Seizure detection algorithm based on fusion of spatio-temporal network constructed with dispersion index. (January 2023)

• Record Type: Journal Article
• Authors: Xiong, Yuhuan; Li, Jinghan; Wu, Duanpo; Dong, Fang; Liu, Junbiao; Jiang, Lurong; Cao, Jiuwen; Xu, Yuansheng
• Is Part Of: Biomedical signal processing and control. Volume 79(2023)Part 2
• Publication Date: 2023-01
• Subjects: Multivariate variational mode decomposition -- Dispersion index -- Weighted horizontal visibility graph -- Seizure detection; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.104155
• Languages: English
• ISSNs: 1746-8094

Abstract

Abstract: Epilepsy is one of the common brain disorders. Traditional seizure detection is done by electroencephalography (EEG) technicians, which takes a lot of time. Therefore, this paper proposes a new method for automatic seizure detection. In this paper, we use multivariate variational mode decomposition (MVMD) method for modal decomposition of multi-channel EEG signals and introduce dispersion index (DI) as a new brain network weight calculation method which is based on dispersion entropy (DE) to extract the features of single-channel temporal brain network and multi-channel spatial brain network for seizure detection. In single-channel temporal brain network, the network is constructed by weighted horizontal visibility graph (WHVG) method which considers the sample points as network nodes. In multi-channel spatial brain networks, DI is used to calculate the correlation between channels and select the threshold to construct the ternary-valued network of channels. We perform the validation on two publicly available datasets. The results show that the classification results for F1, AUC, ACC, PRE, SEN and SPE on CHB-MIT dataset are 97.89%, 97.81%, 97.83%, 97.56%, 98.24% and 97.39%, respectively. In addition, the results for F1, AUC, ACC, PRE, SEN and SPE on Siena scalp dataset are 99.21%, 99.19%, 99.19%, 99.28%, 99.14%, and 99.24%, respectively. The method proposed in this paper achieves good results on both datasets. In general, the joint detection of temporal and spatial networks is promising, and DI can serve as a valid indicator of correlation.