Final Report Summary - NIMBLE (Neuromagnetic imaging: multi-object Bayesian localisation and estimation) Non-invasive investigation of brain activity with functional imaging techniques is a highly interdisciplinary research field with a wide range of actual and potential applications from the pre-surgical evaluation of epilepsy and the conception of treatments for neurological diseases such as Alzheimer's and Parkinson's to the development of futuristic thought-controlled devices like brain computer interfaces. One of the most powerful techniques for such non-invasive investigations is magnetoencephalography (MEG) that records right out of the scalp the magnetic fields produced by the neural activity. Estimation of brain activity from MEG measurements is an ill-posed problem that requires proper use of prior information on the sources in order to be solved. The NIMBLE project was concerned with the development of a novel Bayesian method, based on sequential Monte Carlo techniques (particle filters) for the estimation of neural activity from MEG data and the implementation of this method with state-of-the-art parallel computing techniques. The project resulted in the development of a novel Bayesian model and a novel algorithm for estimation of dipolar sources from MEG data. The novel Bayesian model entails the use of static current dipoles as sources of the measured fields. The new model differs from the previous research of the Fellow in providing an explanation of the neural activity which is much closer to the neurophysiological interpretation of the current dipole model. Quantitative results obtained by analysing experimental MEG data encouragingly support the new model against older models. The novel algorithm which has been specifically devised to compute the developed statistical model, is a combination of sequential Monte Carlo and Markov Chain Monte Carlo steps, which allows estimation of partially and temporary static models in a dynamic setting, a task which is known to be highly challenging. The results of this project can be of interest for at least two different scientific communities. For the MEG community, the project has provided a novel tool for investigation of brain activity. For the statistician community, this research has provided a working example of successful application of recently developed statistical approaches, and an example of an algorithm that is able to estimate temporarily static parameters dynamically.
