BMART: B0 Mapping using Rewind Trajectories
Corey Allan Baron1 and Dwight G. Nishimura1

1Electrical Engineering, Stanford University, Stanford, CA, United States


B0 inhomogeneity leads to image artifacts and/or blurring. These issues can be addressed by using a B0 map, which typically requires an extra scan. In addition to the longer total scan time required, motion occurring between the acquisition of the imaging data and B0 map can lead to misregistration. The proposed method utilizes images reconstructed from rewind trajectories to construct a B0 map. In pulse sequences that already use gradient rewinds (e.g., bSSFP), a B0 map that is inherently registered to the imaging data can be created with no additional scan time.


To utilize trajectory rewinds to create B0 maps that are automatically registered to the imaging data and require no additional scan time: B0 Mapping using Rewind Trajectories (BMART).


BMART B0 mapping requires that the rewind portions of the trajectory, as used for example in bSSFP, are evenly distributed throughout k-space. This criterion can be satisfied for center-out trajectories, where the sampling density of the rewinds often resembles that of undersampled projection imaging.

The proposed B0 map reconstruction technique is illustrated in Figure 1a. The k-space data from the outgoing part of the trajectory (kOUT) and from the rewinds (kREW) have delayed effective TE with respect to each other, and a B0 estimate can be estimated from their difference in phase. However, different portions of k-space have different time delays between kOUT and kREW (Figure 1b). This is accounted for by splitting the k-space data into N bins, where each bin has an effective delay Δti. The outgoing and rewind data is then rebuilt sequentially from the longest Δt (i.e., near k=0) to the shortest (near k-space edge). At the end of each step, the current estimate of B0 (B0i) is used to advance the phase of the images (outgoing and rewind) rebuilt so far to Δti+1. Gridding and low-pass filtering (Hanning window) of the k-space data is performed before applying BMART, and the average sampling time for each gridded sample is used to determine Δti.

Methods and Results

To demonstrate the method, we have applied BMART to data acquired using 3D cones 1 in a phantom, human brain, and heart on a 1.5 T GE Signa Excite. The 3D cones parameters were: TR = 5.6 ms, TE = 0.6 ms, flip 70°, BW = 250 kHz, FOV = 28 x 28 x 14 cm3, 1.2 mm isotropic resolution, 9137 readouts, readout duration = 2.8 ms, 18 cones per heartbeat (for heart scan). The only sequence timing change required was to collect data during the rewinds, which necessitated a 100 μs longer TR for RF transmit/receive switching delays (normally accounted for during the rewinds). The phantom and heart scans used ATR-bSSFP, while the brain scan used spoiled gradient echo (SPGR) to achieve more gray-white matter contrast. Images reconstructed from kREW (direct gridding of all rewind data, with no Δt correction) were compared to those from kOUT to demonstrate sufficient sampling density of the rewind portion of the trajectory (Figure 2).

For validation, the BMART B0 maps in the phantom and brain were compared to B0 maps obtained using dual echo gradient echo (DE-GRE) (Figure 3). The DE-GRE parameters were: TR = 134 ms, TE1/TE2 = 4/6.5 ms, flip 30°, BW = 62.5 kHz, 36 axial slices (3 mm), FOV = 24 x 24 cm2, 1.9 mm isotropic in-plane resolution. B0 correction was performed on the main outgoing trajectories with the BMART B0 maps (phantom and brain: Figure 4, heart: Figure 5). In the cardiac scan, BMART was applied after rigid-body motion correction.


Excellent agreement between the two methods was observed (Figure 3), and deblurring using the BMART B0 map improved image sharpness (Figure 4, 5). The primary disadvantage of BMART over traditional B0 mapping techniques is the lack of freedom in choosing the TE difference. The advantages of BMART are that the B0 map is automatically registered to the data to be corrected, and that for projection-like trajectories no additional scan time, nor appreciable compromises of TE or TR, are required. In addition, BMART may enable highly efficient trajectory design for dedicated B0 mapping, since it requires no “fly-back” time between echoes.


We acknowledge the following funding sources: NIH R01 HL127039, GE Healthcare.


1. HH Wu, PT Gurney, BS Hu, DG Nishimura, and MV McConnell. Free-breathing multiphase whole-heart coronary MR angiography using image-based navigators and three-dimensional cones imaging. MRM 2013;69(4):1083-1093


Figure 1. (a) Data from the outgoing (kOUT) and rewind (kREW) portions of the trajectory are split into N bins of sampling delay Δti. The bins are processed from low-to-high k, phase correcting images along the way (FT = fourier transform). (b) kOUT and kREW for an example spiral trajectory.

Figure 2. Images reconstructed from kREW are comparable to kOUT, albeit with additional blurring due to undersampling.

Figure 3. B0 maps generated using BMART from 3D cones and from dual-echo gradient echo (DE-GRE) are in agreement. The maximum frequency offsets before encountering phase-wrapping were ±180 Hz and ±200 Hz for the 3D cones and DE-GRE scans, respectively.

Figure 4. Improved image sharpness in the phantom (ATR-bSSFP 3D cones) and brain (SPGR 3D cones) is observed after off-resonance correction using the BMART B0 maps.

Figure 5. B0 map obtained using BMART on a segmented, cardiac triggered ATR-bSSFP 3D cones acquisition. Improved vessel contrast is observed with the application of the BMART B0 correction. The profile plot is from the location identified by the yellow line.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)