Skip to main content

CIRtalk 0509: Salma Essafi

Sorry, this content is only available in German!

Sparse Models and 3D Hierarchical Wavelet Knowledge based Muscle Segmentation

Salma Essafi (Ecole Centrale de Paris)
Where: Pokieser Seminarraum (Leitstelle 7F, Ebene 7, AKH Wien Waehringerguertel 18-20)
When: Donnerstag, 28.5.2009, 14:30h


Myopathies are diseases that affect the muscle system, and lead to a severe deterioration of the motoric abilities. These pathologies affect 4 to 6% of the population, i.e 25 to 30 million Europeans. Diagnosis, as well as follow up for a given therapeutic strategy are often performed through biopsy. Magnetic resonance imaging (MRI) allows a non-invasive  observation of the muscle fibers, their texture, and their global structure. This enables the analysis of local properties, as well as understanding the global structural change of muscles affected by a disease. A crucial first step in this analysis is the accurate segmentation of individual muscles. State of the art segmentation methods mainly rely on a clearly  defined topology, and an object boundary characterized by salient features (e.g, edges). However muscle-compounds pose a rather different and new challenge to segmentation algorithms mainly because there is no prominent difference of tissue-properties between neighboring muscles. Border tissues in between muscles are only visible on specific locations, distributed in a very sparse and heterogeneous manner.

We propose a novel representation of prior knowledge for image segmentation, using diffusion wavelets that can reflect arbitrary and continuous inter-dependencies in shape data. The application of diffusion wavelet techniques has, so far, largely been confined to signal processing. In our approach, and in contrast to state-of-the-art methods, during the learning stage we optimize the coefficients, the number and the position of landmarks, and the object topology using geometric -reconstructed surface- constraints, in a coarse-to-fine manner. The resulting paradigm supports hierarchies both in the model and the search space, can encode complex geometric and photometric dependencies of the structure of interest, and can deal with arbitrary topologies. We report results on two challenging medical data sets -heart and calf muscles-, that illustrate the impact of the soft parametrization and the potential of our methodology.