Filtered deconvolution of a simulated and an in vivo phase model of the human brain

Abstract

Purpose: To remove spatial patterns in gradient echo phase images which are caused by susceptibility differences between different tissue types using filtered deconvolution and to evaluate deconvolution effects. Materials and Methods: A realistic simulated susceptibility map of the human brain was built and used to evaluate the effects of filtered deconvolution. The simulated susceptibility map was convolved with a filter kernel representing a magnetic dipole resulting in a simulated phase map. The artificial phase map was superimposed with different noise levels and deconvolved using different deconvolution kernels. The resulting contrast-to-noise ratios between white and gray matter of the deconvolved data provide an estimate for an optimal deconvolution kernel for a given noise level. These results were used to deconvolve an in vivo phase model representing the average of 30 phase data sets and also individual phase data acquired at 7 Tesla. Results: The deconvolved phase model shows a better anatomical agreement with the corresponding magnitude than the original phase model (5% higher κ coefficient). Visual inspection of the deconvolved individual phase shows a more consistent delineation of blood vessels. Conclusion: Filtered deconvolution of SWI phase is possible when an appropriate filter kernel is used. This helps to improve region of interest definition as unrealistic phase patterns are removed. © 2010 Wiley-Liss, Inc.