (EEG) correlates of human emotional experiences is crucial in affective computing.
• Previous research primarily utilized EEG features from localized brain activity.
• A need exists for novel EEG features that capture global brain dynamics related to emotion.

Purpose of the Study:

• To explore brain network-based features derived from EEG microstates for representing emotional experiences.
• To compare the performance of EEG microstate features against conventional spectral power features in decoding emotional valence and arousal.
• To assess the potential of EEG microstates as a promising feature type for affective computing applications.

Main Methods:

• Utilized the publicly available DEAP dataset comprising 32-channel EEG recordings from 32 participants watching music videos.
• Extracted four quasi-stable prototypical EEG microstates and their temporal parameters.
• Employed random forest regression to decode emotional valence and arousal using microstate features and spectral power features.

Main Results:

• EEG microstate features demonstrated superior performance in decoding emotional valence compared to spectral power features (MSE = 3.85±0.28 vs. 4.07 ± 0.30, p = 0.022).
• Microstate features showed comparable performance to spectral power features in decoding emotional arousal (MSE = 3.30±0.30 vs. 3.41 ±0.31, p = 0.169).
• Microstate features offer a global spatiotemporal dynamical perspective of neural activity.

Conclusions:

• EEG microstate temporal parameters represent a promising feature for affective computing.
• Findings suggest a potential new mechanism for understanding human emotion from a network perspective.
• This study highlights the utility of EEG microstates for advancing emotion recognition technologies.