Computational & Technology Resources
an online resource for computational,
engineering & technology publications
PROCEEDINGS OF THE NINTH INTERNATIONAL CONFERENCE ON ENGINEERING COMPUTATIONAL TECHNOLOGY
Edited by: P. Iványi and B.H.V. Topping
Parallel Methods for 3D+t Anisotropic Diffusion of Dynamic Positron Emission Tomography Images
C. Tauber1, M. Chau2 and P. Spiteri3
1UMRS INSERM U930, Université François Rabelais, Tours, France
C. Tauber, M. Chau, P. Spiteri, "Parallel Methods for 3D+t Anisotropic Diffusion of Dynamic Positron Emission Tomography Images", in P. Iványi, B.H.V. Topping, (Editors), "Proceedings of the Ninth International Conference on Engineering Computational Technology", Civil-Comp Press, Stirlingshire, UK, Paper 26, 2014. doi:10.4203/ccp.105.26
Keywords: parallel asynchronous algorithm, subdomain method, iterative method, large scale system, nonlinear filtering, positron emission tomography.
In this paper, accurate and fast computations are proposed for an original set of coupled three-dimensional spatial problems of anisotropic diffusion. A spatiotemporal filter is presented for dynamic positron emission tomography imaging in which diffusion is based on the distances between vectorial representations of the voxel intensity across time. Spatial discretization is proposed combined with explicit or semi-implicit discretization schemes in which the stability of the global scheme is studied. At each time step, it is necessary to solve m large linear coupled algebraic systems for which two levels of parallelism are considered. The first level is due to the fact that the m images can be processed independantly once the coefficient of diffusion is computed. The second level of parallelism appears in the solution of each algebraic system, which can be solved by using parallel synchronous and asynchronous iterative methods. For the previous parallel iterative methods, we analyse the convergence. Experimental parallel simulations concerning 3D+t anisotropic diffusion of simulated dynamic positron emission tomography images are presented.
purchase the full-text of this paper (price £20)