Adaptive segmentation of coronary angiograms
D.P. Kottke, Ying Sun · 2003
The problem of unsupervised segmentation of coronary angiograms is investigated. An algorithm which uses an iterative line search procedure to adapt the segmentation thresholds is proposed. This algorithm is parallel in nature, and could be implemented on a disturbed computer architecture. Based on the angiogram's histogram, two thresholds are obtained which classify the pixels into three types: artery, background, and unclassified, The threshold adaptation is an iterative process. A heuristic line search is conducted throughout the neighborhoods of the unclassified pixels. The results of the search are used to adapt the thresholds. The process is continued until all pixels are classified as either artery or background. The algorithm was implemented on an IBM PC/AT-based imaging system and tested with coronary arteriogram images. Preliminary results demonstrated the algorithm's usefulness in enhancing the arterial structure, even under low signal-to-noise ratios. Furthermore, the segmentation was achieved within a few iterations.>