High spatial resolution distributed current source estimation using magnetoencephalography (MEG)

• Main Authors: Wei-Tang Chang, 張偉棠
• Other Authors: 林發暄
• Format: Others
• Language: en_US
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/64035305430711690476

博士 === 國立臺灣大學 === 醫學工程學研究所 === 98 === Magnetoencephalography (MEG) enables non-invasive detection of weak cerebral magnetic fields by utilizing super-conducting quantum interference devices (SQUIDs). Solving the MEG inverse problem requires reconstructing the strength, locations, and orientations of the underlying neuronal current sources based on the extracranial measurements. Due to the ill-posed nature of the MEG inverse problem, the neuronal current reconstruction is not unique unless additional constraints are imposed. The main theme of this thesis is to improve the accuracy of source reconstruction by imposing specific constraints. This thesis includes two parts: (i) sparse current source estimation using loose orientation constraint, and (ii) spatially sparse source cluster modeling by Compressive Neuromagnetic Tomography (CENT). The assumption of a spatially focal distribution of underlying neuronal current sources can be incorporated into MEG source modeling. One method is the minimum-current estimate (MCE), which achieves spatially focal source estimates by imposing a minimum l1-norm constraint on the distribution of the current sources. The knowledge about the orientations of current sources in MCE can be first estimated by the minimum-norm estimate (MNE) with a loose orientation constraint (LOC). However, MCE is spatially unstable because the l1-norm minimization is sensitive to noise and any errors due to, for example, inaccurate source orientations estimated by MNE with the LOC. Inspired by the MCE minimizing the l1-norm of the source distribution, the first part of this thesis presents a minimum l1-norm estimation source modeling approach with loose orientation constraint l1LOC), which integrates the estimation of current source orientation, location, and strength into a cost function to jointly model the residual error and the -norm of the estimated sources. In simulations and MEG experiments, l1LOC can estimate sources with a smaller spatial extent and can achieve a higher localization accuracy. In the second part of this thesis, we develop a distributed source modeling technique without the assumption on the spatial distribution of sources to be either focal or diffussive. Most inverse problem solvers explicitly favor either spatially focal or diffussive current source patterns. Naturally, in a situation where both focal and spatially extended sources are present, such reconstruction methods may yield inaccurate current estimates. To address this problem, we develop the ComprEssive Neuromagnetic Tomography (CENT) method based on the assumption that the current sources are “compressible”. By combining two complementary constraints of standard and transformed domain sparsity, we obtain source estimates not only locally smooth and regular but also forming globally separable clusters. We study the Laplacian matrix (CENTL) and spherical wavelets (CENTW) as alternatives for the transformation in the compression constraint. For simulated sources of focal, diffuse, or the combination of these two types, the CENT method shows better accuracy on estimating the source locations and spatial extents than the minimum l(1)-norm or minimum l(2)-norm constrained inverse solutions. Different transformations yield different benefits: By utilizing CENT with the Laplacian matrix it is possible to suppress activations extending across two opposite banks of a sulcus. With the spherical wavelet transform, CENT can improve the detection of two nearby yet not directly connected sources. Overall, the CENT method is demonstrated to be a promising tool for adaptive modeling of distributed neuronal currents associated with cognitive tasks.