Cookie Settings

We use cookies to optimize our website. These include cookies that are necessary for the operation of the site, as well as those that are only used for anonymous statistic. You can decide for yourself which categories you want to allow. Further information can be found in our data privacy protection .

Essential

These cookies are necessary to run the core functionalities of this website and cannot be disabled.

Name Webedition CMS
Purpose This cookie is required by the CMS (Content Management System) Webedition for the system to function correctly. Typically, this cookie is deleted when the browser is closed.
Name econda
Purpose Session cookie emos_jcsid for the web analysis software econda. This runs in the “anonymized measurement” mode. There is no personal reference. As soon as the user leaves the site, tracking is ended and all data in the browser are automatically deleted.
Statistics

These cookies help us understand how visitors interact with our website by collecting and analyzing information anonymously. Depending on the tool, one or more cookies are set by the provider.

Name econda
Purpose Statistics
External media

Content from external media platforms is blocked by default. If cookies from external media are accepted, access to this content no longer requires manual consent.

Name YouTube
Purpose Show YouTube content
Name Twitter
Purpose activate Twitter Feeds

In most patients deformations are known to occur during the treatment course. To detect and extract such deformations, image registration needs to be able to describe deforming anatomy inside its transformation model. Such non-rigid image registration methods are also called deformable image registration (DIR). Non-rigid transformations can be described mathematically using various formulations. Therefore various DIR methods can be realized.

Interactive landmark-based DIR

Similar to interactive landmark-based rigid registration method (→ LM rigid), the registered images can be deformed guided by user interaction. Selecting pairs of anatomical landmarks inside both image scans, the user defines supporting points for a transformation described by a superposition of radial basis functions. Different types of bending potentials can be used to construct specific radial basis functions. Bookstein et. al. [1] introduced the now-a-days frequently used thin plate splines (TPS) bending potential to describe image deformations.

This implemented module allows the user to select landmark pairs iteratively while the corresponding transformed image is updated and rendered using the red-green-fusion mode. Since the interactively used registration approach needs to be fast to allow a flowing processing for the user, this module is implemented using hardware-acceleration on the graphic processing unit (GPU).

© dkfz.de

Template Matcher Approach

This fully automated DIR approach consists of different sub-steps. First, small promising image regions (templates) are selected, which hold heterogeneous and unique structures, as candidates for automatic rigid matching in their neighborhood. If a clear increase in the local similarity measure is achieved by optimizing the position of each chosen template, its center and translation are used as supporting vector for the global transformation using thin plate splines.

© dkfz.de

GPU-accelerated ChainMail Registration

The more image point correspondences can be found in an image pair the more precise will be the description of the occurred deformations. Yet, the more supporting vectors are to be processed in the global interpolation scheme, the more time consuming is the solving of equations and worse conditioned the system of equations to solve.

To overcome this problematic relationship, we developed an alternative local transformation model to propagate the local displacements of the found correspondences into its neighborhood [5]. Based on the idea of the 3D Chainmail algorithm, initially developed by S.F.F. Gibson [2], this method, additionally, allows even to model deformations of heterogeneous tissues.

In the first step, comparable to the template matching strategy, small templates are selected and repositioned rigidly until a position with local optimal similarity measure is found. These correspondences are then used to fit a global rigid transformation to account for global shifts and rotations in the anatomy of the patient. Then the remaining vectors, which describe the deformational component, are used as input for the 3D chainmail propagation step. The used tissue parameters in the 3D chainmail dictate how and how far the local deformations are propagated inside the adjacent volume.

To be able to propagate tens of thousands of correspondences the data-parallelizable calculations are implemented to run on GPU hardware. This way, it is possible to reduce the registration duration for typical data sets in radiation therapy treatments to far less than one minute using today’s standard graphics cards (e.g. NVIDIA GeForce Series).

References

[1] Bookstein F (1989) "Principal Warps: Thin-Plate Splines and the Decomposition of Deformations" IEEE Trans.Pattern Anal.Mach.Intell.

[2] Gibson SFF (1996) 3D ChainMail: a Fast Algorithm for Deforming Volumetric Objects. Technical report (TR96-22) Mitsubishi Electric Research Laboratories

Publications

[3] Malsch U, Thieke C, Huber P E, Bendl R (2006) "An enhanced block matching algorithm for fast elastic registration in adaptive radiotherapy" Phys. Med. Biol.

[4] Giske K, Stoiber EM, and Bendl R (2011) GPU based parallelization of deformable registration for dynamic adaptations in radiation therapy. Radiotherapy & Oncology (ESTRO conference)

[5] Giske K (2011) GPU basierte Parallelisierung von deformierbaren Registrierungsverfahren als Grundlage für dynamische Anpassungen von Therapieplänen in der adaptiven Strahlentherapie. Dissertationsschrift, Ruprecht-Karls-Universität Heidelberg

to top
powered by webEdition CMS