Two-way regularization for MEG source reconstruction via multilevel coordinate descent

Tian Siva Tian, Jianhua Z. Huang, Haipeng Shen

Research output: Contribution to journalArticlepeer-review

3 Scopus citations

Abstract

Magnetoencephalography (MEG) source reconstruction refers to the inverse problem of recovering the neural activity from the MEG time course measurements. A spatiotemporal two-way regularization (TWR) method was recently proposed by Tian et al. to solve this inverse problem and was shown to outperform several one-way regularization methods and spatiotemporal methods. This TWR method is a two-stage procedure that first obtains a raw estimate of the source signals and then refines the raw estimate to ensure spatial focality and temporal smoothness using spatiotemporal regularized matrix decomposition. Although proven to be effective, the performance of two-stage TWR depends on the quality of the raw estimate. In this paper we directly solve the MEG source reconstruction problem using a multivariate penalized regression where the number of variables is much larger than the number of cases. A special feature of this regression is that the regression coefficient matrix has a spatiotemporal two-way structure that naturally invites a two-way penalty. Making use of this structure, we develop a computationally efficient multilevel coordinate descent algorithm to implement the method. This new one-stage TWR method has shown its superiority to the two-stage TWR method in three simulation studies with different levels of complexity and a real-world MEG data analysis. © 2013 Wiley Periodicals, Inc., A Wiley Company.
Original languageEnglish (US)
Pages (from-to)545-556
Number of pages12
JournalStatistical Analysis and Data Mining
Volume6
Issue number6
DOIs
StatePublished - Dec 11 2013
Externally publishedYes

Fingerprint

Dive into the research topics of 'Two-way regularization for MEG source reconstruction via multilevel coordinate descent'. Together they form a unique fingerprint.

Cite this