Adaptive Linear Predictor Tracks Implanted Radiopaque Markers

Abstract

Radiopaque markers implanted in the heart's wall permit one to observe absolute motion of the heart in closed-chest animals and humans using X-rays. The resulting data, however, are very tedious to analyze because human operators must digitize the observed marker motion from cineradiographic films (or stored video images). This paper describes an algorithm that tracks the trajectories of markers implanted in the endocardium. The algorithm uses a least-mean-squares adaptive linear predictor. By solving the marker tracking problem, one moves closer to a fully automated endocardial marker data acquisition system. The algorithm is also used to detect errors in data acquired using manual digitizing procedures. Tests run with data from nine dogs showed that the algorithm could successfully track and identify 95-100 percent of the markers in 40 out of the 60 film segments and 90-100 percent of the markers in 52 out of 60. The root-mean-square distance between the identified marker trajectories and the manually determined trajectories, a measure of accuracy, was about one-half the marker size when averaged over all of the film segments tested.