Meeting Banner
Abstract #4994

Eliminating Streaking Artifacts in Quantitative Susceptibility Mapping Using L1 Norm Minimization

Ildar Khalidov1, Tian Liu1, Xiaoyue Chen2, Moonsoo Jin2, Martin Prince1, Yi Wang1

1Radiology, Weill Cornell Medical College, NYC, NY, United States; 2Biomedical Engineering, Cornell University, Ithaca, NY, United States


Quantitative susceptibility mapping (QSM) has been developed as a technique that uses the phase information from the MRI measurements to estimate susceptibility changes in the imaged object. Moreover, it is possible to estimate the magnetic moment of the region of interest, which gives way to quantitative imaging of tracer particles in MRI. However, the inverse problem that needs to be solved to recover the susceptibility map from the phase image is ill-posed: 1), the dipole kernel that links the two maps has a cone of zeros in Fourier domain, and 2), regions of strong susceptibility change have low intensity (and hence, unreliable phase data) due to T2* dephasing. In this work, we use total variation-based regularization to tackle the inverse problem. Compared to original weighted quadratic regularization in [1], the proposed TV regularization significantly reduces the streaking artifacts from the areas of susceptibility change. This is particularly important in animal imaging where eventual air bubbles and/or voxel misclassifications at the segmentation stage could lead to strong under-estimation of the quantities of particles of interest.