Processing and analyzing large medical image sets
Traitement et analyse de grands ensembles d'images médicales
Résumé
Medical images analysis today requires high end computing infrastructures to face the growing need for computing power and distributed medical data management. Medical applications increasingly require the federation of large data sets distributed over the image acquisition centers and their processing (e.g. for epidemiological studies, statistical construction of atlas, medical image analysis procedures validation...). We are interested in the use of production grids which represent affordable computing infrastructures, externalized for the non-specialized clinical centers. Furthermore, grids foster collaborative work by easing the sharing of data, processing, data analysis procedures, and so on. Exploiting grid infrastructures for medical data analysis require to fulfill specific requirements associated to medical image processing such as data protection, large data sets manipulation and easy set up of multiple data analysis procedures. In this defense, we will show how we came from medical image analysis procedures validation to the exploitation of grid infrastructures. We will analyze the needs related to this application area. The work we are doing to interface clinical storage with grid data management systems and to represent and efficiently process data-driven application workflows on a grid infrastructure will then be presented. Probabilistic modeling of the grid execution will be proposed and strategies to schedule and efficiently execute data-intensive workflows on a grid will be derived. Results will be shown on the Bronze Standard, an application to medical image registration algorithms assessment, exploiting both the EGEE production grid infrastructure and the Grid5000 National experimental grid.
Non disponible
Domaines
Informatique [cs]
Origine : Fichiers produits par l'(les) auteur(s)