Multi-way modelling of high-dimensionality electroencephalographic data

The aim of this study is to investigate whether useful information can be e xtracted from an electroencephalographic (EEG) data set with a very high nu mber of modes, and to determine which model is the most appropriate for thi s purpose. The data was acquired during the testing phase of a new drug exp ected to have effect on the brain activity. The implemented test program (s everal patients followed in time, different doses, conditions, etc....) led to a six-way data set. After it was confirmed that the exploratory analysi s of this data set could not be handled with classical principal component analysis (PCA), and it was verified that multidimensional structure was pre sent, multi-way methods were used to model the data. It appeared that Tucke r 3 was the most suited model. It was possible to extract useful informatio n from this high-dimensionality data. Non-relevant sources of variance (out lying patients for instance) were identified so that they can be removed be fore the in-depth physiological study is performed. (C) 2001 Elsevier Scien ce B.V. All rights reserved.