The Ultra Fast Tomography system at the ANKA Synchrotron Light Source at KIT allows to study moving biological objects with high temporal and spatial resolution. The resulting amounts of data are challenging in terms of reconstruction algorithm, automatic processing and computing:

  • Fewer projections from Ultra Fast tomography lead to reconstructions with artefacts using standard algorithms
  • Laborious manual process for data analysis
  • Large amounts of data sets and metadata
  • Computational expensive

ANKA.jpg 

Ultra Fast Tomography beamline at ANKA.

  

BUG frog

Moving biological objects: a living bug (left) and African clawed frogs (right).

 

Sparse Reconstruction

Sparse reconstruction in computed tomography (CT) refers to an estimation of a numerically accurate tomographic image if the projection data is not sufficient for exact image reconstruction according to the Nyquist-Shannon sampling theorem. We developed a new algorithm ART-CS com­bining the algebraic reconstruction technique (ART) and the compressive sampling theory (CS) optimizing the images based on total variation.

Compared to the standard algorithms the results show high quality images even if only 60 instead of 1500 projections are used.

 

 

recon1 recon2 recon3

Left: standard reconstruction with 1500 projections; Middle: standard reconstruction with 60 projections showing artefacts; Right: ART-CS reconstruction using 60 projections.

 

Parallel Computing

The reconstruction of a full volume on a standard workstation using ART-CS requires 11 hours of computing time. For near real time processing a HADOOP cluster connected to the Large Scale Data Facility (LSDF) is used.  After data ingest a reconstruction workflow is started automatically processing the MATLAB programs in parallel. By this the computation time is reduced to 6 minutes [ANKA2].

 

ART-CS and the parallel computation using the LSDF infrastructure are general concepts and can be used for many other tomography systems.

Workflow

An automatic workflow of the data analysis: parallel computing, marked in dark, proves its efficiency and effectiveness; speedup for a data set reconstruction goes up to 120.


[ANKA2]  Yang, X.; Jejkal, T.; Pasic, H.; Stotzka, R.; Streit, A.; van Wezel, J. & dos Santos Rolo, T., Data Intensive Computing of X-Ray Computed Tomography, 21st Euromicro International Conference on Parallel, Distributed, and Network-Based Processing, 2013, 86-93

 

Contact:

KIT, IPE: Xiaoli Yang, Rainer Stotzka, Thomas Jejkal

KIT, IPS: Tomy dos Santos Rolo, Thomas van de Kamp, Julian Moosmann, Ralf Hofmann

Copyright by SWM, KIT – Universität des Landes Baden-Württemberg und nationales Forschungszentrum in der Helmholtz-Gemeinschaft
Templates Joomla 1.7 by Wordpress themes free