Characterising the dynamics of EEG waveforms as the path through parameter space of a neural mass model: Application to epilepsy seizure evolution

• Published in: NeuroImage (Orlando, Fla.) Vol. 59; no. 3; pp. 2374 - 2392
• Main Authors: Nevado-Holgado, Alejo J., Marten, Frank, Richardson, Mark P., Terry, John R.
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.02.2012; Elsevier Limited
• Subjects: Algorithms; Axons - physiology; Bifurcation analysis; Brain - physiopathology; Cerebral Cortex - physiopathology; Cluster Analysis; Computer Simulation; Data Interpretation, Statistical; Databases, Factual; Electroencephalography; Electroencephalography - statistics & numerical data; Epilepsy; Epilepsy - physiopathology; Epilepsy, Generalized - physiopathology; Genetics - statistics & numerical data; Humans; Models, Neurological; Models, Statistical; Mortality; Multi-objective genetic algorithm; Neural mass model; Neural Pathways - physiopathology; Nonlinear dynamics; Nonlinear parameter estimation; Receptors, GABA-B - physiology; Reproducibility of Results; Signal Processing, Computer-Assisted; Studies; Synapses - physiology; Thalamus - physiopathology; Time-domain estimation; Time-domain estimation; Nonlinear parameter estimation; Multi-objective genetic algorithm; Nonlinear dynamics; Neural mass model; Bifurcation analysis
• ISSN: 1053-8119; 1095-9572; 1095-9572
• DOI: 10.1016/j.neuroimage.2011.08.111

In this paper we propose that the dynamic evolution of EEG activity during epileptic seizures may be characterised as a path through parameter space of a neural mass model, reflecting gradual changes in underlying physiological mechanisms. Previous theoretical studies have shown how boundaries in parameter space of the model (so-called bifurcations) correspond to transitions in EEG waveforms between apparently normal, spike and wave and subsequently poly-spike and wave activity. In the present manuscript, we develop a multi-objective genetic algorithm that can estimate parameters of an underlying model from clinical data recordings. A standard approach to this problem is to transform both clinical data and model output into the frequency domain and then choose parameters that minimise the difference in their respective power spectra. Instead in the present manuscript, we estimate parameters in the time domain, their choice being determined according to the best fit obtained between the model output and specific features of the observed EEG waveform. This results in an approximate path through the bifurcation plane of the model obtained from clinical data. We present comparisons of such paths through parameter space from separate seizures from an individual subject, as well as between different subjects. Differences in the path reflect subtleties of variation in the dynamics of EEG, which at present appear indistinguishable using standard clinical techniques. [Display omitted] ► We develop a novel method for estimation of model parameters from EEG. ► The framework can usefully be applied in a wide variety of macroscale brain research. ► The outcome helps to characterise the dynamic transitions observed in EEG. ► Evolution of model parameters over time may inform clinical decisions.