We present a prospectively accelerated 3D black-blood sequence in conjunction with CS reconstruction and show the feasibility of 3D high-resolution (≤ 0.5mm) carotid vessel wall imaging. 3D black-blood strategies are attractive for allowing full coverage carotid imaging with an isotropic resolution. However, the typical voxel size of these scans (0.7x0.7x0.7 mm³) does not match the in-plane resolution generally used in 2D imaging (0.5x0.5 mm²), which may lead to overestimation of wall thickness values and also complicate delineation of plaque components. Imaging at higher isotropic resolution (≤ 0.5 mm) is hampered by reduced SNR or lengthy acquisition times resulting in motional blurring or artifacts. Few studies report on the use of compressed sensing (CS) acceleration for carotid imaging [1,2] and often only apply retrospective undersampling of fully acquired low-resolution data. Here we investigate the use of prospective acceleration as a means to increase SNR efficiency, thereby facilitating 3D high-resolution carotid artery imaging.

Sequence – All scans were performed on a Philips 3T Ingenia MR scanner. A 3D iMSDE TFE with SPIR fat suppression [3] was used as the basic sequence for black blood carotid imaging, using the following parameters: prepulse length: 11.5 ms, m1 = 234 mTms²/m, TR = 10-11 ms, TE = 3.65 ms, flip angle = 8°, TFEfactor = 60.

Data acquisition and Reconstruction – A custom sequence patch was written to enable acquisition of arbitrary ky-kz trajectories. Figure 1 shows a typical sampling pattern where the color indicates the echo number in each TFE shot. For maximizing black-blood efficiency, we choose a radial ky-kz sampling pattern to measure central k-lines directly after the motion-sensitized prepulse. For scan acceleration, a Poisson disk distribution of ky-kz space with a 25x25 calibration center was chosen corresponding to an effective acceleration of R = 4 relative to a full k-space with elliptical shutter. In 4 healthy volunteers, fully sampled data was acquired at 0.7 and 0.5 mm isotropic resolutions, while CS accelerated scans were acquired at 0.5 and 0.4 mm resolutions using 4 and 3 repetitions of the same k-space sampling pattern, respectively. All scans were within a clinically feasible imaging time of 4 min. A summary of geometry and sequence parameters is given in Table 1. Raw data were imported into MATLAB using ReconFrame (Gyrotools, Zurich, Switzerland). CS reconstruction was performed using the open-source BART toolbox [4] using a L1-iterative self-consistent parallel imaging reconstruction technique (L1-SPIRiT) [5]. For the sparse domain, L1 wavelet sparsity with regularization (r = 0.05) was used. In addition, coil sensitivity maps were estimated directly from the center of the acquired k-space data using a calibration region of 60x25x25 voxels. For the 0.5 mm case, either 4 or 2 repetitions were used for signal averaging prior to CS reconstruction.

Data analysis – Reconstructed data were exported as DICOM files and analyzed using Vessel Mass (LUMC, Leiden, Netherlands). Mean vessel wall area from 5 slices in the common carotid artery (CCA) was calculated for several combinations of spatial resolution and number of signal averages. Values from different acquisitions were compared using ANOVA with repeated measures and a post-hoc paired Student’s t-test. A p value < 0.05 was considered statistically significant.

Fig. 3 contains reconstructions of a 0.5mm isotropic acquisition, showing typical CS results at different number of signal averages (NSA) as well as a standard linear reconstruction of fully acquired data. Although imaging time is equal for the full and CS NSA4 case, the latter shows clear improvement in vessel wall visibility, especially in regions with low SNR. This shows that high-resolution carotid artery imaging benefits from acceleration in combination with additional signal averaging. Furthermore, the CS NSA2 case (with an effective acceleration of 2) has equal image quality as compared to the full data case. Axial reconstructions from scans at different resolutions are shown in Fig. 4, in which a true increase in resolution can be appreciated due to improved vessel wall delineation and enhanced structural detail. Quantitative results in Fig. 5 indeed show that the measurements of vessel wall area are significantly reduced from 0.7 mm to 0.5 mm (*), as well as from 0.5 CS to 0.4CS (#). No significant difference in vessel wall area was found between the 0.5 mm full and CS accelerated acquisitions.We have successfully implemented a prospectively accelerated 3D black-blood sequence allowing high-resolution carotid artery imaging. We foresee that this method will increase the application of 3D MRI studying atherosclerotic disease progression, by more accurate assessment of vessel wall dimensions and improved plaque component delineation.