Functional brain imaging: Combining EEG andfMRI using finite element and Bayesian methods

The localization of cerebral activity is a principal goal of functional brain imaging techniques such as functional Magnetic Resonance Imaging (fMRI) and Electroencephalography (EEG). FMRI detects brain activity by measuring the blood oxygenation level dependant (BOLD) effect, and EEG measures the electrical activity of the brain directly. FMRI demonstrates excellent spatial resolution (∼1mm), however its effective temporal resolution (1 ∼ 2 sec) is limited by relatively slow blood hemodynamics. In contrast, EEG can measure the brain activity in msec, but its spatial resolution remains in cm due to the lack of realistic head models and robust inverse procedures. Therefore combining fMRI and EEG premises high spatiotemporal resolution for imaging brain activity.; Most classical source localization (i.e., forward/inverse) techniques in EEG utilize over-simplified multi-layered spherical head models. However the actual human head is far more complicated due to varying thickness and electrical conductivity of different portions within the head. Also fMRI is known to be prone to artifacts caused by spatiotemporally varying structural noise components such as gross head motion, cerebro-spinal fluid pulsation, physiological fluctuations, and changes in magnetic susceptibility. The presence of these artifacts can cause negative and positive false activation, and obscure detection of true activated pixels. Thus, the reliability of the functional images can be diminished.; In this work, novel EEG forward/inverse techniques have been developed using the finite element method (FEM). Automatic construction methods of a realistic finite element head model based on MR images have been devised using the Delaunay tessellation procedure and the semi-automated MR image segmentation technique. To correlate the findings of EEG source localization with those of fMRI, noise and artifacts in fMRI are reduced by a Bayesian processing strategy developed in this study. The techniques are validated through both computer simulation and human studies. The results of Bayesian processing using human visual fMRI data demonstrate its effectiveness in reducing noise and artifacts in fMRI and enhancing the connectivity of activated pixels. The FEM-EEG simulation and human evoked motor potential studies demonstrate the feasibility of novel methods for EEG source localization suggesting a promising approach to combine fMRI and EEG.