The growing use of digital medical imaging gives rise to the question of data archiving and retrieval. Dedicated applications, i.e. telemedicine and image exchange impose severe demands on networking and storage resources. This bottleneck can be moderated using content-orientated image compression which reduces the amount of data while preserving the diagnostic information. This paper presents an new method for automated vessel segmentation which can be used for both, decision support systems and content-orientated image compression of angiograms. Some vasculature detection approaches use differentiating operators which are problematic since they are not immune to the usually high noise level in angiograms. Histogram segmentation ignore most of the knowledge of vessel geometry. Tracking approaches with matched filters are favored here, because they do not possess the above mentioned disadvantages. However, they depend largely on exact signal and noise models, which has not been exhaustively investigated so far. In this article, an accurate vascular model is derived from angiographic imaging equations. The anatomical knowledge of blood vessels can be better exploited using a Kalman filter, which uses an explicit vessel model. This optimizing filter has an additional advantage: Some heuristic thresholds of the tracking process can be optimized with respect to sample data. A further problem is the evaluation of the detected vasculature, which is of prime importance in medical applications. Here, a new automated structural evaluation method of vessel trees by means of graph matching is presented. Tests of the new method showed very good results, also very small and hardly recognizable vessels have been detected. The structural check shows that some errors like oversegmentations can be discovered using structural knowledge.
展开▼