EEG-fMRI Fusion: Adaptations of the Kalman Filter for Solving a High-Dimensional Spatio-Temporal Inverse Problem

Abstract

Recording the dynamics of human brain activity is a key topic for both neuroscience fundamental research and medicine. The two main techniques used, Electro- and Magneto-encephalography (EEG/MEG) on the one hand, functional Magnetic Resonance Imaging (fMRI) on the other hand, measure different aspects of this activity, and have dramatically different temporal and spatial resolutions. There is an important literature dedicated to the analysis of EEG/MEG (REF) and fMRI data (REF). Indeed, the both techniques provide partial and noisy measures of the hidden neural activity, and sophisticated methods are needed to reconstruct this activity as precisely as possible. Adaptive filtering algorithms seem well-adapted to this reconstruction, since the problem can easily be formulated as a dynamic system, but it is only recently that such formulations have been proposed for EEG analysis (Jun et al., 2005), fMRI analysis (Johnston et al., 2008; Murray & Storkey, 2008; Riera et al., 2004), or EEG-fMRI fusion (Deneux & Faugeras, 2006b; Plis et al., 2005).