On spatio-temporal component selection in space-time independent component analysis: An application to ictal EEG

• Published in: 2009 Annual International Conference of the IEEE Engineering in Medicine and Biology Society Vol. 2009; pp. 3154 - 3157
• Main Authors: James, C.J., Demanuele, C.
• Format: Conference Proceeding Journal Article
• Language: English
• Published: United States IEEE 01.01.2009
• Subjects: Algorithms; Artificial Intelligence; Biomedical Engineering - methods; Biomedical measurements; Brain - pathology; Brain Mapping; Covariance matrix; Data mining; Electroencephalography; Electroencephalography - instrumentation; Electroencephalography - methods; Epilepsy; Epilepsy - diagnosis; Epilepsy - physiopathology; Humans; Independent component analysis; Pattern Recognition, Automated - methods; Process control; Reproducibility of Results; Scalp; Signal processing; Signal Processing, Computer-Assisted; Source separation; Time Factors
• ISSN: 1094-687X; 1557-170X; 1558-4615
• DOI: 10.1109/IEMBS.2009.5334034

This paper assesses the use of independent component analysis (ICA) as applied to epileptic scalp electroencephalographic (EEG) recordings. In particular we address the newly introduced spatio-temporal ICA algorithm (ST-ICA), which uses both spatial and temporal information derived from multi-channel biomedical signal recordings to inform (or update) the standard ICA algorithm. ICA is a technique well suited to extracting underlying sources from multi-channel EEG recordings - for ictal EEG recordings, the goal is to both de-noise the EEG recordings (i.e. remove artifacts) as well as isolate and extract epileptic processes. As part of any ICA application, there is an interim stage whereby relevant components (or processes) need to be identified - either objectively or subjectively (usually the latter). In previous work with ST-ICA we used spectral information alone to identify the underlying processes subspaces extracted by the ST-ICA. Here we assess the joint use of spatial as well as spectral information for this purpose. We test this on ictal EEG segments where it can be seen that different underlying processes possess characteristic signatures in both modalities which can be utilized for the clustering (or process selection) stage.