On the use of Spectrally Constrained ICA applied to single-channel Ictal EEG recordings within a Dynamical Embedding Framework

Abstract

Within a dynamical embedding (DE) framework it is possible to extract information on multiple-sources underlying just a single channel recording of electromagnetic brain activity. Independent Component Analysis (ICA) is a technique which, when used in conjunction with DE, can identify and extract statistically independent sources underlying these single channel recordings. However, these powerful techniques still generally require subjective a posteriori analysis in order to visualise neurophysiologically meaningful components in the outputs. For this reason we introduce a variant of ICA known as constrained ICA (cICA) which allows for the extraction of one of many sources underlying the measurement signal, through the provision of a basic reference signal. This constraint can be chosen to reflect neurophysiological prior knowledge of the sources in question given the measured signal. Here we present a technique which allows for the application of spectral constraints on single channel recordings of epileptic EEG data. We show that through a combination of DE and cICA it is possible to extract meaningful information on epileptic seizures and other rhythmic activity from just a single channel of EEG. We further show that accurate extraction of the sources of interest is not critically dependent on the closeness of the measurement channel to the location of the source activity.