Bayesian Denoising and Source Localization for MEG Data

University Of California, San Francisco, San Francisco CA

Investigators

Linked publications & trials

Abstract

DESCRIPTION (provided by applicant): The ability to pinpoint the three-dimensional location of the neural generators involved in various tasks is paramount to many neurophysiological studies, such as gaining a correct understanding of the manifold processes of human cognition, and for preoperative assistance in epileptic foci localization. The broad aim of the proposal is to apply novel statistical signal processing tools to raw MEG data in order to produce more accurate localizations. The proposed scheme consists of two approaches, the first of which concerns taking advantage of the statistical relationships among neural signals in order to reduce the inherent noise and interference in raw MEG data. In this approach the resulting denoised signals are localized using one of the standard algorithms. The second approach involves combining the localization and the denoising algorithm into a single functional unit. While it has recently become popular to use Independent Component Analysis (ICA) for denoising, there are several drawbacks to this approach, e.g., model order selection and the determination of which components correspond to signals of interest and which are interference. The proposed method uses a novel Bayesian inference formulation that is not hindered by these deficiencies.