System identification methods for dynamic models of brain activity

• Published in: Biomedical signal processing and control Vol. 68; p. 102765
• Main Authors: Griffith, Tristan D., Hubbard, James E.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.07.2021
• Subjects: EEG dynamics; Modal decomposition; System identification; Modal decomposition; EEG dynamics; System identification
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2021.102765

•System identification methods are explored for the analysis of EEG dynamics.•The resultant models yield an eigenmode decomposition of the emergent dynamics which capture the spatio-temporal dynamics elegantly.•The eigenmodes present as both traveling and standing waves.•This modal decomposition may be used in a biosecurity subject identification task. There is a broad need to better understand the dynamics of neural activity in both space and time. Rigorous modeling methods are needed to improve the analysis of brainwave dynamics. Two system identification algorithms, Output Only Modal Analysis (OMA) and Dynamic Mode Decomposition (DMD), are modified for use on neural dynamics and compared. An example application is included. The system identification methods are applied to estimate state space models for neural dynamics. The modeling technique results in a reduced order modal decomposition of the behavior of the brain. The resultant eigenmodes can be non-orthogonal and complex, capturing the emergent space time dynamics. We apply the modeling method to the Database for Emotion Analysis using Physiological Signals (DEAP) and the EEG Motor Movement/Imagery Dataset (EEGMMI) in a biosecurity application. It is shown that there are common modes shared among all subjects, regardless of stimuli. Further, the modal decompositions may be used to distinguish subjects from one another in a subject identification biosecurity task. The accuracy of the OMA eigenmode model is 100%, while the accuracy of the DMD eigenmode model is 96%. Output only system identification techniques are an appropriate rigorous modeling method for EEG data. The structured modeling procedure offers new opportunities for cognitive modeling and affective computing.