N400 extraction from a few trials of EEG data using spatial and temporal-frequency pattern analysis. (6th November 2019)

• Record Type: Journal Article
• Authors: Li, Bowen; Liu, Zhiwen; Gao, Xiaorong; Lin, Yanfei
• Is Part Of: Journal of neural engineering. Volume 16:Number 6(2019:Dec.)
• Publication Date: 2019-11-06
• Subjects: event-related potential -- N400 -- spatial and temporal-frequency pattern -- dictionary learning -- low-rank regularization; Neurosciences -- Periodicals; Biomedical engineering -- Periodicals; 612.8
• DOI: 10.1088/1741-2552/ab434c
• Languages: English
• ISSNs: 1741-2560

Abstract

Abstract: Objective . N400 plays an important role in the studies of cognitive science and clinical neuropsychology diseases. However, it is still a challenge to extract the N400 component from a few trials of electroencephalogram (EEG) data. Approach . A method was proposed to analyze the spatial and temporal-frequency patterns of N400 in this study. First, resampling-average difference was used to enhance the signal-to-noise ratio (SNR) of N400 in EEG samples. Next, dictionary learning was utilized to adaptively select the wavelet bases corresponding to event-related potentials (ERPs) rather than spontaneous EEG activities and obtain the temporal-frequency patterns of ERPs. Finally, the low-rank constrained sparse decomposition was exploited to remove the spontaneous EEG activities and to learn the ERP spatial patterns, and the number of ERPs was also automatically determined. Simulation N400 datasets with different SNR levels and real N400 datasets of 15 subjects were used to evaluate the performance of the proposed method. Main results . The results indicated that the proposed method accurately extracted the N400 component from a few trials of EEG data, and a significant difference of extracted N400 waveforms was observed between two experiment conditions. Significance . In the proposed method, the resampling-average difference significantly enhanced the SNR of EEG samples. Combined with the dictionary learning, the low-rank constrained sparse decomposition effectively removed the spontaneous EEG activities and automatically selected the correct ERP components.