Schweser, F., Deistung, A., Sommer, K., & Reichenbach, J. (2011). Superfast Dipole Inversion ( SDI ) for real-time Quantitative Susceptibility Mapping ( QSM ).
Schweser, F., A. Deistung, K. Sommer, and J. Reichenbach. “Superfast Dipole Inversion ( SDI ) for Real-Time Quantitative Susceptibility Mapping ( QSM )” (2011).
Schweser, F., et al. Superfast Dipole Inversion ( SDI ) for Real-Time Quantitative Susceptibility Mapping ( QSM ). 2011.
INTRODUCTION – Magnetic susceptibility is an intrinsic physical tissue property, which recently became accessible in vivo by quantitative susceptibility mapping (QSM) . The practical applicability of QSM, however, is currently hampered by the enormous numerical complexity and computational cost associated with the QSM reconstruction including phase unwrapping, background correction, and field-to-source inversion. Computation times between 30 minutes and several hours for a single dataset currently impede clinical routine application and hinder promising applications, such as QSM-based quantitative fMRI. In this study, we present and analyze a novel QSM framework, Superfast Dipole Inversion (SDI), that not only allows reconstruction of susceptibility maps from raw (wrapped) gradient-echo (GRE) phase data in near real-time, but also enables processing of phase images with open-ended fringe lines , which often occur in practice due to improper combination of multichannel images and which have hitherto hindered QSM. In addition, an approach is presented that allows compensating for intrinsic underestimation of susceptibilities, which has been reported by several authors (e.g. ). SDI was evaluated with respect to computation time and reconstruction accuracy based on numerical simulation and volunteer data. MATERIALS AND METHODS – Algorithm: SDI relies on an extension of the background field correction algorithm SHARP  and on a combination of SHARP with Thresholded K-Space Division (TKD)  for field-to-source inversion. Schweser et al.  have recently demonstrated that SHARP may be improved by using the Laplacian kernel instead of a solid sphere kernel during the convolution and deconvolution steps. Schofield and Zhu  have shown that the Laplacian φ 2 ∇ of the phase may be calculated directly from wrapped phase images, w φ , by using a special non-linear operator L1: w L φ φ 1 2 = ∇ . The first