Abstract
Magnetic Resonance Imaging is a measurement method which produces representative data of the insight of human bodies. In this work a retrospective correction method for the inhomogeneity artifact is composed of other approved image processing techniques, extended and evaluated.
The proposed method performs denoising and simultaneous bias estimation by the TGV-L1 Primal-Dual algorithm for volumetric data. Bias correction is done by solving a Poisson Equation via a direct form solution in the cosine domain. The algorithm is implemented for general TGV order and efficiently calculates most operations in parallel.
Several image types are processed including 3D MR measurement data. The results include a quantitative comparison to ground truth data and metric values. Additionally a qualitative evaluation by intensity profile line plots and an estimate of the probability density functions is given.
Under the assumption of piecewise constant objects of interest and a slowly and smoothly varying bias field the proposed method successfully estimates higher-order bias fields. The method outperforms the reference method N4ITK in several aspects and may improve the performance of other imaging tasks, and could be applied to several other imaging modalities.
The proposed method performs denoising and simultaneous bias estimation by the TGV-L1 Primal-Dual algorithm for volumetric data. Bias correction is done by solving a Poisson Equation via a direct form solution in the cosine domain. The algorithm is implemented for general TGV order and efficiently calculates most operations in parallel.
Several image types are processed including 3D MR measurement data. The results include a quantitative comparison to ground truth data and metric values. Additionally a qualitative evaluation by intensity profile line plots and an estimate of the probability density functions is given.
Under the assumption of piecewise constant objects of interest and a slowly and smoothly varying bias field the proposed method successfully estimates higher-order bias fields. The method outperforms the reference method N4ITK in several aspects and may improve the performance of other imaging tasks, and could be applied to several other imaging modalities.
Originalsprache | englisch |
---|---|
Qualifikation | Master of Science |
Gradverleihende Hochschule |
|
Betreuer/-in / Berater/-in |
|
Publikationsstatus | Veröffentlicht - 2016 |