A fast skeleton finder for parallel hardware

The authors describe a fast skeleton finder for artery trees in coronary angiograms. Based on the skeleton finder proposed by T. Nguyen et al. (1986), a parallel version was developed for a multi-transputer cluster. The aim is the 3-D reconstruction of the coronary artery tree from biplane angiograms using artificial neural networks. For the neural network approach an algorithm for feature extraction was selected and modified, which can extract the main features of the artery free segments such as location, width, intensity, and direction. This algorithm overcomes user interaction, runs in parallel and is robust to artifacts resulting from background structures. The algorithm can extract the main features of the artery tree in less than 16 seconds on a cluster of ten transputers