Brain States That Encode Perceived Emotion Are Reproducible but Their Classification Accuracy Is Stimulus-Dependent

Area of Science:

• Neuroscience
• Cognitive Science
• Affective Science

Background:

• The brain state hypothesis suggests a unique neural pattern for each image-induced emotion.
• Multivariate pattern analysis (MVPA) studies support this, but show inconsistent classification accuracies.
• Unaccounted variance in MVPA studies of affect processing needs investigation.

Purpose of the Study:

• To investigate inter-study variability in functional magnetic resonance imaging (fMRI)-derived affective brain states.
• To identify and account for sources of variance affecting classification accuracy.
• To validate affective brain states using an independent measure of autonomic arousal.

Main Methods:

• Demonstrated strong inter-study correlations of affective brain states across studies with different subjects and stimuli.
• Developed a method to control for the spatial distribution of perceived affective properties.
• Validated brain state predictions using skin conductance response (SCR).

Main Results:

• Inter-study correlations of affective brain states were strong despite differences in stimuli and subjects.
• Controlling for the spatial distribution of affective properties significantly improved valence (56% to 85%) and arousal (61% to 78%) classification accuracies.
• Brain states weakly predicted SCRs (r = 0.08), but prediction improved threefold (r = 0.25) for stimuli with classifiable arousal.

Conclusions:

• The spatial distribution of affective properties is a critical, previously unaccounted source of variance in MVPA studies of affect.
• Controlling for this distribution enhances the reliability and accuracy of classifying affective brain states.
• Affective brain states show predictive validity for autonomic arousal, particularly when stimulus properties are well-defined.