To evaluate several static RF shimming¹ calculation strategies for maximizing the lowest transmit field strength in myocardium at 7T.It has been demonstrated that 7T human first-pass myocardium perfusion is feasible², but the high transmit energy required by saturation to compensate large RF inhomogeneities is a major limiting factor. The required energy is most sensitive to the regions of the lowest transmit field strength, while regions of the highest strength can be saturated with very little energy. In such cases, transmit field uniformity is less critical as a shimming target than maximizing the lowest strength. The B1 shimming optimization cost function corresponding to the lowest strength is not smooth, leading to impractically long calculation if all transmit magnitudes and phases are optimized simultaneously. Hence it is necessary to find an optimization strategy that is both practically fast and reasonably effective.

The shimming calculation took relative B1 maps, from each of 8 transmitting channels, as input. The target was to find a combination of transmit magnitude and phase values for all channels that maximizes the minimum combined relative B1 in selected region of interest (ROI). Five methods were evaluated.

1. Phase matching³ (Mp). For each channel, the negative of the average phase of the relative B1 values in ROI was used as the transmit phase, and the magnitude was fixed at one (full power).

2. Phase only optimization (Op). The magnitudes were fixed at one. All phases were optimized simultaneously in a constrained nonlinear optimization.

3. Phase matching and magnitude optimization (MpOm). All magnitudes were optimized simultaneously after Mp.

4. Phase and magnitude optimization (Omp). All phases and magnitudes were optimized simultaneously. If a global maximum is found, it will be the best possible solution. However, the cost function was not smooth, and most iterations ended at local maximums, depending on the initial values they started with. With high degrees of freedom, an exhaustive search is not practical either. As a compromise, 5 or 6 values were tested for each degree of freedom to calculate a large number of results. 20 sets of magnitude and phase values giving best results were used as initial values for 20 iterations of optimization. The best solution from these 20 iterations was used as the output and also as a reference for evaluating other methods.

5. Two-stage phase and magnitude optimization (Omp2). Four channels with high relative B1 or narrow phase distribution were manually selected as dominant channels. Omp was applied to these channels as the first stage. In the second stage, the dominant channels were treated as one channel, combined as optimized in the first stage, in a second Omp with all the other channels.

Fast low-angle shot (FLASH) images were collected (imaging parameters listed in Fig.1) in a single breath-hold, transmitting on one channel at a time. An absolute B1 map was acquired in a separate breath-hold transmitting on all channels. The absolute B1 map was used to remove the receiving sensitivity variation in the FLASH images to calculate relative B1 maps.

For each subject, the minimum combined relative B1 in ROI from each method was divided by the value from Omp to obtain a dimensionless measure. Mean and standard deviation were then calculated over all subjects.

Twelve healthy male volunteers (age 33±5 years, weight 76±5 kg) were scanned with a whole-body 7T scanner (Siemens) and an eight-channel transceiver array⁴. Myocardial ROIs were manually segmented.

All shims were calculated in MATLAB (MathWorks) using the fminimax function. A computer with two 8-core CPUs (Intel Xeon E5-2650) and 64 GB RAM was used for all calculation.

Fig.2 shows relative B1 maps from one volunteer.

Typically around 130 pixels were included in each ROI. For each slice, it took about 3 days for Omp, and less than 10 seconds for other methods.

Fig.3 shows the minimum relative B1 from each method.

The good performance of Omp2 may be attributed to the fact that the positions of transmit elements relative to the heart were fairly consistent over all volunteers. It was also observed that for most subjects shims from the first stage of Omp2 were already the same as those from the reference method.

Shimming multiple slices of the myocardium simultaneously can be more challenging. This may require further investigation.

The results suggest that the proposed method is able to consistently produce near optimal shims, and it is fast enough for in situ implementation.