Compared with single-shot EPI, interleaved EPI (iEPI) DWI can provide higher spatial resolution and less geometric distortions. However, physiological motion can cause phase inconsistency among different shots, thus inducing ghost artifacts in DWI images. Several methods, performed either in the image ¹ or k-space ^2-4 domain, have been proposed to solve this problem with extra acquired navigator data. However, the navigator is usually acquired with a lower bandwidth in the phase encoding (PE) direction than the image echo, which can cause different distortion levels. In this study, the effects of such distortion on the performance of the image or k-space based reconstruction are investigated.

Sequence The sequence diagram for navigated spin echo EPI DWI acquisitions is shown in Fig. 1(a). The echo spacing of the navigator (T_nav) is first set to the same as the image echo. Then it is reduced to 1/R_nav of the echo spacing of the image echo, where R_nav is the reduction factor. Noticeably, to ensure enough sampling points for a navigator, when R_nav>1, the number of PE lines for the navigator is increased. The corresponding sampling trajectories for different R_nav (R_nav = 1, 2, 3) are shown in Fig. 1(b). Meanwhile, the sampling bandwidth in the readout (RO) direction and the slope time of the RO gradient for the navigator are unchanged.

Reconstruction The iEPI DWI reconstruction can be performed either in the image or k-space domain. In the image domain, the low resolution navigator image is first reconstructed, and then the shot-to-shot phase variations are calculated and corrected ¹. While in the k-space domain, data from different shots can be taken as encoded by both phase variations and coil sensitivities, and the missing data of each shot are recovered through convolution, like GRAPPA ^2-5. In this study, image reconstruction using image-space sampling function (IRIS) ¹ is used as the image based reconstruction, while synergistic image reconstruction with phase variation and sensitivity (SYMPHONY, previously named SEPARATE) ^{2, 4} is used as the k-space based reconstruction. It should be noted that for IRIS used here, no image registration between navigators and DWI images is performed. After the reconstruction, the SNR is calculated using the pseudo-multiple replica method ⁶ with 128 repetitions.

Experiments All scans were performed on a Philips 3.0T Achieva TX MRI scanner (Philips Healthcare, Best, The Netherlands) using a 32-channel head coil. All human studies were performed under IRB approval from our institution. The imaging parameters were as follows: FOV = 240 × 240 mm², acquisition voxel size = 1 × 1 × 4 mm³, 4 slices with gap = 16mm, 8 shots with 29 echoes per shot, no partial Fourier, TE = 77ms, TR = 3s, scan time = 1.75min, diffusion preparation was applied in 3 orthogonal directions with b = 800 s/mm², NSA = 1. For different R_nav, the acquisition matrix sizes of the navigator are summarized in Table 1.

Fig. 2(a) shows the navigator images for different R_nav. For a better illustration, all navigator images are zero-padded to 240 × 240. The slice consisting of frontal lobe of the brain (indicated by green line in Fig. 2(b)) is chosen for display, where distortions are severe at the air-tissue interfaces (indicated by red arrow head). A T2 weighted TSE image with the same resolution 1 × 1 × 4 mm³ is shown in Fig. 2(c) as reference. As shown, with R_nav increasing, the navigator distortion is largely reduced.

Fig. 3 shows a comparison between DWI images reconstructed with the image (a) or k-space (b) based methods for different navigator distortion levels. With a severely distorted navigator, the DWI image reconstructed from the image based method is largely corrupted (yellow arrow heads in Fig. 3(a)). When the navigator distortion is minor (R_nav=3), the image based method can achieve comparable performance as the k-space based method and even show a slightly higher SNR, as shown in Fig. 3(c) and (d). In general, the k-space based method works well for different navigator distortion levels.

In interleaved EPI DWI, the k-space based reconstruction is more tolerant to navigator distortions, compared with the image based method. This indicates that the k-space method can be a better choice when the image registration between image and navigator is not easily implemented.