Publikationsserver der RWTH Aachen University
Generalisierte Bildrekonstruktion für die Magnetresonanz-Bildgebung
Abstract
dc:descriptionHardware development for modern magnetic resonance imaging is reaching technological and physiological limits. Further acceleration of measurement experiments, as well as special methods for the acquisition of additional information, therefore require the use of advanced measurement protocols; these already have an inherently increased computational complexity of image reconstruction, and are especially sensitive to artifacts. Correction of multiple artifact sources using a single image reoncstruction method is impossible so far, especially in combination with so-called sensitivity encoding, which allows additional decreases in imaging time. In this work a generalized image reconstruction is proposed, which for the first time allows a simultaneous correction of multiple artifact sources in magnetic resonance imaging, including the combination with sensitivity encoding. Algorithms are developed which permit the computation of the generalized image reconstruction in acceptable time with modest requirements to the computation hardware. To this end, the iterative method of conjugate gradients for the solution of very large linear systems of equations is used. Further acceleration is achieved by discretization of acquired artifact parameters, and the use of the fast Fourier transform, as well as the parallelization of the method on various levels of implementation. The latter ist greatly facilitated by the highly regular structure of the base operations of the the generalized image reconstruction. Imaging experiments which sample the spatial frequency spectrum on an irregular, rather than on a conventional cartesian grid, can also be accelerated by introducing a two-dimensional variant of the fast Fourier transform for non-cartesian data sets. The generalized image reconstruction method was extended by a regularization of the underlying linear system of equations for the treatment of underdetermined or ill-posed image reconstruction problems. Simulations and real magnetic resonance experiments show, that the generalized image reconstruction works very satisfactory over a wide range of artifact levels, and is stable in the presence of noise. Even the most general variant can be calculated on standard computation hardware with moderate memory storage of a few megabytes in under one hour. Special cases of the generalized image reconstruction can already be calculated in very short time. One example with real data from a real-time magnetic resonance system shows the practical applicability of the method with currently unsurpassed reconstruction times of under one second. The apparent image quality of the reconstruction is generally very good. The normalized relative error of the reconstruction compared with an undistorted reference image can be reduced by at least a factor of 2. Even strong ghosting artifacts can be almost completely eliminated. Remaining artifacts caused by the reconstruction itself can be suppressed effectively by regularization. In summary, the generalized image reconstruction presents itself as a versatile method capable of replacing a whole palette of special image reconstruction and correction methods, and which opens up new possibilities for magnetic resonance imaging by combining the correction of multiple artifact sources. Performing the method is already possible today using standard computation hardware. The generalized image reconstruction is additionally suitable for implementation on dedicated computation hardware because of its inherent ability to be parallelized. It is therefore to be expected that it will become practical to use it in the near future.
Degree
thesis:*- Grantor dc:publisher
- Publikationsserver der RWTH Aachen University
- Year dc:date
- 2003
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kannengießer, Stephan
- Contributors dc:contributor
-
- Noll, Tobias G.
Subjects
dc:subject × 8Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- ger
Identifiers
dc:identifier.*- OAI identifier oai:identifier
- oai:publications.rwth-aachen.de:61970