Correction of local deformations in fMRI by 3D non-linear warping in map-slice-to-volume approach

Zntroduction Recently, we have demonstrilted the capability of retrospectively mapping a slice image into rin anatomical reference volume (i.e. map-slice-to-volume, MSV) from the same subject using a simple rigid body transformation of $x degrees of freedom (DOF) [l]. In practice, deformation in a sliQe image is often presented as a nonlinear warping problem that his not limited to the imaging plane. Out-of-plane warp is typically encountered in multislice fMRI data due to the local field induced deformations or localized outof-plane motion artifacts. This study presents correction of the 3D warping artifacts localized in the regions that undergo local changes that result in spatial displacement of voxels typically observed in single shot ac@isitions [2]. The local warping solution is implemented follolwing the rigid body MSV transform that determines the patient’s~ rigid head motion causing in-plane and out-of-plane displacemend, Methods A slice image from time series EPI data was mapped into a high resolution reference yolume space, T1-weighted MRI volume data using a registration algorithm implementing either affine or a control pbint based thin-plate-spline (TPS) function. The automated registration is driven by the mutual information (Z) computed from the gray values of the image pair. By allowing the optimizer to move the control points in the homologok 3D space to positions that minimize the cost metric @&=-I), the iterative process determines the transformation coefficients for a coordinate mapping. For each iteration the affine or TPS warp transformation performs trilinear interpolation in which the original voxel grey values are used. This approach ensures that there is no undesired intensity change from interpolation effects such as round-off or smoothing across iterations. Results Retrospective image registration was performed to map a GREEPI slice image into a volumetric SPGR MRI data acquired from a 3T MRI system. The tota?ly automatic registration routine for each fMRI slice can be initiaipd using an initial three control point set, i.e., rigid body MSV, atid then can automatically increment DOF to multiple control point TPS warping transforms. The six DOF rotate/translate transfoF determines the spatial slice profile relative to the volumetric refqrence data at the time of acquisition. This process accommodates’ the patient’s head movement [l]. The following warping solution is initiated by using the optimized rigid body model to map the control points (> 4 pts) in the reference space into the homologous space. The TPS warping is subsequently computed, agati, by optimizing MI through iterative movement of the control points in homologous space. Improvement in registration using TPS warping solution is demonstrated in Fig. 1 as, compared to rigid-body in l(a). Deformatiotis in the center and the posterior edge are not removed by rigid body and the improved mapping with TPS solution is also indicated by lower MI index. Note that the final mapping of an acquired slice image into the-anatomical reference space is achieved through ?D warping, that corrects out-of-pime as well as in-plane local deformations besides the patient head motion.