Restoration for RF Induced Spatial Intensity Non-uniformities in MRI Data with Statistical, Structural, and Physical Constraints

• Applicant: Professor Dr. Michael Bock, since 3/2014
• Subject Area: Medical Physics, Biomedical Technology ; Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
• Term: from 2014 to 2020
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 237079734

MRI reconstruction assumes uniform Radio-Frequency (RF) fields. Modern MRI systems generate the RF fields with integrated coil arrays that give rise to higher resolution as well as higher Signal to Noise Ratio (SNR). However, at the same time they give rise to an inhomogeneous RF field that leads to intensity non-uniformities across an image. Non-uniformity artifacts are also more pronounced in high field MRI and complicate reconstruction, confound tissue properties, compromise diagnostic quality, and any further automated image analysis. This proposal intends to address this problem with the development of methods for the estimation of primarily the receive non-uniform sensitivity of a single coil as well as of multiple coils in phased arrays. It will remove non-uniformity artifacts in single contrast images as well as jointly for images of multiple contrasts. The effect of this artifact on the statistics of intensity co-occurrences as well as on cross-co-occurrences between images will be removed non-parametrically. The statistics will be restored both with Wiener filtering as well as together with sparsity constraints. Constraints from statistics of intensity differences in space will also be considered both parametrically as well as non-parametrically. The single contrast parametric constraint will be in terms of the total variation. The multi-contrast parametric constraint will be in terms of the Laplace-Beltrami operator. The non-parametric constraint based on intensity differences will be expressed with general image features and will be performed with Wiener filtering. The restored statistics will be forced to the imaging data. The accuracy of the statistical restoration will also consider the valid signal regions of the images and partial volume artifacts. The restoration will be constrained with RF non-uniformity field mappings that will be obtained during the acquisition phase. The non-uniformity will be assumed to be smooth in space.The combinations of the various constraints including the one from the RF field map will be validated extensively. They will also be compared with other post-acquisition restoration methods. The validation will be performed with data that includes the BrainWeb 1,5 T phantom as well as brain anatomic data acquired at 3,0 T and 7,0 T at various MRI systems. The brain data will be from healthy volunteers as well as from patients suffering from multiple sclerosis and Alzheimers. Validation will also be performed with 3,0 T chemical shift imaging for fat/water reconstruction with data obtained from both a physical phantom and from obese volunteers over an extensive body region. The numerical implementation of the restoration will be iterative and will be expedited with a parallel implementation.

• DFG Programme: Research Grants
• Participating Institution: Deutsches Krebsforschungszentrum (DKFZ); Otto-von-Guericke-Universität Magdeburg; Institut für Physik
• Participating Person: Professor Dr. Oliver Speck
• Ehemaliger Antragsteller: Dr. Stathis Hadjidemetriou, Ph.D., until 2/2014