Recordings of neural activity, such as EEG, are an inherent mixture of different ongoing brain processes as well as\nartefacts and are typically characterised by low signal-to-noise ratio. Moreover, EEG datasets are often inherently\nmultidimensional, comprising information in time, along different channels, subjects, trials, etc. Additional information\nmay be conveyed by expanding the signal into even more dimensions, e.g. incorporating spectral features applying\nwavelet transform. The underlying sources might show differences in each of these modes. Therefore, tensor-based\nblind source separation techniques which can extract the sources of interest from such multiway arrays, simultaneously\nexploiting the signal characteristics in all dimensions, have gained increasing interest. Canonical polyadic\ndecomposition (CPD) has been successfully used to extract epileptic seizure activity from wavelet-transformed EEG\ndata (Bioinformatics 23(13):i10ââ?¬â??i18, 2007; NeuroImage 37:844ââ?¬â??854, 2007), where each source is described by a rank-1\ntensor, i.e. by the combination of one particular temporal, spectral and spatial signature. However, in certain scenarios,\nwhere the seizure pattern is nonstationary, such a trilinear signal model is insufficient. Here, we present the application\nof a recently introduced technique, called block term decomposition (BTD) to separate EEG tensors into rank-(Lr, Lr, 1)\nterms, allowing to model more variability in the data than what would be possible with CPD. In a simulation study, we\ninvestigate the robustness of BTD against noise and different choices of model parameters. Furthermore, we show\nvarious real EEG recordings where BTD outperforms CPD in capturing complex seizure characteristics.
Loading....