Accelerated Computation of Regularized Estimates in Magnetic Resonance Imaging
Friday, March 21, 2014|
1:00pm - 3:00pm
Add to Google Calendar
About the Event
Magnetic resonance imaging (MRI) is a non-invasive medical imaging modality that uses electromagnetic fields. Accurate estimates of these fields are often used to improve the quality of MR imaging techniques. Regularized estimators for such fields are robust and can provide high quality estimates but often at a significant computational cost. In this work, we explore several of these estimators with a focus on developing novel minimization methods that reduce their computation times. First, we explore regularized receive coil sensitivity estimation and present several algorithms, based on augmented Lagrangian methods, that minimize the quadratic cost function in half the time required by a preconditioned conjugate gradient (CG) method. Second, we accelerate the minimization of the nonconvex cost function associated with regularized main magnetic field inhomogeneity estimation by developing two methods, both based on optimization transfer principles, that reduce the computation time by a factor of 30 compared to the existing method. Third, we present a novel alternating minimization method that uses augmented Lagrangian methods to accelerate the computation of the compressed sensing based water-fat image reconstruction problem by at least 10 times compared to the existing nonlinear CG method.
Sponsor(s): Prof. Jeffrey Fessler
Open to: Public