1. Emotion recognition based on phase-locking value brain functional network and topological data analysis.
- Author
-
Wang, Zhong-min, Li, Sha, Zhang, Jie, and Liang, Chen
- Subjects
- *
ELECTRIC network topology , *EMOTION recognition , *LARGE-scale brain networks , *TOPOLOGICAL entropy , *MACHINE learning , *PLURALITY voting , *BODY area networks - Abstract
Traditional threshold-based methods in brain functional network analysis have some drawbacks. First, the process of determining thresholds is often based on trial and error, lacking standardization, and exhibiting strong subjectivity. Second, this subjectivity may lead to the loss of emotion-related information, limiting a comprehensive understanding and accurate identification of underlying neural processes. To overcome these problems, the persistent homology (PH) theory based on topological data analysis was introduced in this study, and a PH-based framework for emotion recognition in functional brain networks was proposed. Firstly, the EEG signals were divided into five frequency bands (δ , θ , α , β , and γ ) and segmented into multiple time series using non-overlapping sliding windows. Secondly, considering the coupling relationship between brain regions in different emotional states, the degree of phase synchronization between channels was calculated using the phase-locking value (PLV), and the PLV-based brain functional network was constructed. The PLV brain functional network was then analyzed with PH to extract multiple persistent topological features and combine them into richer feature vectors. These persistent topological features include persistence landscapes, Betti curves, persistent entropy, amplitudes, and non-diagonal points. Finally, these persistent feature vectors are used as inputs to a classifier and used for emotion recognition using machine learning algorithms and majority voting methods. Experimental analyses were conducted on the DEAP dataset to evaluate the proposed model. Further validation of the model was implemented on the DREAMER dataset and SEED dataset. The results show that the model achieves good results in EEG emotion recognition. The average accuracy reached 89.94, 87.61, and 83.24%, respectively. In this study, we extracted potential features of EEG data by applying PH to the exploration of functional brain networks, which avoided the reliance on manually determined thresholds and enabled a more comprehensive and accurate method of emotion recognition. Compared with traditional functional brain network methods, our method retains more original information related to functional brain networks in a stable and threshold-free manner. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF