The inverse mapping problem is well-studied in magnetoencephalography (MEG) domain, where measurements recorded from a small number of sensors (100-300) are used to infer the currents in a much higher dimensional brain space (1,000-50,000 vertices). The driving theme is that the patterns observed in neuronal signal responses, when presented with various stimuli, can give insight into the functional mapping of human brain. While inverse mapping algorithms are critical in arriving at the correct estimates of electrical currents in brain space, functional mapping of brain requires further analysis of these inferred signals. In this paper, we assume that state of the art methods for inverse mapping provide a reasonable solution in the brain space, and use the measurements in the brain space for taking the next step toward functional analysis of brain. We propose a set of analytical tools to characterize and analyze event related neuronal activity measured in brain via MEG. We propose a joint entropy minimization based formulation that makes no assumptions on the underlying canonical response at a given measurement point in the brain for a given stimulus. We model the neuro-electrical activity for a given cognitive task as an ARMA process. We then compare the event related neuronal activities using a distance measure in the space of dynamical models that takes into account the activities from multiple measurement points in the brain simultaneously. This approach can handle individual variations across subjects, can model and handle various measurement noises and phase off-set in the neuronal activities. We evaluated our method on a dataset of MEG activity obtained during an anticipatory attention task conducted by the Dynamic Neuroimaging Lab at UCSF. Preliminary results show that this framework offers low computational complexity while providing excellent classification performance.

Keywords: Magnetoencephalography (MEG), Comparative Analysis and Charaterization of event related neuronal activity, Functional brain mapping, Clustering, Time Series Modeling, Classification, Linear Dynamical Models, Mutual Information, Entropy Minimization




Download Full History