Tubular structures

The segmentation of tubular structures from 3D medical image data is the essential basis for many computer-assisted applications such as operation planning and the development of an individualized ventilation strategy. Our project has as its objective the development of a robust segmentation procedure to achieve precise results with few user interactions in a clinically feasible amount of time.

Often the image data are noisy and the vessels are pathologically altered. To achieve reliable segmentation in such data records, we are working on a statistical approach of Bayesian tracking. In this procedure a statistical model is used to iteratively estimate at each particle of the central line of the tubular structures how likely it is for a vessel to be found at this location. The tracking process is then continued recursively for those particles with high probabilities.

A great advantage of this approach is that it permits the integration of prior information and various statistical models to estimate the central line. We have developed a comparability measurement for the tracking of coronary arteries which functions robustly even in pathologically altered vessels. Most of the tracking procedures described in the literature are limited to a single vessel. We are working on being able to detect the bifurcations automatically and thus segment the vascular trees with few user interactions. Furthermore, Bayesian tracking is computationally intensive. An objective of our research is to optimize the algorithm to permit clinical use.

Another very promising approach for vessel segmentation is the gradient vector flow algorithm. The procedure is characterized by its robustness against image noise. In this procedure the energy of a gradient vector field is minimized. The calculated matrix image is then searched for specific eigenvalue ratios that indicate the presence of tubular structures. This information combined with the gray scale image is used to decide whether or not a specific image point belongs to a vessel.

Selected publications

• [Wang2010]: Xin Wang, Ivo Wolf, Philipp Hartmann, Tobias Heimann, Hans-Peter Meinzer, Ingmar Wegner. Ein gradientenflussbasiertes Ähnlichkeitsmaß für das Tracking von Gefäßen in medizinischen Bilddaten In Deserno TM, Handels H, Meinzer HP, Tolxdorff T (eds). Bildverarbeitung für die Medizin 2010. Heidelberg: Springer (2010) 236-240.