To implement and evaluate retrospective motion correction using SNAVs interleaved within a 3D imaging pulse sequence.SNAVs are k-space navigators that can be used to measure 3D rigid-body motion.^1,2 They are faster to acquire than image-based navigators^3-5 and unlike optical tracking,⁶ they do not require additional hardware. The clinical applicability of the preRot-SNAV² technique was limited because it required 26 s of “no motion” during baseline acquisition. We have developed a method to acquire the required baseline in only 2.5s. In this study we use this rapid baseline strategy paired with a novel SNAV interleaved 3D gradient echo pulse sequence to demonstrate retrospective correction of brain images.

SNAV-interleaved imaging sequence. The navigated image sequence is a modified spoiled gradient echo sequence (SPGR-SNAV). This sequence acquires an SNAV after every four Cartesian image lines. The built-in SNAV has a radius of 0.40 cm^-1 and 2508 sample points.

Data acquisition. Three volunteers were each scanned twice using the SPGR-SNAV sequence: once using the single-channel head coil and once with the 8-channel head coil. The volunteers performed step-wise motion; they were instructed to move approximately every 50 s of the 7.5 min scan. Prior to these motion scans the required 2.5s baseline scan was acquired. An additional no-motion reference image was acquired after all motion trials. The sequence parameters used were as follows: matrix size = 256x160x124, TE/TR = 3.9/15 ms; image bandwidth = 62.5 kHz, SNAV bandwidth =125kHz, flip angle = 8°; slice thickness = 1.5mm; field of view = 24x24x18.6 cm.

Motion correction. All 6 rigid-body motion parameters were extracted from 4960 interleaved navigators. To correct the images, phase shifts were first applied in order to correct for the measured translations; next, the 3D coordinates of the phase-corrected data were rotated based on the measured rotation. K-space data were then interpolated at the transformed coordinates using spline interpolation.

Representative results of retrospective motion correction using a single channel birdcage coil are shown in Fig. 1, which compares a reference (no motion) image (Fig. 1a) to a motion-corrupted image (Fig. 1b) and retrospectively motion-corrected images (Fig. 1c). For this experiment the volunteer was asked to rotate their head in both the θ_x and θ_z directions (nodding and axial). The measured motion agrees well with the intended motion; we see step-like rotations about X and Z with accompanying Z and X translations at the time-points of directed motion. The uncorrected image acquired during motion has severe artifacts, as expected. Excellent correction of these artifacts is observed in the corrected image (Fig. 1c). The measured rotations and translations are shown Fig. 2a and 2b respectively.

Representative results of motion correction with another volunteer using an 8-channel head coil are shown in Fig. 3. Once again retrospective motion correction was successful, with the corrected images shown in panel (c) demonstrating fewer motion-related artifacts when compared to the uncorrected images in (b). Substantial correction of in vivo head motion up to 4° and 4 mm was observed in six acquired data sets with 3 volunteers and 2 different receive coils.

This work demonstrates, for the first time, retrospective motion correction in vivo using SNAVs. Nearly motion-artifact-free images were achieved during head rotations of several degrees and corresponding translations of up to 4 mm.

In all cases, the profiles derived from the SNAV motion measurements agree with the intended motion. Occasional jitter, of up to 0.4 mm and 0.5 degrees, is seen in the motion profiles. This jitter, which can likely be reduced with further optimization of the measurement method, did not appear to affect the ability of SNAVs to perform motion correction.

Only retrospective correction has been demonstrated in this study. Prospective motion correction keeps the image coordinate system fixed relative to the object and thus unlike retrospective motion correction avoids gaps in k-space that occur due to object rotation. A small rotation range was used for this study in order to minimize these gaps. A benefit of retrospective correction, however, is that it ensures that the original image is always available. Both methods have advantages and disadvantages; it is likely that the most suitable method will be application dependent. SNAVs are practical for both retrospective and prospective motion correction.

With further optimization and speed-up of the SNAV processing code, we expect that the gradient orientation can be updated within 35 ms of the start of SNAV acquisition. This is faster than the current image-based approaches,^3-5 and makes SNAVs a promising motion correction technique, which could be applied both prospectively and retrospectively, with gradient echo images.