How about a Bayesian M/EEG imaging method correcting for incomplete spatio-temporal priors

Carsten Stahlhut, Hagai T. Attias, Kensuke Sekihara, David Wipf, Lars Kai Hansen, Srikantan S. Nagarajan

Research output: Contribution to journalConference abstract in journalResearchpeer-review

87 Downloads (Pure)

Abstract

In this contribution we present a hierarchical Bayesian model, sAquavit, to tackle the highly ill-posed problem that follows with MEG and EEG source imaging. Our model facilitates spatio-temporal patterns through the use of both spatial and temporal basis functions. While in contrast to most previous spatio-temporal inverse M/EEG models, the proposed model benefits of consisting of two source terms, namely, a spatio-temporal pattern term limiting the source configuration to a spatio-temporal subspace and a source correcting term to pick up source activity not covered by the spatio-temporal prior belief.
We have tested the model on both artificial data and real EEG data in order to demonstrate the efficacy of the model. The model was tested at different SNRs (-10.0,-5.2, -3.0, -1.0, 0, 0.8, 3.0 dB) using white noise. At all SNRs the sAquavit performs best in AUC measure, e.g. at SNR=0dB AUC is, 0.985 (sAquavit) and 0.857 (Bolstad et al., 2009).
Our results demonstrate that the sAquavit model is capable in balancing spatio-temporal prior guidance and source correction estimation to obtain superior estimates relative to current inverse methods.
Original languageEnglish
JournalJournal of Cognitive Neuroscience
VolumeSupplement
Pages (from-to)260
Number of pages1
ISSN0898-929X
Publication statusPublished - 2013
EventCognitive Neuroscience Society Annual Meeting 2013 - San Francisco, United States
Duration: 13 Apr 201316 Apr 2013

Conference

ConferenceCognitive Neuroscience Society Annual Meeting 2013
CountryUnited States
CitySan Francisco
Period13/04/201316/04/2013

Fingerprint Dive into the research topics of 'How about a Bayesian M/EEG imaging method correcting for incomplete spatio-temporal priors'. Together they form a unique fingerprint.

Cite this