MULTIVIEW REGISTRATION OF CARDIAC TAGGING MRI IMAGES

This paper introduces a new method based on k-nearest neighbors graphs (KNNG) for bringing into alignment multiple views of the same scene acquired at two different time points. This framework is applied to cardiac motion estimation from tagging MRI sequences. Features acquired in each view are collected in a high dimensional feature space and an efficient estimator of alpha-joint entropy (alphaJE) is used for selecting the optimal alignment. In order to register 4D datasets, an analytical expression of the alphaJE estimator was derived, enabling a fast implementation of gradient based optimization. The technique was tested in a set of six sequences and the results compared with respect to manual measurements made at tag crossing points, obtaining good accuracy and low processing times compared to published state of the art methods