Using dynamical embedding to isolate seizure components in the ictal EEG

Abstract
A system for isolating seizure components in the epileptic electroencephalogram (EEG) is presented. The method of independent component analysis (ICA) is implemented to decompose multichannel recordings of scalp EEG known to contain epileptic seizures into their underlying independent components (ICs). The ICs are treated as a convenient expansion basis and in order to identify the relevant seizure components amongst the ICs, a series of dynamical embedding matrices are first constructed along each IC. By observing the change in structure of the singular spectra obtained by performing a singular value decomposition on each consecutive dynamical embedding matrix, it is possible to track changes in the underlying complexity of each IC with time. The change in complexity is linked to the change in entropy that can be calculated from each consecutive singular spectrum. The change in complexity, coupled with the topographical distribution of each IC, allows seizure-related components extracted by the ICA process to be subjectively identified. The method has been applied to four seizure EEG segments, and in each case probable seizure components were identified subjectively. As a proof-of-principle study, the method indicates that ICA coupled with dynamical embedding may be useful as a tool in pre-processing seizure EEG segments.