Functional imaging techniques have certainly changed the way neuroscientists are looking at the brain. It is now possible to produce images of the brain in action to study cognitive processes involved in e.g. perception, motor action, language and memory. Clinical research also starts to beneficiate from these advances owing to imaging studies on the origin of epilepsy seizures and the neural substrates of functional impairments due to severe neurological and psychiatric disorders.
Localization of brain areas involved is largely addressed by several powerful techniques such as functional magnetic resonance imaging (fMRI) and positon emission tomography (PET). However accessing, non invasively, the dynamics of the electrophysiology of the brain at the spatial scale of large neural cell assemblies is a major challenge. Electro and magnetoencephalography (EEG and MEG, respectively) are two recording techniques that produce, through image reconstruction techniques, sequences of brain functional images at the millisecond range. These techniques respectively measure the electric potentials and the magnetic induction produced by neural assemblies at the scalp. Because the number of sensors has increased and the technology has improved during the last decade, it is now possible to consider the joint estimation of the localization and of the time dynamics of neural activity from these scalp measurements.
Localizing the generators of electrical potentials and magnetic fields measured outside a closed volume conductor (i.e. the head) is an ill-posed inverse problem with no unique solution. However, specific constraints based on legitimate priors from neurophysiology and the individual neuroanatomy processed from structural MR images can help – once formulated in a choice of regularization approaches – both localize and estimate the dynamics of brain activity in many situations.
We will go through a number of forward and inverse modelling approaches that have been adopted by the MEG and EEG community: from analytical volume conductor models to Finite Element numerical techniques; and from linear to Bayesian estimators based on spatio-temporal Markov Random Field modelling of cortical generator patterns.
These models and inverse methods will be illustrated by a choice of experimental data ranging from validations with a real-skull phantom and simultaneous recordings with implanted electrodes in epilepsy, to various protocols in experimental psychology.
Copyright. All Rights Reserved.