Real-time support vector classification and feedback of multiple emotional brain states

Area of Science:

• Affective neuroscience
• Neuroimaging
• Brain-computer interfaces

Background:

• Determining emotional states from brain activity alone is a key challenge in affective neuroscience and emotional disorder treatment.
• Previous methods lacked real-time processing and robust feature selection for emotion decoding.

Purpose of the Study:

• To investigate the feasibility of decoding discrete emotional states using functional magnetic resonance imaging (fMRI) signals.
• To develop and validate a real-time brain-computer interface (BCI) for emotion recognition.

Main Methods:

• Utilized an online support vector machine (SVM) classifier to recognize emotions from fMRI data.
• Implemented real-time head motion correction, spatial smoothing, and a novel 'Effect mapping' feature selection method.
• Performed offline whole-brain and region-of-interest classification in individuals recalling emotional episodes.

Main Results:

• Successfully decoded two discrete emotional states (happiness, disgust) and later three (happiness, disgust, sadness) with robust prediction rates.
• Subjective reports confirmed participants engaged in emotion imagery, and the classifier decoded emotions, not arbitrary brain states.
• Identified the frontal cortex as critical for emotion induction via imagery.

Conclusions:

• Real-time fMRI-BCI can decode emotional states, offering visual feedback for potential clinical applications in treating dysfunctional affect.
• The developed methods provide a foundation for objective emotion assessment through brain activity measurement.
• This technology opens new avenues for understanding and intervening in emotional processing disorders.