UC Berkeley

The dynamics of neural information processing are complicated, and determining the sequence and mode of activation is equally important to determining which areas activate. Magnetoencephalography (MEG) and electroencephalography (EEG) noninvasively measure the electromagnetic fields directly generated by neuronal currents in the human brain. The potential to accurately localize these signals has emerged with the advent of dense, whole-head sensor arrays. Adaptive spatial filtering techniques such as beamforming are often used reconstruct the sources of MEG/EEG activity. I have developed extensions to apply beamforming to more experimental paradigms as well as a method for more accurate validation with intracranial EEG.

Beamformers poorly resolve brain sources that are strongly correlated temporally with one another, as might be expected for an auditory experiment. I presented a method to reject the contribution of potentially interfering sources in a user-defined suppression region while allowing for source reconstruction at other specified regions. Performance of the algorithm was validated with data from simulations and an auditory MEG experiment.

Few methods exist for localizing spectral power changes with MEG. I described a novel method that uses beamformers optimized for time-frequency source reconstruction from MEG data. The performance of the method was demonstrated with simulated sources and was also applied to real MEG data from a finger movement task. Modulations in both the beta band and, importantly, the high gamma band were revealed in sensorimotor cortex and found to be statistically significant across subjects. These results were additionally validated by intracranial EEG data from two epilepsy patients. Another compelling finding was high frequency activity (30-300 Hz) in the cerebellum.

Finally, while intracranial recordings are considered the gold standard for validating noninvasive measurements, often electrode locations are not precisely known. One common method, CT-MRI coregistration, may result in a localization error of more than 10 mm. To address this, I developed a procedure to link preoperative MRIs, surgical photographs, and postimplant X-rays with projective transforms.