0379

Three Vencs for the Price of Two: Efficient Multi-Venc Phase-Contrast MRI for Improved Velocity Dynamic Range
Liliana Ma1,2, Kelvin Chow1,3, Can Wu4, Alireza Vali1, Michael Markl1,2, and Susanne Schnell1

1Department of Radiology, Northwestern University, Chicago, IL, United States, 2Department of Biomedical Engineering, Northwestern University, Evanston, IL, United States, 3Cardiovascular MR R&D, Siemens Medical Solutions USA, Inc., Chicago, IL, United States, 4Philips Healthcare, Gainesville, FL, United States

Synopsis

Single-venc 4D flow MRI is inherently limited by the need to set a velocity encoding sensitivity (venc), where velocity(v)>venc results in velocity aliasing and v<<venc results in elevated noise. Thus, we propose a novel multi-venc encoding scheme for reconstruction of a triple-venc 4D flow dataset from our previously described 7TR dual-venc sequence. This triple-venc dataset can be used to improve velocity unwrapping without increasing scan time. The aim of this study was to systematically evaluate the utility of 7TR triple-venc velocity encoding to improve velocity dynamic range and further decrease velocity noise beyond the capabilities of current dual-venc methods.

Purpose

Comprehensive evaluation of cardiovascular disease often requires measurement across a wide velocity dynamic range (VDR) to quantify high velocity jets (400 cm/s) near adjacent low circulation zones (10 cm/s). However, evaluation by conventional single-venc 4D flow MRI is inherently limited by the need to set a velocity encoding sensitivity (venc), where velocity(v)>venc results in velocity aliasing and v<<venc results in elevated noise. To address this limitation, we previously developed a dual-venc 4D flow MRI sequence, using a shared reference scan followed by two 3D acquisitions in 7 repetition times (TR).1,2 The conventional dual-venc reconstruction used the non-aliased high-venc data to unwrap the low-venc data with a higher velocity-to-noise ratio (VNR). However, current dual-venc unwrapping algorithms are often limited by a high-to-low venc ratio of ~2:1 to avoid problematic multiple phase wraps. Specifically, Zwart et al. found that while decreasing the low-venc increases the VNR, it in turn increases the v to venc ratio, increasing noise sensitivity and limiting the reconstruction algorithm.3 Multi-venc encodings have shown utility in improving VDR at the expense of further increased scan or reconstruction time.4,5

Thus, we propose a novel multi-venc encoding scheme for reconstruction of a triple-venc 4D flow dataset from our previously described 7TR dual-venc sequence. This triple-venc dataset can be used to improve velocity unwrapping without increasing scan time. The aim of this study was to evaluate the utility of a prototype 7TR triple-venc velocity encoding implementation to improve the VDR and VNR beyond the capabilities of current dual-venc methods, using a highest-to-lowest venc ratio of 3:1 and a bi-conditional unwrapping algorithm.

Methods

Intrinsic multi-venc encoding: Time-resolved velocities were calculated for the three vencs as shown in Figure 1. Multi-venc reconstruction employed a shared reference scan to calculate flow images with venc1 (lowest venc, reference scan in TR1 subtracted from TR2-4, red) and venc2 (TR5-7-TR1, green). An additional set of phase-difference images utilizing 6 TRs (TR2-7) yielded venc3 (highest venc, blue). The desired change in first-order gradient moment for venc1 (I), ΔM1i(I), i= x,y,z, was equally distributed between the reference scan and subsequent venc1 scans , with venc2 (II) first moments calculated based on the same reference. The resulting venc3 (III) depended on the change in first moment between the venc1 and venc2 scans, $$$venc3_i=\frac{\pi}{\gamma(M_{1i}^{(I)}-M_{1i}^{(II)})} $$$. This new scheme enables selection of variable ratios of the three vencs (Fig. 2A), which can increase VDR, and thus further improve VNR.

Multi-venc reconstruction: Data from phase-difference images for all 3 vencs were used to un-alias the venc1 scan to generate a single, unwrapped dataset with the favorable VNR of the lowest venc. Using a venc3:venc2:venc1 ratio of 3:1.5:1, for a given voxel, if |venc1,2-venc3| was within the 2*venc1,2±ε boundary (Figure 2B2), venc1 was considered aliased. Multi-venc phase-difference reconstruction including Maxwell correction for all 3 vencs was implemented in the scanner's online reconstruction pipeline. Offline processing included eddy current correction.

In-vitro flow phantom validation: Multi-venc 4D flow MRI was tested in-vitro using a pulsatile flow pump and a 3D-printed aorta with an 80% coarctation in the descending aorta. All phantom data were acquired with the following multi-venc sequence parameters: venc1/venc2/venc3=110/165/330 cm/s, TR/TE=5.0/2.6 ms, FA=15°, resolution=(2.4 mm)3. A 4TR single-venc (SV) scan was used for comparison with: venc=330 cm/s, TR/TE=4.7/2.2 ms, FA=15°, resolution=(2.4 mm)3.

