Parallelization on a 8-GPU server of iterative 3D reconstruction algorithm for dental X-ray tomography

Abstract : In dental numeric 3D imaging, classic filtered backprojection (FBP) methods, for exemple, Feldkamp (FDK), is too sensible to the number of projections and noise, also limited in metal artifacts and beam-hardening reduction, so we use a 3D iterative least squares algorithm. Because of the huge size of 3D volumes (512 3 to 1024 3 voxels) and sinograms (512*512*360), it was too long to be applied in clinical medical imaging. However, thanks to the high performance of computing acceleration by GPU, its appearance in clinic is possible. In our iterative algorithm,we have two principal operations, projection and backprojection. According to our first resultats, the computation time of these operations on GPU is less in two orders (/100) than that on CPU, in particular for backprojection. With Multi-GPU (8 GPUs, Tesla C1060) in a Carri system's server, it decreases by a significant factor compared to single GPU. My Ph.D job consists of developping an iterative reconstruction method to reduce or remove the artifacts above, and obtain a segmentation of reconstructed volume as well.
Document type :
Poster communications
Complete list of metadatas

https://hal.archives-ouvertes.fr/hal-01851992
Contributor : Nicolas Gac <>
Submitted on : Tuesday, July 31, 2018 - 1:59:55 PM
Last modification on : Monday, October 28, 2019 - 5:40:35 PM
Long-term archiving on : Thursday, November 1, 2018 - 1:32:47 PM

File

gac_28229.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01851992, version 1

Collections

Citation

Long Chen, Nicolas Gac, Thomas Rodet, Colombe Maury. Parallelization on a 8-GPU server of iterative 3D reconstruction algorithm for dental X-ray tomography . PUMPS summer school, Jul 2012, Barcelone, Spain. ⟨hal-01851992⟩

Share

Metrics

Record views

52

Files downloads

47