To implement and evaluate blip reversed non-segmented 3D EPI pseudo-continuous arterial spin labeling (PCASL) for reduced susceptibility artifacts.Pseudo-continuous arterial spin labeling (PCASL) is the method of choice in arterial spin labeling due to its superior labeling efficiency and SNR as well as compatibility with RF transmission hardware [1]. For the readout module of ASL, non-segmented (single-shot) 3D EPI acquisition offers higher SNR efficiency and reduced variation in signal arising from phase inconsistencies in multi-shot techniques. However, one drawback is image distortion resulting from static field inhomogeneity and susceptibility. Among distortion correction techniques, blip reversal provides a simple yet effective means for correction without the need to explicitly obtain field inhomogeneity maps[2,3]. It has been successfully applied to DWI using 3D EPI acquisition[4].

3D single-shot EPI acquisition with PCASL was modified so that the first set of tagged and control images were acquired with k_y space traversed from –k_y to +k_y (blip-up) while a second pair was acquired with k_y space acquisition reversed (blip-down). Eight volunteers were scanned on a 3T Philips Achieva scanner using a 8 channel head coil. The PCASL 3D EPI sequence used had the following parameters: PCASL: label dur = 1.65 s, label delay = 1.8 s; ACQ: FOV = 24 × 20 cm, res: 3×3×4mm³, EPI TR/TE = 22/11ms, centric k_z encoding, SENSE (y) = 2.5, 30 slices, phase encoding (R/L), spectral-spatial excitation pulse θ=25° with optimal flip angle train for reduced blurring[5], 34 dynamics, scan time: 4:50. A single background suppression (BS) adiabatic inversion pulse was used at TI=1.8 s after the initial saturation pulse. In Bloch simulations, the normalized magnetization was 0.85, 0.96 and 0.52 in gray matter(GM), white matter(WM) and cerebrospinal fluid(CSF), respectively, without this inversion pulse and was 0.37, 0.62 and 0.09 with the inversion pulse.

The post-processing pipeline included the following steps: (a) 3D rigid registration of images to the first acquired BU or BD control image using FLIRT [6]. (b) All labeled and control BU, BD images were summed to get 4 sets of images. (c) Distortion correction[2,3] was performed using information from the four sets. (d) Finally, control images were subtracted from labeled images to get reduced distortion cerebral blood flow (CBF) images.

To ascertain reduction in distortion, a GM mask was prepared from relatively distortion free 3D MPRAGE images after 3D registration, skull stripping [7] and segmentation [8] for each slice and overlaid with CBF images. The number of voxels which overlapped in matched GM from MPRAGE and CBF images from the three sets (BU, BD and distortion corrected) was counted over the entire brain for all eight volunteers. Student’s t-test (matched, 2-tailed) was done to check for significant differences between the voxels overlapping the GM mask.

Figure 1 shows six slices (every fifth slice of 30) of distortion corrected CBF maps. Non-segmented 3D EPI achieves full brain coverage in <5 minutes.

Figure 2 shows the comparison between the distortion corrected images and the uncorrected images in a single slice. In the control images, the distortion along the phase encoding direction (R/L) is clearly evident, particularly in the frontal lobes and ventricles. The asymmetry of the frontal horns and the signal pileup in the frontal white matter is corrected by following application of the distortion correction algorithm. A similar distortion is seen in CBF images. The bottom row shows the overlap of the cortical gray matter mask with the thresholded perfusion map. Ideally, all voxels in the gray matter mask should correspond to highly perfused tissue.

There were 5% more voxels matching with the mask in the distortion corrected images when compared with BU and 3.2% more when compared to BD images. There were significant differences between the BU and BD CBF maps when compared with the distortion corrected CBF maps (p=0.0001 and p=0.018, respectively) indicating significantly improved localization of the CBF signal.

No additional scan time was required when compared with standard EPI acquisition since a number of dynamic phases are always employed for ASL. By interleaving the BU-BD acquisitions, we reduced misregistration between the two due to motion. Spectral-spatial RF excitation pulse was favored over a fat saturation (FS) pulse since residual fat signal shifted into the brain (when FS is used) made distortion correction difficult. Additional signal loss due to signal dephasing in regions of strong field offset was not taken into account.Performing distortion correction on non-segmented 3D EPI acquisition should provide comparable localization of CBF as segmented acquisition based techniques while providing higher SNR efficiency.