Long-term subject motion is a major challenge for Diffusion Tensor Imaging (DTI). It is common practice to perform a retrospective motion correction, but overall results are likely to improve using a prospective motion correction to minimize the motion to be corrected by post processing. Registration-based prospective motion correction is well established in BOLD imaging¹, but application to diffusion-weighted (DW) imaging² is problematic due to large contrast variation between images. Previous work^3,4 proposed intermediate, pseudo-trace-weighted (psTW) images for fast, real-time image registration with low sensitivity to the contrast variation; this approach was shown to provide reliable motion estimates for b-values up to 3000s/mm². This abstract describes, for the first time, the use of these intermediate, psTW images to perform prospective motion correction in DW imaging.

Before scanning, the diffusion-gradient directions are reordered to provide a new acquisition scheme in which each set of three consecutive directions are highly mutually orthogonal. During the scan, a psTW image is calculated from the current and the two preceding DW images by taking their geometric mean. The psTW images are used as input for a rigid-body motion detection related to the first psTW image using a least-squares cost-function. This results in rapidly calculated motion estimates without the influence of contrast variation between DW images.

The calculated results are then fed back to the sequence and used to adapt the slice position and orientation during the ongoing scan (Fig. 1). In this procedure, the motion estimates are used to derive a linear transformation, related to the average slice position and orientation⁵ of the three images used to calculate the psTW image. This includes a quaternion multiplication to acquire the new orientation and a linear shift, weighted with the measured rotation, to get the new position. To maintain a consistent relationship between imaging and diffusion gradients, the diffusion-gradient direction is also adapted to fit the new imaging plane.

The method was evaluated using a dedicated DW imaging sequence with 64 diffusion-gradient directions at 3T (Skyra, Siemens Healthcare). A b-value of b=1000s/mm² and GRAPPA factor of 2 were used (TE=86.47ms/TR=5524.80ms) at 48 slices with 2mm isotropic voxel size. No correction was applied for frequency drift during the scan. As gold standard reference, b=0s/mm² images were interleaved between the DW images to estimate the true subject motion; these were not used by the prospective motion correction. Three individual subjects were scanned; all with and without enabled motion correction. There was no head restraint or intentional head movement.

Figure 2 demonstrates the reduction of contrast variation by calculating psTW images (right) from the acquired DW image volumes (left).

Figure 3 shows the estimated motion parameters from the interleaved b=0s/mm² images with and without the prospective motion correction based on psTW images. The motion correction was able to reduce the motion to a band of approximately ±0.2mm/±0.2°. In contrast, the graphs without motion correction show the typical long-term motion of the subject. During the scans, there was a significant frequency drift, leading to a corresponding image shift in the phase-encoding direction without motion correction (Fig. 2a/d/g). This component of apparent motion was also removed when scanning with the prospective correction procedure. However, the graphs also demonstrate a low-level fluctuation in motion parameters, which is not evident in the uncorrected data; particularly noticeable in the rotation parameter plots of figures 2c and 2e.

The related trace-weighted images and color FA maps (Fig. 4, proband two) indicate an improvement of overall image quality due to the enabled prospective motion correction.

The fluctuations in the rotational motion parameters are not fully understood. The motion detection estimates of previous work showed smoother curves using the psTW images, which implies a systematic error in the correction of the imaging slices.

The temporal resolution of the motion correction was lowered by the interleaved b=0s/mm² images, which were used as gold standard reference. Without these, the temporal resolution of the method is only affected by the moving window of three averaged DW images. The update rate is the same as the temporal resolution of the measurement.

The motion detection using intermediate, pseudo-trace-weighted images was successfully transferred to a prospective motion correction using the motion parameters to correct slice position and orientation of ongoing scans. The long-term subject motion was reduced to a band of approximately ±0.2mm for translations and ±0.2° for rotations, shown by the gold standard reference, without the need of increasing the total scan time or changing the set of diffusion-gradient directions. The method increases the image quality of calculated parameter maps and fiber-tracking analyses.