A novel robust Student’s t-based Granger causality for EEG based brain network analysis

• Published in: Biomedical signal processing and control Vol. 80; p. 104321
• Main Authors: Gao, Xiaohui, Huang, Weijie, Liu, Yize, Zhang, Yinuo, Zhang, Jiamin, Li, Cunbo, Chelangat Bore, Joyce, Wang, Zhenyu, Si, Yajing, Tian, Yin, Li, Peiyang
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.02.2023
• Subjects: Brain-network Estimation; EEG; Emotion analysis; Granger Causality Analysis; Student’s t-distribution; Brain-network Estimation; Emotion analysis; Student’s t-distribution; EEG; Granger Causality Analysis
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2022.104321

•Developed a novel Granger causality inference based on Student’s t-distribution.•Quantitatively verified its robustness through both simulation study and real EEG application.•Significantly improved the performance of EEG-based directed brain networks for the recognition of emotions.•Revealed the brain-network-topology differences between various emotional states.•Discovered the lateralization differences of brain networks in emotion processing between genders. Granger-causality-based brain network analysis has been widely applied in EEG-based neuroscience researches and clinical diagnoses, such as motor imagery emotion analysis and seizure prediction. However, how to accurately estimate the causal interactions among multiple brain regions and reveal potential neural mechanisms in a reliable way is still a great challenge, due to the influence of inevitable outliers such as ocular artifacts, which may lead to the deviation of network estimation and the decoding failure of the inherent cognitive states. In this work, by introducing Student’s t-distribution into multivariate autoregressive (MVAR) model, we proposed a novel Granger causality analysis to suppress the outliers influence in directed brain network analysis. To quantitatively evaluate the performance of our proposed method, both simulation study and motor imagery EEG experiment were conducted. Through these two quantitative experiments, we verified the robustness of our proposed method to outlier influence when applying it to capture the inherent network patterns. Based on its robustness, we applied it for EEG analysis of emotions and assessed its efficiency in offering discriminative network structures for emotion recognition and discovered the biomarkers for different emotional states. These biomarkers further revealed the network-topology differences between male and female subjects when they experienced different emotional states. In general, our conducted experimental results consistently proved the robustness and efficiency of our proposed method for directed brain network analysis under complex artifact conditions, which may offer reliable evidence for network-based neurocognitive research.