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

Abstract

An important question that confronts current research in affective neuroscience as well as in the treatment of emotional disorders is whether it is possible to determine the emotional state of a person based on the measurement of brain activity alone. Here, we first show that an online support vector machine (SVM) can be built to recognize two discrete emotional states, such as happiness and disgust from fMRI signals, in healthy individuals instructed to recall emotionally salient episodes from their lives. We report the first application of real-time head motion correction, spatial smoothing and feature selection based on a new method called Effect mapping. The classifier also showed robust prediction rates in decoding three discrete emotional states (happiness, disgust and sadness) in an extended group of participants. Subjective reports ascertained that participants performed emotion imagery and that the online classifier decoded emotions and not arbitrary states of the brain. Offline whole brain classification as well as region-of-interest classification in 24 brain areas previously implicated in emotion processing revealed that the frontal cortex was critically involved in emotion induction by imagery. We also demonstrate an fMRI-BCI based on real-time classification of BOLD signals from multiple brain regions, for each repetition time (TR) of scanning, providing visual feedback of emotional states to the participant for potential applications in the clinical treatment of dysfunctional affect.