Emerging PET/MR systems seem promising for oncology, particularly in assessing identification of heterogeneous and complex tumor tissues. It becomes thus conceivable to optimize the treatment planning with regard to tumors local activities (aka dose painting). To this end, tumors and organs at risk (OAR) contouring reveals to be of the utmost importance. If therapists may be assisted today by proved algorithms for inter-modality registering and for OAR contouring, the question of the objectively and quantitatively aided detection of tumors appears still opened. The aim of this thesis is to tackle the detection of specific tissues in a fully 3D context where all desired modalities (anatomical and functional) are used together to build a joint co-segmentation rather than common fusion of paired sliced images. From a methodological point-of-view, we propose to put at the center of the structured modeling the emerging concept of random non-exchangeable partitions, leading to a neat basis for the construction of a multi-modality hierarchical joint segmentation.

This position is open until it is filled.

Département: DM2I (LIST)
Laboratory: Laboratoire Modélisation et Simulation des Systèmes
Start Date: 01-06-2015
ECA Code: SL-DRT-15-0444
Contact: eric.barat<στο>cea.fr