Deformation tracking by multi-resolution gradient descent method

摘要

The goal of image registration is to determine a spatial transformation that will bring into correspondence (i.e. register) pairs of homologous points in two given images. We study the deformation tracking of the myocardiac muscle in 2D ICE images, which are always perturbed by a strong "speckle" noise. Speckle noise is non-Gaussian with a high noise-to-signal ratio, which is usually assumed to be decorrelated from frame to frame and for different pixels within a single frame. We reconstruct dynamic myocardial deformations through elastic registration of sequences of ICE images, by minimization of a cost functional constructed by stochastic modeling of speckle noise.