Probabilistic methods for neural source reconstruction from MEG data.

摘要

The field of brain imaging has exploded in the past two decades due to new technological developments mostly on the hardware/acquistion end. Consequently, neuroscientists have been excited to study all aspects of brain function and the general public have been excited to hear what results come from such studies. Magnetoencephalography (MEG) is a brain imaging method which passively detects the naturally occurring magnetic fields outside the head resulting from neural cells' activity in the brain. The MEG sensors record direct neural activity with millisecond resolution. However, there is inherently no unique solution to determining where exactly in the brain the neural activity was located that produced the brainwaves measured by the sensors. With certain general assumptions, a reasonable estimate of the location can be made. However, measurement noise and other artifacts, including heartbeat and eyeblinks, swamp the signals of interest, making localization nearly impossible. In this dissertation, two novel methods are proposed which improve neural source estimate by removing the effects of noise and interference. These new methods are tested against standard methods for both simulated and real data and show improved performance. These methods are further tested on real data obtained from primates with the ultimate goal of using electrophysiological data from these primates to compare the MEG localization with the true location. Finally, an example is shown of one way to combine neural activity measured by MEG with a method for measuring white-matter anatomical connections obtained by diffusion tensor imaging.