A relaxation algorithm for segmentation of the endocardial surface from cine CT

A relaxation algorithm has been developed for automated segmentation of the endocardial surface from contrast cine-CT (computer tomography) images. The image is contoured at an initial density threshold, and a one-dimensional edge operator is applied orthogonally to each point of the contour. Output from the operator is used to generate a histogram, the mode of which identifies a new threshold. The image is contoured again at the new threshold and the process is repeated. Iteration continues with successive threshold estimates converging to a stable value in the region of the endocardial surface. Computer-determined thresholds compare favorably with manual segmentation while reducing processing time and increasing reproducibility