Combination of PCA and undecimated wavelet transform for neural data processing

EEG based emotion recognition using minimum spanning tree

Emotion is a fundamental factor that influences human cognition, motivation, decision making and social interactions. This psychological state arises spontaneously and goes with physiological changes that can be recognized by computational methods. In this study, changes in minimum spanning tree (MST) structure of brain functional connectome were used for emotion classification based on EEG data and the obtained results were employed for interpretation about the most informative frequency content of emotional states. For estimation of interaction between different brain regions, several connectivity metrics were applied and interactions were calculated in different frequency bands. Subsequently, the MST graph was extracted from the functional connectivity matrix and its features were used for emotion recognition. The results showed that the accuracy of the proposed method for separating emotions with different arousal levels was 88.28%, while for different valence levels it was 81.25%. Interestingly, the system performance for binary classification of emotions based on quadrants of arousal-valence space was also higher than 80%. The MST approach allowed us to study the change of brain complexity and dynamics in various emotional states. This capability provided us enough knowledge to claim lower-alpha and gamma bands contain the main information for discrimination of emotional states.