Phase-Driven Finite Element Model for Spatio-temporal Tracking in Cardiac Tagged MRI

MRI tissue tagging[1] of the heart produces noninvasive markers within the muscle wall that can be used to measure motion and deformation (Fig. 1a). However, the widespread use of tissue tagging has been limited by time-consuming image post-processing. This paper outlines a method for automatic 3D tracking of LV motion from 2D MRI images acquired from multiple views. Since the tags form a repetitive pattern in the images, the local phase can serve as a material property that can be tracked. We first derive displacement from local image phase, a quantity previously used for estimating disparity between two 2D images in stereo vision [2, 8] or to measure relative deformation, or strain [7]. Displacement estimates have also been used for least square fitting of a 2D affine motion model to angiography images [5]. In the current work, we use phase-based displacement estimates to drive a 3D finite element deformable model of the left ventricle (LV). Our model enables reconstruction of 3D motion from 2D images and stabilization of the phase-derived displacement estimates, which are susceptible to noise and image artifacts. The basic framework for this model has been previously applied to 3D motion reconstruction of the right ventricles from user tracked tags [4, 3]. We here describe the geometry and constraints that enable phase-based displacement estimates to be used for automatically recovering 3D motion from tagged MR images.