Emotion Recognition and Dynamic Functional Connectivity Analysis Based on EEG

Although emotion recognition techniques have been well developed, the understanding of the neural mechanism remains rudimentary. The traditional static network approach cannot reflect the entire brain activity at the time scale. Instead, a newly introduced temporal brain network is an optimal approach which can be used to investigate the dynamic functional connectivity (FC) of the human brain in different emotion states considering the time-varying brain regional interaction. In this study, we focused on emotion recognition and dynamic FC analysis with SEED dataset. First, multiband static networks were computed by the phase lag index (PLI). Then, subject-independent discriminative connection features of such static networks were selected to recognize the positive, neutral, and negative emotion types. In addition, we constructed the temporal brain network by sorting the static network according to time sequence. The experimental results show that the beta band is the most suitable for emotion recognition due to the best accuracy of 87.03%. And, the frontal and the temporal lobes are more sensitive to brain emotion-related activities. Moreover, we find the spatio-temporal topology of dynamic FC shows the small-world structure. Notably, the positive emotion is more distinguishable in the temporal global efficiency, especially between positive and neutral emotion states. Our findings provide new insight into the emotion-related brain regional coordination evolution and show the potential of dynamic FC for the investigation of the emotion-related brain mechanism.