Signal processing approaches to MEG data analysis: Comparison of localization methods and neural synchrony detection

Functional neuroimaging modalities improve our knowledge about normal brain function and the cause of pathological diseases. Among these Magnetoencephalography (MEG) is a very promising modality since it provides unique insights into the spatio-temporal dynamics of neural activation in the human brain. Users of MEG are faced with a vast array of inverse methods that can be used to process their data, yet there are few guidelines available to decide which might be appropriate for a given problem. Receiver Operating Characteristics (ROC) has been the most widely used for assessing system performance for lesion detection in x-ray and nuclear medicine imaging. Here we describe an objective task-based framework for comparing inverse methods using the free response receiver operating characteristic (FROC), which is the most suitable approach among the other variations of ROC. The FROC framework is adapted to functional neuroimaging so that it favors localization methods that give clinically acceptable and meaningful results. We show the application of this objective evaluation criteria by comparing the performance of a variety of localization methods used in MEG studies for known cortical activation.; One of the significant challenges for neuroscientific research is solving the mystery of the binding problem. Despite considerable efforts, the question of how the brain orchestrates the large scale integration of specialized brain regions into a functional whole remains unanswered. Over the years, linear methods have been extensively used to investigate the role of specialized brain regions in a variety of tasks. Synchronized activity of different brain regions is accepted to be an important mechanism for large scale integration and a variety of nonlinear methods have been used to investigate such interactions. Higher order statistics have found a place in EEG data analysis for detecting phase coupling between frequency bands. Although brain signals are known to be highly non-stationary, most of these studies have focused on nonparametric applications despite the penalty of low frequency resolution. This thesis introduces a novel method for parametric cross-bispectrum estimation for a higher frequency resolution in short time window EEG/MEG analysis.