Parallélisation sur GPU d'un algorithme de reconstruction 3D bayésien en tomographie X. - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2009

Parallélisation sur GPU d'un algorithme de reconstruction 3D bayésien en tomographie X.

Résumé

An important number of image reconstruction algorithms are implemented in the literature on X-ray CT (Computed Tomography) data. The main family of methods are analytical, mainly filtered backprojection which are implemented generally in medical imaging for their fast reconstruction time. The limits of these methods appear when the number of projections is small, and/or not equidistributed around the object. In this specific context, iterative algebraic methods are implemented. A great number of them are mainly based on least square criterion. We propose a regularized version of iterative algorithms to improve results. The main problem that appears when using iterative algebraic methods is the computation time and especially for projection and backprojection steps. We propose to implement some steps of the iterations on GPU hardware. We present an original method based on a Bayesian statistical method for 3D tomographic reconstructions. The main interest is to apply it in a context of non-consistent data sets, for example with a small number of projections. We show a good quality of results and a significant speed up of the calculation with GPU implementation.
Fichier principal
Vignette du fichier
GPU_Workshop_Obernai_mai_2009.pdf (162.88 Ko) Télécharger le fichier
Presentation_Oral_Colloque_GPU_Obernai_GAC_Nicolas.pdf (1.79 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)
Format : Autre

Dates et versions

hal-00504759 , version 1 (21-07-2010)

Identifiants

  • HAL Id : hal-00504759 , version 1

Citer

Nicolas Gac, Alexandre Vabre, Ali Mohammad-Djafari, Fanny Buyens, Samuel Legoupil. Parallélisation sur GPU d'un algorithme de reconstruction 3D bayésien en tomographie X.. Développement d'applications de calcul intensif sur carte graphique pour l'imagerie moléculaire, May 2009, Obernai, France. ⟨hal-00504759⟩
232 Consultations
252 Téléchargements

Partager

Gmail Facebook X LinkedIn More