IMPROVING RIGID HEAD MOTION CORRECTION USING PARALLEL IMAGING

点击翻译按钮获取中文摘要

This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. The subproject and investigator (PI) may have received primary funding from another NIH source, and thus could be represented in other CRISP entries. The institution listed is for the Center, which is not necessarily the institution for the investigator. Introduction. Involuntary patient motion is still a great challenge in MRI. Specifically, in the elderly and pediatric patient population or in patients whose medical conditions (tremor, seizure, stroke) preclude them to hold still, effective strategies to compensate for motion are paramount. In this study, a variant of parallel imaging is introduced that can correct k-space inconsistencies arising from rigid body motion (rotation or translation). This motion correction scheme first identifies the degree of motion, corrects the k-space data accordingly and thereafter employs an augmented conjugate gradient based iterative image reconstruction to synthesize missing data in k-space. The method is described and verified in simulated interleaved EPI and spiral images scans as well as in vivo using bi-density spiral scanning. Materials and Methods: Reconstruction  Generally, an object rotation in image space is paralleled by a similar rotation of k-space data, whereas translations are reflected by linear phase rolls. If these motion components are known, k-space data can be corrected for but usually leading to a fragmentation of k-space. This, in turn, gives rise to significant ghost artifacts in the final image. Our correction builds upon an augmented version of an iterative SENSE reconstruction1 and is performed as follows: 1) counter-rotating k-space data by applying the corresponding rotation matrix to the k-space trajectory coordinate points of each profile/interleave prior to gridding. 2) Rotating the coil sensitivity map that enters the encoding matrix E1 for each profile/interleave. This rotation is necessary because even if the object is rotated back to its desired position, different regions of the object have been exposed to different coil sensitivities during the acquisition. 3) Correcting the altered sampling density after rotation. In this study, Voronoi tessellation has been used to derive the new sampling density from the rotated k-space trajectories. 4) Phasing the data to account for translation by applying the correction term pcorr(?) = exp{-j(2??x/FOVx) (kx(?)/[kx,max-kx,min])  j(2??y/FOVy)(ky(?)/[ky,max-ky,min])} to the original k-space data prior to gridding. Motion detection  Various methods exist to derive the extent of translational and rotational motion from MR data. In this study, the motion information was extracted from navigator echoes. The navigator information can be derived from the scan trajectory itself (i.e. self-navigating trajectories) or alternatively from a separate acquisition that provides a low resolution image. Here, a multi-grid registration approach was used that finds the maximum Pearson correlation between a reference image and individual navigator images and provided a reliable estimate of the amount of rotation and translation relative to the reference image (average over all images). To increase robustness and to improve the accuracy of co-registration this step was repeated at least twice. Experiments  Synthetic data for interleaved spiral and EPI acquisitions (8 interleaves) were generated by using inverse gridding operations2 on a motion corrupted phantom. For each of the eight interleaves a random head rotation (range ¿30¿) and translation (range ¿15mm) was generated. Prior to the inverse gridding step, each of the individually rotated and shifted images were multiplied by coil sensitivities simulating receiver coil sensitivities from six coils that were attached around the circumference of the object. In vivo validation was performed in 3 healthy volunteers using T2w spin echo scans with an interleaved spiral-in/spiral-out readout and an 8-channel head coil. The spiral-in part (3-5ms duration) provided for each interleaf data a low resolution navigator image (322). The spiral-out part was a normal interleaved spiral acquisition: TR/TE=4,000ms/85ms, slice thickness/ gap=4/1mm, 17 slices, FOV=24cm, matrix=256, interleaves = 32, and NEX=1. The receiver bandwidth for the spiral acquisition was +/- 125kHz. During each experiment the volunteers were asked to rotate and/or shift their heads at three increasing levels of motion (no, mild [~¿15¿], and moderate [~¿25¿] motion ). References: 1Pruessmann K, et al. MRM 46: 638-51, 2001; 2Rasche V, et al. IEEE TMI 18: 385-92, 1999. Acknowledgements: This work was supported in part by the NIH (1R01EB002771), the Center of Advanced MR Technology at Stanford (P41RR09784), Lucas Foundation.