FUNER: A Spatio-Temporal Penalization Approach for the MEG Inverse Problem

• Activity Number: 466
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): T. Siva Tian*+ and Zhimin Li
• Companies: University of Houston and The University of Texas Health Science Center at Houston
• Keywords: Spatio-temporal data ; Group Lasso ; Basis functions ; Inverse problem ; MEG

Abstract

The inverse problem encountered in magnetoencephalography (MEG) studies refers to reconstructing signal sources given limited scalp-recorded data. In many cases, the underlying signal sources of EEG/MEG measurements are expected to be relatively smooth in the temporal domain but relatively sparse in the spatial domain. We propose a FUNctional Expansion and Regularization (FUNER) approach that expands the signal source time courses by their temporal basese, which are then combined with the spatial information of the signal. Then the inverse problem is transformed to a linear regression problem with grouped predictors. Each group contains temporal and spatial information of one signal source. An iterative group Lasso algorithm is proposed to perform an L2-penalty within groups and an L1-penalty between groups. FUNER is shown to be effective on simulations and human MEG studies.