A Comparative Study of Minimum Norm Inverse Methods for MEG Imaging
The majority of MEG imaging techniques currently in use fall into the general class of (weighted) minimum norm methods. The minimization of a norm is used as the basis for choosing one from a generally infinite set of solutions that provide an equally good fit to the data. This ambiguity in the Solution arises from the inherent non-unique-ness of the continuous inverse problem and is compounded by the imbalance between the relatively small number of measurements and the large number of source voxels. Here we present a unified view of the minimum norm methods and describe how we can use Tikhonov regularization to avoid instabilities in the solutions due to noise. We then compare the Performance of regularized versions of three well known linear minimum norm methods [51  with the non-linear iteratively reweighted minimum norm method [11 and a Bayesian approach described in our companion paper (“MEG-based Imaging of Focal Neuronal Current Sources,” Phillips J.W., Leahy R.M., Mosher J.C.).
KeywordsMinimum Norm Tikhonov Regularization Bayesian Technique Phantom Data Magnetic Source Image
Unable to display preview. Download preview PDF.
- M. Hämäläinen, R. Hari, et al., “Magnetoencephalography - theory, Instrumentation, and applications to noninva sive studies of the working human brain,” Rev Mod Phys, vol. 65(2), pp. 413–497,1993.Google Scholar
- P.C. Hansen, Regularization Tools A Matlab Package for Analysis and Solution of Discrete lll-Posed Problems, available in Postscript from http://www.mathworks.com/, March 1993
- R.D. Pascual-Marqui, C.M. Michel, D. Lehmann, “Low resolution electromagnetic tomography: A new method for localizing electrical activity in the brain,” Int. Journ. ofPsyc., vol. 18, pp. 49–65, 1994Google Scholar