Personal tools

Priors and biomedical source separation


Jump to: navigation, search

Title Priors in the biomedical source separation problems
Speaker Phlypo Ronald
Institution UNSA/CNRS
Date 2008/06/27
Time 14:30
Location Anfiteatro do IEETA
Country Portugal


The focus of this talk will be on the inclusion of priors into the framework of Blind Source Separation (BSS). Including priors allows for the permutation ambiguity inherent to the BSS-based models to be (partially) alleviated. Solving the permutation ambiguity is highly usefull for registrations in high dimensional spaces (such as EEG/MEG), since the computational load can be limited to the estimation of a component/subspace instead of the estimation of a basis for the complete space with subsequent posterior component selection. Fixing the order in which the sources are estimated is also useful in deflation based approaches, where sources are estimated one after another and the estimation error is known to propagate. We will also show that with the right choice of priors and under slightly restrictive conditions, the maxima of some contrast functions can be found algebraically.

To illustrate the use of the contrasts, clinical data will be used, including cerebral and cardiac noninvasive measurements. We will show that by choosing an appropriate contrast in function of the priors, the algorithms are able to eliminate ocular artefacts from the electro-encephalographic (EEG) recordings and to focus on the atrial activity in electro-cardiographic (ECG) recordings.