Tag point classification in tagged cardiac MR images

This paper presents a tag point classification algorithm for use in a new technique for tracking tag lines in tagged cardiac magnetic resonance (MR) images. Instead of tracking tag lines from frame to frame in an image sequence with active contours or similar techniques, a set of candidate tag points are detected in each image and then classified as either a false positive or belonging to a particular tag line. The advantage of this approach is that the tag point positions are not pre-smoothed during tracking, allowing smoothness constraint to be applied only in the deformation model fit to the tag points. Results of a preliminary validation experiment on human cardiac MR data are presented that show a classification accuracy of 97.86%