Skip to Main Content
This paper characterizes factors affecting the accuracy of the correlation coefficient (CC) template matching algorithm, as applied to motion tracking from two-dimensional real-time coronary artery magnetic resonance images. The performance of this algorithm is analyzed in the presence of both random and systematic error. In the presence of random error, it is shown that a necessary and sufficient condition for accurate motion tracking is a large CC difference-to-noise ratio (CCDNR). The CCDNR itself is in turn affected by five factors: image and template size, image and template structure, and the magnitude of the noise. Techniques are introduced for manipulating some of these factors in order to increase the CCDNR for greater motion tracking accuracy. In the presence of superimposed systematic error it is shown that, while large CCDNR is necessary, it alone is not sufficient to ensure accurate motion tracking. Techniques are developed for improving motion tracking accuracy that minimize the effects of systematic error, while maintaining an adequate CCDNR level. The ability of these techniques to improve motion tracking accuracy is demonstrated both in phantoms and in coronary artery images.