Spatially coherent clustering applied to cardiac angiograms

Digitized X-ray angiograms of the left ventricle were clustered using a weighted Euclidean distance criterion with the constraint that the pixels in a cluster are spatially coherent. A three-dimensional feature space was used: position along x and y coordinates and image intensity. The spatial coherence constraint is based upon the belief that pixels within a cluster should be coherent in a spatial as well as an intensity sense. The results indicate that the clusters separate the left ventricle from other anatomical objects in the image and that this approach should be useful in cardiac volumetric analysis and wall-motion studies.