This thesis aims at improving the statistical power of group comparison in medical imaging studies. Group comparison are widely used, for instance to automatically detect effects of pathologies (like Alzeimer's disease) on brain structures. Image registration is thoroughly used during group comparison, and consists in finding the most accurate spatial transformation between two anatomies. One of the contribution of this work will be to inject information extracted from multiple modalities like functional or diffusion MRI into registration to improve its accuracy. Then, the benefice of such multi-modal and multi-structural registration framework will be evaluated on databases of Alzheimer patients compared to normal controls, and biomarkers of the pathology will be determined.
In medical imaging studies, one often needs to compare images of hundreds of patients to images of hundreds of normal controls to detect abnormalities caused by a pathology. These abnormalities can be seen as deviations from the normal distribution of a particular structure in terms of position, surface/volume, or overall shape.
Image registration is the central operation in group studies and consists in computing a spatial transformation mapping one subject's anatomy onto another. It should be the most accurate possible. Very often, image registration relies on a unique image modality while many of them are now routinely available (anatomical, functional and diffusion MRI to quote a few). The goal of this work will consist in developing a registration framework able to handle several modalities to produce the most accurate spatial transformations, and to further improve the statistical power of group comparisons.
The objectives of this thesis are twofold:
To develop a framework for multi-modal and multi-structure image registration to benefit from the maximum of information available when registering two anatomies. Such approach will be compared to the mono-modal version for validation.
To compare populations of subjects based on this registration framework. First, the framework will be used to compute numerical atlases of the human brain anatomy containing all structures used during registration. Second, those atlases will serve to automatically detect statistically significant differences between populations, for instance by comparing the distributions of each population around the atlases using parametric or non-parametric statistical methods.
Applications of this work include the detection of biomarkers of neurological pathologies like Alzheimer's disease.
The first step will consist in a proof of concept giving evidence that image registration accuracy is improved by considering multiple structures and imaging modalities instead of one. Second, a review of the most interesting modalities (anatomical, functional and diffusion MRI) and structures (cortical surface, functional regions, neural fibers) will be made. The registration framework will be adapted to include structures of different kind (lines, surfaces, volumes). Finally, given databases of modalities and structures extracted in Alzheimer patients, numerical atlases of those will be computed and compared to a population of normal controls to infer automatically whether the pathology has a significant impact on one of them.
The PhD candidate will integrate the INRIA team PARIETAL which is located in Neurospin, a research center dedicated to brain imaging.
A MSc in computer science or equivalent is required. Excellent programming skills in C/C++ and python are expected. Expertise in medical image processing would be highly appreciated.
Status of funding
Vendredi 11 juin 2010 15:33:26 CEST
Vendredi 11 juin 2010 15:33:26 CEST
Aucun fichier joint à cette fiche
Ecole Doctorale Informatique Paris-Sud
Nicole Bidoit Assistante
Stéphanie Druetta Conseiller aux thèses
ED 427 - Université Paris-Sud
UFR Sciences Orsay
Bat 650 - aile nord - 417
Tel : 01 69 15 63 19
Fax : 01 69 15 63 87
courriel: ed-info à lri.fr