Detection and analysis of oscillatory patterns in multisensor data by time-frequency hidden Markov models: an application to EEG
Résumé
A model combining time-frequency representation, hidden Markov models and multichannel signal processing is developed, for the characterization of locally oscillatory phenomena in multisensor signals, such as in particular EEG and MEG signals). The model is based upon a multichannel MDCT transform of the signal; the multichannel MDCT coefficients are modelled by a hidden Markov chain, that is common to all sensors. The coefficients are distributed according to a multivariate Gaussian distribution (that depends on the hidden state), with a Kronecker-structured covariance matrix. Corresponding estimation algorithms are described and demonstrated, on both simulated and real data. The relevance of the proposed approach is demonstrated on a case study on rest EEG signals, for which it is shown that the model parameters can discriminate between control subjects and patients with multiple sclerosis.
Origine | Fichiers produits par l'(les) auteur(s) |
---|
Loading...