Noise reduction in estimating cardiac deformation from marker tracks.

Deformation of the cardiac wall is measured by using optical or radiopaque markers attached to the wall. When digitized images are used, the accuracy of the measurement of a marker position is limited by pixel resolution and the size of the marker. The spatial accuracy is improved by singular value decomposition (SVD) filtering. This filtering procedure is based on the assumption that displacements of markers are mutually related because they are embedded in a common continuum. In a computer stimulation with 48 markers in 51 video frames, the accuracy of the measurement of a marker position improved from 0.14 to 0.045 (SD) pixel. In an open-chest animal experiment, with markers on the surface of the heart, the deformation patterns were extracted more clearly using SVD filtering, while mutually related high-frequency components were not suppressed. In a 50-frame sequence of 256 X 256 video images of a 45 mm X 35 mm deforming surface with 50 markers of 8 pixels in diameter, the marker position resolution improves from 0.1 to 0.03 (SD) pixel (6 microns). Strain is determined with an accuracy of 0.002 over a distance of 30 pixels (6 mm).