High resolution segmented diffusion weighted imaging (DWI) can profit from prospective motion correction using external tracking (1). However, the use of tracking markers may cause inadequate correction and even additional motion artifacts. In this work, real-time volumetric registration (PACE, 2) is used to detect and correct for residual errors during prospective motion correction.

All experiments were conducted on a 3T Tim Trio Siemens system. External tracking was performed using an in-bore camera system (Metria Innovation Inc. Milwaukee, USA, 3) and provides the transformation matrix A_XPACE. The Siemens PACE algorithm provides real-time volumetric registration, and was modified to provide feedback for scans without diffusion weighting only (transformation matrix A_PACE, representing an error term). Position information from both sources was combined (A_final):

A_final = A_XPACE * A_PACE^-1 [1]

This represents a re-initialization of the marker position relative to the object.

A spin echo DWI sequence with segmented EPI readout was modified as follows:

- External position information is used to continuously update the imaging volume according to the volunteer’s position.

- Multiple b = 0s/mm² scans (b0) are distributed evenly through the measurement.

- During the measurement, each b0-volume is reconstructed and registered towards the first b0-volume.

- Relative position information from this registration is applied in addition to external tracking data (as per Eq.[1]) to update the orientation of subsequent scans in real-time. When a selected motion threshold is detected by the volumetric registration (A_PACE), a subset of the measurement is repeated.

- A maximal number of repetitions are enforced to ensure acceptable scan duration.

While the b0 images can be reconstructed online, image reconstruction of the segmented diffusion weighted data was performed offline using MUSE (4) to correct for phase differences after diffusion weighting. In addition to a higher signal to noise ratio, the segmentation of the EPI readout has the advantage of substantially reducing distortions which is beneficial for the PACE algorithm.

A phantom experiment was conducted to display the new functionality. During the DWI acquisition the phantom was first moved to a new position. Subsequently, external tracking was altered by introducing position information from a second marker, creating intentional position errors.

Three in vivo acquisitions were performed:

a) No motion, for comparison

b) Motion, without correction

c) Motion, with the new combined prospective correction

A custom made mouthpiece was used for marker fixation. In addition to the head motion performed in experiment (c), a volunteer was instructed to remove and put back the mouthpiece in the middle of the acquisition to introduce a marker shift.

Sequence parameters: TR/TE: 4000/77ms, matrix: 128², 2 segments, PF 6/8, FOV: 192mm, 33 slices (2mm, 100% gap), DW: 1000s/mm², 30 directions, 32 channel coil.

Figure 1 a-d displays 4 time points of a single slice of the b0 volume acquired during the phantom experiment. The first image (1a) shows the phantom in the original position. In (1b) the phantom was moved manually. This movement was prospectively corrected using data from the tracking system. The movement of the small air bubble at the top of the phantom displays the changing orientation. The third time-point (1c and c*) shows the acquired images after switching tracking to the second marker, including the position offset introduced (1c) and the volume re-aligned with PACE (1c*). In (1d) the prospective correction data was adapted to the intentionally modified marker position (as per Eq.[1]).

Figure 2 shows diffusion tensor data (color maps) acquired during the in vivo experiments. Figure 2a shows the dataset without motion; in (2b) head movement leads to a substantial loss in data quality. (2c) When the new combined prospective correction approach is used, data quality can be restored despite the shift in marker position.

Figure 3 shows the positional differences detected by the PACE algorithm during the in vivo scan. The peak at the third time-point (1.5mm and 0.5deg) indicates the removal of the mouthpiece.We demonstrate that intrinsic volumetric information can be used to detect and correct errors in external tracking during prospective correction of motion during DWI. Another potential application for the combination of XPACE and PACE are fMRI time series. Of note, the limited accuracy of PACE navigators seems to be sufficient to correct errors in the relatively low resolution (>1.5mm) of routine DWI and fMRI scans. We will further explore whether the method can be applied to high resolution structural scans using other navigator methods.