Iterative minimization procedure with uncompressed local SAR estimate

1. A method for providing an image of a subject in a magnetic resonance imaging MRI system with parallel transmission (pTx), comprising:

a) providing a localizer scan of the subject;

b) determining body anatomy from the localizer scan;

c) matching the body anatomy with at least one body model of a plurality of body models;

d) calculating B 1 + and B 0 maps from the body anatomy;

e) using electric fields from the at least one body model and constraints to simultaneously determine a plurality of excitation pulses for a pTx performed as an optimization problem with a single objective term consisting of all time-averaged quantities including local SAR, global SAR, and average power and constraints on instantaneous quantities including flip angle inhomogeneity resulting from each pulse and instantaneous power applied to each channel, wherein local SAR matrices at all voxels are included in the objective term without compression;

f) performing MRI on the subject using the determined excitation pulses;

g) monitoring global SAR in real time to ensure all time-averaged constraints are satisfied simultaneously; and

h) providing an image from the performing the MRI on the subject and monitoring the global SAR in real time.

2. The method, as recited in claim 1 , wherein more than one body model is used for SAR estimation, performed by concatenating SAR matrices from the voxels across all body models.

3. The method, as recited in claim 1 , wherein only the magnitude of excitation is determined and the phase of excitation is not determined.

4. The method, as recited in claim 1 , wherein flip angle inhomogeneity is nonconvex with freedom in choice of spokes locations and target phase.

5. The method, as recited in claim 1 , wherein the optimization problem is decomposed into two subproblems to determine the pulses iteratively.

6. The method, as recited in claim 5 , wherein a first subproblem of the two subproblems consists of an unconstrained minimization of a piecewise positive semidefinite function.

7. The method, as recited in claim 6 , wherein a second subproblem of the two subproblems consists of a projection operation onto a nonconvex set of feasible pulses.

8. The method, as recited in claim 6 , wherein the first subproblem is solved with Kelley's cutting plane algorithm.

9. The method, as recited in claim 8 , wherein a subgradient and value of the time-averaged cost function in each iteration of Kelley's cutting plane algorithm is computed with a vectorized oracle.

10. The method, as recited in claim 9 , wherein the vectorized oracle exists on a GPU for additional acceleration.

11. The method, as recited in claim 7 , wherein the projection operation is performed in parallel for each RF pulse.

12. The method, as recited in claim 7 , wherein the nonconvex projection operation is accomplished through sequential convex projections onto ellipsoids with fixed k-space locations and target phases.

13. The method, as recited in claim 1 , wherein the step of calculating B 1 + and B 0 maps from the body anatomy is performed on a scanner while the step of matching the body anatomy with at least one body model of a plurality of body models is computed on a workstation simultaneously in parallel.

14. The method, as recited in claim 1 , wherein RF pulse design is performed in parallel with non-pTx scanning.

15. The method, as recited in claim 1 , wherein electric fields of at least one body model of the plurality of body models has been precalculated, and wherein using the electric fields, comprises using the precalculated electric fields of the at least one matched body model.