Non-parametric orthogonal slice to volume deformable registration: Application to PET/MR respiratory motion compensation

The recent advance of hybrid PET/MR system enables MR-based motion correction of PET data. In this paper, we propose a PET/MR motion compensation (MC) strategy and a non-parametric orthogonal slice to volume deformable registration technique for respiratory motion estimation from the MRIs. In our imaging strategy, a static 3D MRI is acquired before the PET acquisition, and a series of dynamic 2D MRIs are acquired in sagittal and coronal planes by turns during the PET acquisition. The 2D MRIs in orthogonal orientations are then registered with the static 3D MRI to derive a 3D+t deformation field for PET MC. Unlike most previously reported works in the literature, our MC strategy does not rely on respiratory gating, and therefore is able to address irregular and/or varying breathing patterns. We validated our approach using MRI data acquired from 5 volunteers and synthetic simultaneous PET data, demonstrating up to 30.5% improvement in registration accuracy and sharper motion corrected PET images over the reported MC approaches.