IRM de diffusion du Q-space : Acquisition et pré-traitements

Emmanuel Caruyer 1
1 ATHENA - Computational Imaging of the Central Nervous System
CRISAM - Inria Sophia Antipolis - Méditerranée
Abstract : The overall goal of this thesis is to develop novel methods for the acquisition and the processing of diffusion magnetic resonance images (MRI), to provide new insights into the structure and anatomy of the brain white matter in vivo. Diffusion MRI is a non-invasive technique that measures locally the diffusion of water molecules. The latter are hindered by tissue structure, and therefore the characterization of water molecules displacement gives information on the nature, orientation, microstructure of the underlying tissue. Because of the strong anisotropy observed in white matter fiber tracts, this tool is most popular for the analysis of brain connectivity. One of the modality of acquisition and reconstruction, called diffusion tensor imaging, is now an established tool in research and clinical applications, for the detection of neural diseases and for pre-operative planning. Being model-based, the diffusion tensor cannot describe complex intra-voxel configurations, with multiple populations of fibers crossing. Since then, for a finer description of water molecules displacement, model-free approaches have recently been proposed, aiming at overcome the limitations of the diffusion tensor. Most of these techniques are still extremely demanding in acquisition time, and involve challenging reconstruction problems. The first part of this thesis proceeds from a description of the tissue microstructure, and a physical explanation of the origin of acquired diffusion signal. We give a review of the reconstruction methods and corresponding acquisition techniques in diffusion MRI. Several reconstruction methods are presented, and are categorized into model-based and model-free techniques. The first contribution of this thesis is related to the parametric reconstruction of the diffusion signal in a continuous basis of functions. We develop on a previous proposed method called Spherical Polar Fourier basis, and propose a continuous basis with a significant reduction of the dimension for the same power of description. We also derive the expression of the Laplace regularization operator in this basis, for a better robustness to noise. The second contribution is also related to the reconstruction of the diffusion signal, and the orientation distribution function, with a special focus on clinical setting. We propose a real-time reconstruction algorithm based on the Kalman filter to reconstruct the ODF in constant solid angle. We develop on top of the Kalman filter a motion detection algorithm, based on a monitoring and statistical analysis of the Kalman filter residuals. We are able to give a precise and sensitive motion detection, at no additional cost on the on-line acquisition system, as compared to systems based on camera and computer vision. The two last contributions are related to the acquisition methods in diffusion MRI, in particular for single and multiple $q$-shell experiments. We first describe a geometric approach to generate angular uniform schemes, that offer optimal angular coverage per shell and as a whole. Then we investigate on the link between the choice of a parametric basis of functions, and the design of sampling protocols. We give explicit methods to generate sampling schemes with minimal condition number, for the reconstruction in spherical harmonics (in Q-ball imaging) and the reconstruction in the modified spherical polar Fourier basis, proposed in this thesis. The conclusion of this approach is that the sampling method should be driven by the physical constraints of the scanner, and at the same time by the choice of a specific basis to represent the diffusion signal, and with an overall uniform coverage of the space of sampling directions, for a good rotational invariance. The new sampling schemes generated with this technique are available for download from my web page.
Complete list of metadatas

Cited literature [138 references]  Display  Hide  Download
Contributor : Emmanuel Caruyer <>
Submitted on : Friday, November 9, 2012 - 10:02:40 AM
Last modification on : Saturday, May 11, 2019 - 1:26:44 AM
Long-term archiving on : Sunday, February 10, 2013 - 3:40:08 AM


  • HAL Id : tel-00750144, version 1



Emmanuel Caruyer. IRM de diffusion du Q-space : Acquisition et pré-traitements. Medical Imaging. Université Nice Sophia Antipolis, 2012. English. ⟨tel-00750144⟩



Record views


Files downloads