Magnetoencephalography: Reading Your Mind (with magnetic fields)
Katharine Lee
Crofton House
Floor Location : S 049 D

Magnetoencephalography, or MEG, is a noninvasive technology used to measure brain activity. MEG measures the magnetic fields the brain emits when neurons fire. It does so with hundreds of coils set at different angles around the head that act as receptors to register minuscule magnetic fields created by neuronal currents. MEG can localize the origin of these magnetic fields to see which areas of the brain respond to different types of stimuli. MEG determines the “when” and “where” of these fields to see the amount of brain activity in response to a stimulus.

My study aims to use MEG to map brain activity patterns in the visual cortex and use those patterns to predict what a person is seeing while being scanned. A similar study was performed by Gallant et al. using fMRI and dynamic images. The issue with fMRI is the indirect method it uses to measure neuronal activity. fMRI uses BOLD (blood oxygenation level dependent) signals to examine brain activity. BOLD signals, occur at slow timescales (in seconds) limiting fMRI’s ability to reconstruct dynamic time-dependent images. For this reason, MEG, a process that is measured in milliseconds and looks directly at neuronal activity, may be more suited to reconstruct dynamic images.

Several hours worth of MEG data were collected while participants watched natural dynamic stimuli (clips of movies). Data from 3 participants were collected and analyzed individually. The data were collected in an unshielded environment using a 151 channel CTF MEG System. The open environment created a lot of noise in the data, as even a passing car outside the building could be enough to create a detectable change in the magnetic field. To eliminate the noise, I passed the data through a low pass filter of 40Hz and a high pass filter of 1Hz. I also down sampled the data to 300Hz to make processing faster as well as simplified my data to just the occipital channels (those related to visual input).

I analyzed the data in three different ways. Each had a different objective. The first method tried to achieve the initial objective: image reconstruction from patterns in the visual cortex. The second method attempted to determine if the Fusiform Face Area (FFA) could be found by comparing the movie frames with faces compared to non-face frames. This may be one of the few studies using dynamic movie frames to locate the FFA. The last method used a MATLAB self-organizing neural network algorithm to process the MEG data. The results from the second method were the most successful, resulting in the localization of the FFA. The first and third methods yielded more ambiguous results but provide valuable information for future analysis. Future data collections will be undertaken in a magnetically shielded room to reduce environmental noise. Despite the noise challenges, dynamic naturalistic stimuli was successfully used to identify the FFA using MEG.