DTI relies on EPI acquisitions which are plagued by geometric distortions along the phase-encode (PE) direction¹. A common correction, such as FUGUE² in FSL (FMRIB Software Library), employs a B0 field map, requiring extra non-EPI scans to generate the map. This has the disadvantage of requiring more scan time and it is motion sensitive beause if the subject undergoes head motion between the B0 mapping scans and the DTI acquisition, B0 maps cannot properly align with the distorted images for effective correction. A more general phase-based method of distortion correction is called PLACE (Phase Labeling for Additional Coordinate Encoding)³. PLACE relies on the phase difference between EPI-based images that differ only by an extra PE blip and, unlike with B0 maps, no phase unwrapping is required. PLACE has previously been applied to DTI using extra EPI scans with varying PLACE blips^4,5 or by acquiring two DTI datasets (with and without blip) and extracting the PLACE information from the non-diffusion weighted images (nDWI)⁶. In this work, we modify the DTI acquisition and embed PLACE blips, in alternating fashion, across the set of nDWIs of the DTI data set. As such, the distortions can be reverted without additional scan time. We demonstrate that PLACE can be embedded successfully in a typical brain DTI scan despite the complications of multi-channel SENSE-like acceleration and partial k-space. We show that when motion is present between B0 mapping and DTI scans, this embedded PLACE yields superior distortion corrections compared to those obtained with conventional B0 map-based correction schemes.

The stock pulse sequence for DTI is slightly modified to include a PLACE blip: the amplitude of the pre-winder gradient is adjusted by 2∆ky (where ∆ky is the PE EPI blip). This embedded PLACE blip is incorporated for even-numbered nDWI. This modification does not change the sequence timing nor does if affect the magnitude nDWIs. Complex-valued images were reconstructed off-line using Orchestra SDK (GE research tool, Waukesha, WI) and phase information from the PLACE embedded nDWI data was used to perform the distortion correction on all images in the DTI data set⁶.

To test the sequence, three volunteers were scanned with consent as per institutional REB. All acquisitions were performed on a 3T-MR750 GE scanner (General Electric, Waukesha, WI) using an eight-channel receive-only head coil. A standard DTI protocol was used: N_nDWI = 5, N_DWI = 23, and b = 1000 s/mm². Other scanning parameters were TE=82ms, TR=6.8s, FOV=25.6cm, Nx=Ny=128, slice thickness = 4mm, 45 slices, partial k-space, ASSET=2. To demonstrate the effects of motion, B0 maps were obtained once, from two gradient echo scans taken prior to the DTI acquisitions (TE₁/TE₂=6.5ms/8.5ms). Then, two PLACE-embedded DTI data sets were acquired with the volunteer undergoing realistic head motion in between scans. Motion metrics were obtained using FSL's avscale⁷. For full brain coverage at 4mm isotropic resolution and TR=1s, this added 2.5min of scan time. Undistorted, reference T2-weighted FSE images were also acquired at each head position (T2_DE_FSExL, TE1/TE2 = 11.152ms/89.216ms, TR=2.5s). To assess the accuracy of the distortion correction, FSL's FLIRT⁷ was used with 6 degrees of freedom, to coregister the head positions to an anatomical reference. Edges from a coregistered average nDWI, before and after distortion correction, were overlayed on the FSE T2W reference image.

In the absence of motion, the embedded PLACE, requiring no extra scans nor phase unwrapping, is able to correct geometric distortions as well as FUGUE. In the case of motion occurring between the B0 mapping and DTI scans, distortion correction by FUGUE is worsened, in all cases, as the B0 map no longer applies to the new head position, despite coregistration to account for the motion. In contrast, embedded PLACE is not affected, i.e., geometric distortions are corrected. Table 1 lists metrics for the motion occurring between the B0-map scans and the second DTI acquisition. Figure 1 displays representative results where edges of the distortion corrected nDWIs are overlayed on the FSE reference. Zoomed regions are highlighted below each image. For volunteer 1, edges match best for PLACE at the posterior brain border and gyri. For volunteer 2, the edges of the ventricles are better matched with PLACE. Quantitative analysis of the geometric distortion is underway using white matter segmentation and FA maps, for a more rigorous comparison. Also, more subjects and motion are being tested.Embedding PLACE blips on nDWIS within the DTI set allows for effective distortion correction of the images. This method requires no extra scans and it is more robust to motion than using extra scans for B0 mapping.