Fast Helical-Spiral Spherical Navigator Echoes for 3-D Motion Detection and Inter-Scan Alignment

We have employed a rapid (< 1⁄4 sec) spherical navigator echo (SNAV) to prospectively modify image acquisitions in serial MRI examinations. Detected motions using the faster helical-spiral SNAV and a computer-controlled phantom are now of comparable accuracy to previous results using a slower SNAV comprised of many circular trajectories. The faster SNAV also performs well in a pre-scan registration application for aligning image volumes acquired on different scanners at different times. Correlations of base and SNAV-aligned image volumes were comparable to or exceeded results achieved using a retrospective image registration algorithm. Methods The helical-spiral SNAV consists of two sequential acquisitions taking < 1⁄4 sec in total. As depicted in Fig. 1, each echo begins at the “equator” and moves towards the “pole” of the k-space sphere. It stops after sampling 85% of the hemisphere’s area in order to avoid exceeding slew rate limitations. A second echo samples the other hemisphere providing a total of more than 2,000 samples. Compared to another version of the SNAV (1-2) comprised of 128 orbital echoes, the helical-spiral SNAV is much faster, but it has less than 14% the total number of samples of the slower SNAV. The faster SNAV was tested in two applications: (i) detection of motions executed by a computer-controlled phantom and (ii) inter-scan registration to align image volumes collected of an object on different magnets. SNAVs of the computer-controlled phantom were collected at 7 positions (baseline, 2 sagittal rotations, 2 axial rotations, and 2 z translations). This allowed for 49 SNAV registrations, between every pair of SNAV data sets, including some compound rotation or rotation + translation cross comparisons. In the inter-scan registration experiments, repeated 25 slice volumes with 5mm thickness and 0 mm spacing were collected along with SNAV data on a clinical scanner. The phantom object was then reimaged on a different scanner at 3 distinct positions: (i) re-landmarked as best as possible (ii) intentional small compound motion (iii) intentional large compound motion. Unaligned and SNAV-aligned volumes were collected on the second scanner. The unaligned and SNAV-aligned image volumes were also retrospectively aligned with a 3-D rigid-body, mutual-information-based image registration algorithm. 3-D correlation coefficient measurements were computed for the repeated scan, unaligned and SNAV-aligned volumes, and outputs from the image registration algorithm. Results Table 1 summarizes the motions detected by the faster SNAV in the computercontrolled phantom experiment. The intended motions were nominally 0, 3, or 5 degrees or millimeters. The magnitude of detected motion in cross comparisons between positions should therefore be 0, 2, 3, or 5. All of the detected motions are within the expected known error of the computer-controlled phantom, ±0.13 ̊ on sagittal rotations, ±0.5 ̊ on axial rotations, and ±0.2 mm for z translations (1). Table 2 shows the 3-D correlation coefficient values for the inter-scanner alignment experimental data as well as with the outputs of the registration algorithm. SNAV alignment increases the correlation in all cases, and in the cases of small and large motion for the phantom, SNAV alignment outperforms retrospective image based registration. Also, image based registration performed after SNAV alignment always achieved higher correlations than either technique alone. DiscussionThe approach described should greatly facilitate radiological interpretation of serialMRI studies. An imaging sequence can be acquired in near perfect spatial alignmentwith its earlier counterpart with a negligible time penalty (< 1⁄4 sec). The faster SNAVperforms well in the inter-scan alignment application with results comparable toretrospective image registration. In vivo experiments are underway, and future workwill address questions of how small anatomical changes affect performance in theinter-scan alignment application. The fast SNAV could also potentially be used forprospective correction in fMRI or for interview motion correction. References1. Welch EB, Manduca A, Grimm RC, Ward HA, Jack Jr. CR. Sphericalnavigator echoes for full 3D rigid body motion measurement in MRI. MagnReson Med 2002;47:32-41. 2. Welch EB, Manduca A, Grimm RC, Ward HA, Jack Jr. CR. Automatic inter-exam image alignment using spherical navigator echoes. In: Proceedings ofthe 10 annual meeting of ISMRM, Honolulu, Hawaii, 2002. p. 377. AcknowledgementsThis research was supported by NIH grant AG19142.Fig. 1 The SNAV’s helical-spiraltrajectory. More than 2,000samples are collected by two successive hemispherical echoes. Table 1. Detected motions with computer-controlled phantom NominalMotionSagittalRotation ( ̊)AxialRotation ( ̊)Z Translation(mm) 0.0 (n=27) 0.0±0.05 0.1±0.110.1±0.092.0 (n=2) 1.9±0.00 1.5±0.001.9±0.073.0 (n=10) 3.2±0.06 3.0±0.122.8±0.115.0 (n=10) 5.2±0.08 4.5±0.114.7±0.16 Fig. 2 Absolute difference images for a selected central slice ofthe phantom image volume. The difference between the slice from the base image volume and the corresponding slice in theunaligned large-compound-motion volume is clearly visualized(a). After alignment using the detected misregistrations fromthe helical-spiral SNAV, the absolute difference is reduced (b). Table 2. 3-D Correlation Coefficient Values UnalignedUnalignedRegisteredSNAVAlignedSNAVRegistered Re-landmark 0.916 0.997 0.996 0.997Small Motion 0.903 0.990 0.994 0.998Large Motion 0.825 0.969 0.986 0.996ab 1056Proc. Intl. Soc. Mag. Reson. Med. 11 (2003)