In-vivo application: Multi-venc 4D flow MRI was evaluated in-vivo in 4 healthy volunteers (all male, age=26, 37, 56, 75 y.o., venc1/venc2/venc3 = 50/75/150 cm/s, TR/TE=5.5/3.1 ms, FA=7°, typical resolution = 2.4×2.4×2.7 mm3). One volunteer had an additional SV scan (venc=150 cm/s, TR/TE=4.9/2.5 ms, FA=7°, resolution=2.4×2.4×2.6 mm3). All scans were acquired on a 1.5T MAGNETOM Aera system (Siemens Healthcare, Erlangen, Germany) with k-t GRAPPA, R=5.

Results

Multi-venc reconstruction and unwrapping were performed successfully in all datasets. Compared to SV 4D flow MRI, multi-venc encoding provided increased VNR (153% and 165% in the phantom and one volunteer, respectively) and improved streamline integrity (Figure 3). Flow waveforms in Figure 4 depict successful multi-venc unwrapping in areas of predictable flow, however artifacts due to intravoxel dephasing could not be recovered. Figure 5 shows improved multi-venc unwrapping compared to dual-venc unwrapping, however, we still noticed some missed or incorrectly unwrapped voxels (red arrow).

Discussion and Conclusions

The proposed efficient multi-venc encoding scheme has the potential to extend VDR beyond a traditional dual-venc algorithm. Future investigations will include an improved unwrapping algorithm that considers voxel-by-voxel velocities in all three vencs simultaneously (i.e. model fitting), comparison to a 7TR DV acquisition with venc2=venc3 and traditional unwrapping, as well as further investigations of the temporal blurring and correlative noise effects of venc3 using a simpler phantom.

Acknowledgements

Grant support by NIH F30HL137279, NIH R21 AG055954, and AHA 16SDG30420005.

References

1. Schnell, S., et al., Improved assessment of aortic hemodynamics by kt accelerated dual-venc 4D flow MRI in pediatric patients. JCMR, 2016. 18(1): p. 1.

2. Schnell, S., et al., Accelerated dual-venc 4D flow MRI for neurovascular applications. JMRI, 2017. 46(1): p. 102-114.

3. Zwart, N.R. and J.G. Pipe, Multidirectional high‐moment encoding in phase contrast MRI. MRM, 2013. 69(6): p. 1553-1563.

4. Binter, C., et al., Bayesian multipoint velocity encoding for concurrent flow and turbulence mapping. MRM, 2013. 69(5): p. 1337-1345.

5. Knobloch, V., et al., Mapping mean and fluctuating velocities by Bayesian multipoint MR velocity encoding‐validation against 3D particle tracking velocimetry. MRM, 2014. 71(4): p. 1405-1415.

Figures

Multi-venc 4D flow sequence with a shared reference scan between venc1 and venc2 in the read, phase, and slice (x, y, z) directions, and combined use of venc1 and venc2 scans for reconstruction of the venc3 data. The red lines connect the two scans used to calculate venc1 phase difference images and the green lines the venc2 images. The associated first moment of each TR is depicted above the gradient waveforms, and these moments are referenced in the equations above the blue lines representing calculation of the venc3 phase-difference images.

A, Relations of venc1, venc2, and venc3 for multi-venc encoding. As venc2 increases relative to venc1, venc3 approaches venc1. On the opposite extreme, as the vencs get closer together, venc3 approaches infinity. The blue dot and arrow represent the venc ratios used in this study. B, the unwrapping procedure used. While the traditional dual-venc unwrapping algorithm finds a difference in velocities at a high and low venc, the two-condition algorithm used in this study compares vencs1 and 2 to venc3, and unwraps based on if either venc1 or 2 is aliased. ε represents soft boundaries around the aliasing threshold.

Diastolic streamlines comparing a single-venc 4D flow scan and the unwrapped venc1 scan. Arrows show the improvement in streamline integrity, as well as depiction of low flow, helical recirculation patterns in the ascending aorta of an older volunteer (top row) and in the complex flow associated with the 80% coarctation in the aortic phantom (posterior view, bottom row).

Representative flow curves for one volunteer (A) and the pulsatile phantom (B) in the ascending (AAo) and descending (DAo) aorta. AAo curves are relatively well-matched. A, volunteer’s DAo venc2 and venc3 curves show aliased behavior, while the unwrapped venc1 curve again aligns with the unaliased curves. C, Yellow and Purple purple lines represent semi-automatically-placed, quantification locations in A and B. Maximum intensity projections (MIPs) show significant intravoxel dephasing in the DAo of the 80% coarct phantom that limited the unwrapping algorithm in the DAo (B). Note: MIP used for single plane visualization, but dephasing was more significant on individual images.

Representative volunteer images of aliased and un-aliased aorta velocities in three velocity directions at one slice and time point. Top row corresponds to the x-direction, middle, y, bottom, z. A,Columns 1-3 represent velocity images for venc1 = 50 m/s, venc2 = 75 cm/s, and venc3 = 150 m/s. UW = unwrapped. B shows unwrapping using the algorithm in Figure 2B1, with venc1 as the low-venc and venc3 as the high. C shows unwrapping using the multi-venc algorithm (Fig. 2B2). Black arrows show improved un-aliasing in C compared to B. Red indicates incorrect/incomplete unwrapping, requiring further investigation of unwrapping algorithms.

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)
0379