Analysis and visualization of 4D medical images using self-organizing maps and clustering
|Autoren|| Petra Hobelsberger|
|Editoren|| F. Leberl|
|Titel||Analysis and visualization of 4D medical images using self-organizing maps and clustering|
|Buchtitel||Vision with Non-Traditional Sensors - Proc. 26th Workshop of the Austrian Association for Pattern Recognition (ÖAGM/AAPR)|
Segmentation of high resolution 4D images occurring in medical examinations is a very time consuming and troublesome task. We have applied a hierarchical approach, consisting of self-organizing maps (SOMs) in combination with different clustering algorithms, to the problem of tracking the heart muscle on a series of 3D CT-images taken from the beating heart.In this paper we show how SOMs can be used to create a initial segmentation of a single 3D image, and how this segmentation can be propagated to subsequent time steps. The abstraction introduced by the representation through a self-organizing map can further be used to speed up image segmentation and help to simplify user interaction, as all results can be propagated from one time step to another. Preprocessing such as compression is applied to additionally increase computation speed.