An Approach to Dynamic Modelling and Topographic Feature Extraction of Wake EEG
We consider advanced pattern processing approaches for extracting information from biomedical signals. Signals such as the electroencephalogram are characterised by nonstationarity, a poor signal-to-noise ratio, and a signal which is composed out of many subcomponents. In an application domain such as investigating ‘vigilance’ we have to develop techniques which are capable of addressing some of these problems. In this paper we develop a singular subspace and an independent components dynamical systems embedding of single channel EEG time series which are then analysed for data clustering by a topographic neural network approach known as NeuroScale.
The work is novel in several respects: the use of unsupervised data (most work in this area is through evoked potential response experiments), the treatment of EEG by short window embeddings directly from the time domain(most work averages 30 second segments of power spectra), the comparison of ICA and PCA embeddings on single channel data (most ICA work so far reported has employed multiple sensor channels) and the use of a ‘NeuroScale’ architecture for topographic feature extraction.
KeywordsFeature Space Finite Impulse Response Singular Vector Independent Component Analysis Radial Basis Function Network
Unable to display preview. Download preview PDF.
- Nunez PL. Neocortical Dynamics and Human EEG Rhythms. Oxford University Press, NY., 1995.Google Scholar
- Roberts S and Tarassenko L. The Analysis of the Sleep EEG using a Multilayer Network with Spatial Organisation. IEE Proceedings F:, 139(6):420–425, 1992.Google Scholar
- Siegwart DK, Tarassenko L, Roberts SJ, Stradling JR, and Partlett J. Sleep Apnoea from Neural Network Post Processing. In 4th IEE International Conference on Artificial Neural Networks, 1995.Google Scholar
- Makeig S and Jung T-P. Changes in Alertness are a Principal Component of Variance in the EEG spectrum. Technical Report, Naval Health Research Centre, 1995.Google Scholar
- Makeig S, Jung T-P, and Sejnowski T. Using Feedforward Neural Networks to Monitor Alertness from Changes in EEG Correlation and Coherence. In Advances in Neural Information Processing Systems 8, 1996.Google Scholar
- Lowe D. Embedding the EEG. In SPIE’s Aerospace/Defence Sensing and Controls Conference: Applications and Science of Artificial Neural Networks III, 1997.Google Scholar
- Lowe D. Characterising Complexity by the Degrees of Freedom in a Radial Basis Function Network. NeuroComputing Journal, Special Issue on Radial Basis Functions, 1997.Google Scholar
- Lowe D and Hazarika N. Complexity modelling and stability characterisation for long term iterated time series prediction. In 5th IEE International Conference on Artificial Neural Networks, volume Conference Publication number 440, pages 53–58. The Institute of Electrical Engineers, 1997.CrossRefGoogle Scholar
- Lowe D and Tipping ME. Neuroscale: Novel topographic feature extraction with radial basis function networks. In M.C. Mozer, M.I. Jordan, and T. Petsche, editors, Advances in Neural Information Processing Systems 9, pages 543–549. Cambridge, Mass: MIT Press, 1997.Google Scholar