My research activities are concentrated around the design and use of anatomical atlases for various applications ranging from the segmentation of structures in the brain to the voxel-wise comparison of different imaging modalities at a population level, e.g. detect differences due to a pathology or a specific treatment. I describe in related pages some of my main research publications.
Robust Detection of White Matter Abnormalities in DTI
New modalities such as Diffusion Tensor Imaging (DTI) are of great interest to characterize brain diseases, such as multiple sclerosis. However, many sources can result in a bias in DTI comparison, e.g. errors in non linear registration or artifacts (distortion) in the acquisitions. Moreover, many studies use scalar information derived from the tensor instead of the tensor itself. We are therefore interested in developing new methods to robustly compare the full DTI of a patient to controls. More details are available here.
New modalities such as Diffusion Tensor Imaging (DTI) are of great interest to characterize brain diseases, such as multiple sclerosis. However, many sources can result in a bias in DTI comparison, e.g. errors in non linear registration or artifacts (distortion) in the acquisitions. Moreover, many studies use scalar information derived from the tensor instead of the tensor itself. We are therefore interested in developing new methods to robustly compare the full DTI of a patient to controls. More details are available here.
Locally Most Similar Atlas Construction
High anatomical variability may corrupt atlas-based segmentation results by generating potentially large registration errors. To overcome this drawback, we are highly interested in the construction of an atlas whose anatomy is as similar as possible to a given patient to delineate. Moreover, we are also interested in obtaining efficient methods so that they can be used in practice. More details are available here.
High anatomical variability may corrupt atlas-based segmentation results by generating potentially large registration errors. To overcome this drawback, we are highly interested in the construction of an atlas whose anatomy is as similar as possible to a given patient to delineate. Moreover, we are also interested in obtaining efficient methods so that they can be used in practice. More details are available here.
Image Registration (Locally Affine Framework)
This work was motivated originally by the atlas-based segmentation of anatomical structures. We introduce a new registration algorithm to register in a robust manner predefined anatomical regions. These regions are then registered using a local affine transformation on each region and the global transformation may be computed using the log-euclidean framework on affine transformations. These choices render the algorithm very fast and robust. More details are available here.
This work was motivated originally by the atlas-based segmentation of anatomical structures. We introduce a new registration algorithm to register in a robust manner predefined anatomical regions. These regions are then registered using a local affine transformation on each region and the global transformation may be computed using the log-euclidean framework on affine transformations. These choices render the algorithm very fast and robust. More details are available here.
PhD Thesis: Atlas-Based Segmentation for Radiotherapy
My PhD topic was about the design and use of anatimocal atlases for the segmentation of brain structures and head and neck structures. These segmentations were then used for the planning of radiotherapy treatment. We introduced in this PhD thesis several methodological advances also described here, such as locally affine registration, atlas construction from a database, and validation methodologies. Atlas-based segmentation method and more details on the PhD thesis are available here.
My PhD topic was about the design and use of anatimocal atlases for the segmentation of brain structures and head and neck structures. These segmentations were then used for the planning of radiotherapy treatment. We introduced in this PhD thesis several methodological advances also described here, such as locally affine registration, atlas construction from a database, and validation methodologies. Atlas-based segmentation method and more details on the PhD thesis are available here.
Head and Neck Atlas Construction and Evaluation
Atlas-based segmentation is an efficient and precise way of delineating many structures of a single organ, such as the brain. In this work, we aimed at developing an atlas for segmentation for another region, where cancers are frequent: the head and neck region. To cope with large deformations between the anatomies of the various patients, one must be careful of the registration strategy when building the atlas. We have therefore presented a methodology to create a head and neck atlas and evaluate the registration algorithms used in this process. More details are available here.
Atlas-based segmentation is an efficient and precise way of delineating many structures of a single organ, such as the brain. In this work, we aimed at developing an atlas for segmentation for another region, where cancers are frequent: the head and neck region. To cope with large deformations between the anatomies of the various patients, one must be careful of the registration strategy when building the atlas. We have therefore presented a methodology to create a head and neck atlas and evaluate the registration algorithms used in this process. More details are available here.
Incorporating Deformability Statistics in Non-Rigid Registration
Due to its multi-subject nature, the non-rigid registration problem is generally difficult. The brain variability between subjects is indeed not well known. We have therefore introduced a method to learn brain deformability from a database of images. This deformability can be either as a scalar field or a tensor field, introducing directional deformability. We then proposed a method to take this a priori deformability into account when registering new images. This method showed improved results especially when utilizing tensor-based deformability. More details are available here.
Due to its multi-subject nature, the non-rigid registration problem is generally difficult. The brain variability between subjects is indeed not well known. We have therefore introduced a method to learn brain deformability from a database of images. This deformability can be either as a scalar field or a tensor field, introducing directional deformability. We then proposed a method to take this a priori deformability into account when registering new images. This method showed improved results especially when utilizing tensor-based deformability. More details are available here.
Taking into Account Pathologies in Image Registration
The registration of two images may be corrupted when one image presents lesions and the other image does not present any lesions. Regular registration algorithms indeed suppose that images when registered have corresponding tissues. This presence of lesions or absence of tissues due to a surgical procedure therefore needs to be taken into account to avoid registration errors. To tackle this problem, we have introduced a general method to take into account a lesion mask in a registration algorithm. More details are available here.
The registration of two images may be corrupted when one image presents lesions and the other image does not present any lesions. Regular registration algorithms indeed suppose that images when registered have corresponding tissues. This presence of lesions or absence of tissues due to a surgical procedure therefore needs to be taken into account to avoid registration errors. To tackle this problem, we have introduced a general method to take into account a lesion mask in a registration algorithm. More details are available here.
