Integrated segmentation and non-linear registration for organ segmentation and motion field estimation in 4D CT data

A. Schmidt-Richberg*, H. Handels, J. Ehrhardt

*Corresponding author for this work
23 Citations (Scopus)


Objectives: The development of spatiotemporal tomographic imaging techniques allows the application of novel techniques for diagnosis and therapy in the medical routine. However, in consequence to the increasing amount of image data automatic methods for segmentation and motion estimation are required. In adaptive radiation therapy, registration techniques are used for the estimation of respiration-induced motion of pre-segmented organs. In this paper, a variational approach for the simultaneous computation of segmentations and a dense non-linear registration of the 3D images of the sequence is presented. Methods: In the presented approach, a variational region-based level set segmentation of the structures of interest is combined with a diffusive registration of the spatial images of the sequence. We integrate both parts by defining a new energy term, which allows us to incorporate mutual prior information in order to improve the segmentation as well as the registration quality. Results: The presented approach was utilized for the segmentation of the liver and the simultaneous estimation of its respirationinduced motion based on four-dimensional thoracic CT images. For the considered patients, we were able to improve the results of the segmentation and the motion estimation, compared to the conventional uncoupled methods. Conclusions: Applied in the field of radiation therapy of thoracic tumors, the presented integrated approach turns out to be useful for simultaneous segmentation and registration by improving the results compared to the application of the methods independently.

Original languageEnglish
JournalMethods of Information in Medicine
Issue number4
Pages (from-to)344-349
Number of pages6
Publication statusPublished - 01.12.2009


Dive into the research topics of 'Integrated segmentation and non-linear registration for organ segmentation and motion field estimation in 4D CT data'. Together they form a unique fingerprint.

Cite this