3D Spherical Segmentation and Registration Toolkit

"Current algorithms work on a high amount of matching elements from both datasets. Because of this a denser (more specific form of) information should be used, representing characteristic information of both datasets. The aim of this project was to match different CT coronary datasets taken from different patients and to registrate those coronary CT scans with a reference coronary dataset. Usually this has to be done by an assistant or a doctor manually to improve the comparison of datasets e. g. finding an unhealthy increase of heart muscles, finding the transformation for PETs taken at the same time as the CT or speeding up the diagnosis by segmenting and registering known areas. Blood cavities seem to be the most characteristic information if coronary CT datasets have to be registered. Searching for a good representation for blood vessels, spheres seem to be applicable because blood vessels in human bodies appear in a tubular form which can be represented by placing spheres along a line following the centers of the blood vessel. Possible problems like creation of the graph edges and disjointed blood vessels at bifurcations have to be solved to create a representative data. After this step the blood vessels can be represented by joining the volumes of the spheres along the line. Using this line of spheres, the blood vessels are represented by a very dense amount of graph edges and nodes. The graph nodes of both graphs are then used to create and use a matrix, projecting one dataset to the reference dataset. The important part is to find efficient algorithms for the computation of the underlying graph as well as a fast determination of the best matrix projecting one CT scan to the reference CT scan."

Matching in 2D - Left: Reference dataset, Right: Destination dataset