Recalage non rigide en imagerie cérébrale : méthodes et validation

Abstract : This thesis is concerned with the non rigid registration of medical images, either monomodality or multimodality. In the first part, we have developed a method to estimate a $3D$ dense deformation field between the brain of two different subjects. This approach is based on the estimation of the optical flow, expressed in a robust framework. We have also proposed a minimization scheme, which is both multiresolution and multigrid. The multigrid minimization is based on an adaptive partition of the volume that is constrained with a segmentation of the brain. We have evaluated the algorithm on simulated data and on a database of $18$ subjects. We have then extended the registration method to non-rigid multimodality registration. EPI acquisition, which is commonly used for fMRI data, introduces geometrical distortions that disturb significantly the interpretation of functional data. We intend to correct the distortions by performing a non rigid registration between the fMRI acquisition and the MRI volume. Therefore, the energy-based framework and the multigrid minimization scheme have been modified with an appropriate similarity measure. We have proposed an evaluation framework for non-rigid registration of brains of different subjects. $5$ methods, among which classical and modern methods, have been tested on a database of $18$ subjects. We have designed global and local criteria to assess the quality of the registration. Finally, we have studied the cooperation between segmentation and registration. On the one hand, we showed that a segmentation process can be easily guided by a coarse registration. The segmentation becomes fully automatic, more robust, more precise and the computation time is significantly reduced. On the other hand, we investigated the introduction of sparse cortical constraints in the registration process presented in the first part. The energy based framework makes it possible to incorporate naturally the matching of cortical sulci. We showed the benefits of that cooperative approach on a database of $18$ subjects.
Mots-clés :
Document type :
Theses

https://tel.archives-ouvertes.fr/tel-00536029
Contributor : Pierre Hellier <>
Submitted on : Monday, November 15, 2010 - 10:06:02 AM
Last modification on : Wednesday, April 11, 2018 - 1:51:41 AM
Long-term archiving on : Wednesday, February 16, 2011 - 2:45:14 AM

Identifiers

• HAL Id : tel-00536029, version 1

Citation

Pierre Hellier. Recalage non rigide en imagerie cérébrale : méthodes et validation. Modélisation et simulation. Université Rennes 1, 2000. Français. ⟨tel-00536029⟩

Record views