Overcoming Limitations of Virtual Observation Points in pTx using IMPULSE
Mihir Pendse1 and Brian K Rutt1

1Stanford University, Stanford, CA, United States


We analyze the performance of the IMPULSE pTx optimization algorithm, which allows SAR-aware pulse design without virtual observation points (VOP) compression. We compare performance of IMPULSE with conventional optimization methods using VOPs and compare different values of the overestimation parameter. We show that IMPULSE results in elimination of the time-intensive compression step without significantly increasing the time for real-time optimization. Additionally by eliminating the overestimation error from the VOP compression, IMPULSE is able to achieve better mitigation of local SAR hotspots after optimization than VOP-based methods.


Controlling local SAR in parallel transmission (pTx) is an important safety concern. When designing pTx pulses to counteract B1+ inhomogeneity, local SAR must be incorporated into the optimization to limit the generation of hotspots. The pulse design must also be done rapidly while the patient is on the table, using B1+ maps measured from the patient. Since local SAR must be evaluated over the entire patient volume, in the form of SAR matrices at each of possibly more than 100000 voxels, incorporating SAR into the pTx pulse optimization can result in impractically long optimization times. For this reason, the Virtual Observation Points (VOP) [1] concept was developed as a way of compressing the large number of SAR matrices into a smaller set of VOP matrices, and thereby speeding up the pulse optimization step. The first drawback of this approach is that the VOP compression step itself is computationally demanding, requiring several minutes of computation time, and therefore must be done prior to the patient scan; this is typically accomplished by calculating SAR matrices using electromagnetic simulation software and a realistic body model, followed by VOP compression. In addition to requiring substantial computation time, making “on-the-table” patient-specific SAR-aware pTx impractical to date, the use of VOPs introduces an error into the SAR estimation which can result in a suboptimal pTx pulse. Here we show the quantitative benefits of the IMPULSE pTx optimization algorithm [2] which operates on uncompressed SAR matrices, thereby not requiring VOPs . We demonstrate that IMPULSE enables pTx optimization on patient-specific uncompressed SAR estimates and achieves better elimination of SAR hotspots.


IMPULSE finds the pTx channel weightings and spokes locations that minimize SAR subject to a specified tolerance on flip angle inhomogeneity (FAI) [1]. Here, we compared this highly efficient SAR-aware pTx optimization algorithm against a conventional VOP-based algorithm [3]. A key feature of IMPULSE is that it scales favorably with number of SAR matrices and therefore enables optimization without the VOP compression step. Two sets of VOP matrices were generated, with overestimation parameter values of 0.025 and 0.15, respectively. The benefits of IMPULSE over the VOP-based designs were assessed based on both computation time and resulting SAR performance. B1+ and electric field distributions were generated using Sim4Life FDTD simulation package (Zurich MedTech AG, Zurich) employing the Duke body model from the Virtual Family (IT’IS Foundation, Zurich) and an 8 channel coil model. Pulses were designed to excite an axial slice, and L-curves of SAR vs FAI were generated for each method using MATLAB (MathWorks, Natick, MA) by varying the FAI tolerance.


Figure 1 shows the L-curves for the two algorithms, with two overestimation parameters for the VOP-based algorithm, demonstrating that there is a SAR penalty associated with the use of VOPs. With a strict overestimation parameter (0.025), there is a penalty of approximately 6% compared to using uncompressed matrices. Although this SAR penalty is small, the computation time for VOP compression is significant in this case (620 s). For more relaxed overestimation parameter (0.15) , the compression time is decreased to 40s, making on-the-table patient-specific pTx pulse design feasible; however, the SAR penalty compared to IMPULSE increases to approximately 27%. Figure 2 shows representative SAR maps (maximum intensity projection through the whole head/neck), which indicate that a significantly increased hotspot results with this relaxed parameter. The table in Figure 2 shows that IMPULSE achieves the fastest total computation time while achieving the lowest peak local SAR.


IMPULSE has a clear advantage over the use of VOP-based methods in terms of both total computation time and the SAR performance of the optimized pulse. For VOP-based methods, there is a tradeoff that can be controlled with the overestimation parameter. Low values improve SAR performance at the expense of longer compression time whereas high values allow for rapid compression but with poorer SAR performance; however, in both cases performance is worse than using IMPULSE. Additonally, IMPULSE enables patient-specific SAR estimation, which ultimately will be more accurate than using body models with pre-computed E-fields and VOPs. There are several possible methods to generate the necessary tissue electrical properties for the patient on the table. With modern-day GPU parallelization of the electromagnetic simulation step, it is conceivable that the total time needed for tissue model generation, EM field simulation, and pTx pulse design could be reduced to a few seconds, i.e. fast enough to enable SAR-aware pTx while the patient is in the scanner. The elimination of the time-consuming VOP step using IMPULSE makes “real-time” patient-specific SAR-aware pTx a realistic possibility.


Research support from the NIH (P41 EB015891, 1 S10 RR026351-01A1), GE Healthcare, and Zurich MedTech AG (Sim4Science program).


[1] Eichfelder, MRM 2011;66:1468–1476 [2] M. Pendse and B. Rutt. IMPULSE: A generalized and scalable algorithm for joint design of minimum SAR parallel transmit RF pulses. Proceedings ISMRM 23:5008, 2015 [3] Hoyos-Idrobo, Andres, et al. "On variant strategies to solve the magnitude least squares optimization problem in parallel transmission pulse design and under strict SAR and power constraints." IEEE transactions on medical imaging 33.3 (2014): 739-748.


Figure 1: L-curves of SAR vs FAI using uncompressed SAR matrices (IMPULSE) and using VOPs with overestimation parameter of 0.025 and 0.15

Figure 2: Computation times for both compression and optimization and SAR maximum intensity projection (including peak local SAR values) for pulses designed with a 5% flip angle inhomogeneity tolerance.

Proc. Intl. Soc. Mag. Reson. Med. 25 (2017)