Text optional Institutsname Prof Dr Hans Mustermann www

  • Slides: 12
Download presentation
Text optional: Institutsname Prof. Dr. Hans Mustermann www. fzd. de Mitglied der Leibniz-Gemeinschaft

Text optional: Institutsname Prof. Dr. Hans Mustermann www. fzd. de Mitglied der Leibniz-Gemeinschaft

Fastest Data Processing in Image Reconstruction for Compton Camera Imaging DTSP-Workshop on “Ultra-fast data

Fastest Data Processing in Image Reconstruction for Compton Camera Imaging DTSP-Workshop on “Ultra-fast data transfer and reconstruction” (pillar 2) Burg Obbendorf, Jülich May 9 th - May 10 th Sebastian Schöne, Radiation Physics, HZDR Text optional: Institutsname Prof. Dr. Hans Mustermann www. fzd. de Mitglied der Leibniz-Gemeinschaft

Objectives within the project – CCI Image Reconstruction Ultrafast data transfer and reconstruction •

Objectives within the project – CCI Image Reconstruction Ultrafast data transfer and reconstruction • Intelligent programmable hardware • Near detector optical signal transmission • Fastest data processing with highly parallel architectures Technologies for assembling highly integrated detectors 3 D ASICs Mixed-signal ASICs 3 D / high-Z sensors Packaging and interconnection technologies • Innovative detector structure materials • • Detector types • Fast photon and X-ray detectors • Diamond detectors • Detectors for thermal neutrons • Compact gas detectors Cross cutting activities Member of the Helmholtz Association Sebastian Schöne • Radiation Physics • HZDR

Objectives within the project – CCI Image Reconstruction Why Compton camera imaging (CCI) ?

Objectives within the project – CCI Image Reconstruction Why Compton camera imaging (CCI) ? • Tumor radiation therapy • Protons (and light ions) • More local dose deposition w. r. t. photon irradiation • Dose monitoring • Prompt gammas • By means of Compton cameras A. Müller, Geant 4 simulations • SPECT Member of the Helmholtz Association Sebastian Schöne • Radiation Physics • HZDR

Objectives within the project – CCI Image Reconstruction Principle of Compton camera imaging R

Objectives within the project – CCI Image Reconstruction Principle of Compton camera imaging R axis Q, Eg g q q Scatter q P, L 1 Pscatter Absorber R, L 2 Papex Pabsorber Member of the Helmholtz Association Sebastian Schöne • Radiation Physics • HZDR

Objectives within the project – CCI Image Reconstruction Our prototypes C. Golnik, CZT-Setup T.

Objectives within the project – CCI Image Reconstruction Our prototypes C. Golnik, CZT-Setup T. Kormoll, CZT-LSO-Setup Member of the Helmholtz Association Sebastian Schöne • Radiation Physics • HZDR

Objectives within the project – CCI Image Reconstruction Imaging workflow Study object x Measurement

Objectives within the project – CCI Image Reconstruction Imaging workflow Study object x Measurement y = A(x) Measurement series y Reconstruction x’ = A-1(y) 1 st Construct model A of device Reconstructed image x’ ≈ x 2 nd Optimize image x’ Member of the Helmholtz Association Sebastian Schöne • Radiation Physics • HZDR

Objectives within the project – CCI Image Reconstruction 1 st Construct model A of

Objectives within the project – CCI Image Reconstruction 1 st Construct model A of device • Measured data handled as distributions + • y=A(x) high dimensional: y in R 8, x in R 4 • Additional influences • Cross sections • Camera geometry • Medium memory consuming • …. • High time consumption • ~1 s/event/core • Assumption: 10 k events/s • Assumption: 100 k filtered events per recording Member of the Helmholtz Association Sebastian Schöne • Radiation Physics • HZDR

Objectives within the project – CCI Image Reconstruction 2 nd Optimize image x’ 22

Objectives within the project – CCI Image Reconstruction 2 nd Optimize image x’ 22 Na point @ (0, 4, 7) cm Eventfilter 1275 ke. V +/- 20% • Standard algorithms exist • Less complex • Less time consuming • High memory consumption Summed backprojection 1, 2, …, 7, 50, 100, 500, 800 events • e. g. operate on n*10 G floats MLEM, Iteration 1, 2 … 25 800 events Member of the Helmholtz Association Sebastian Schöne • Radiation Physics • HZDR

Objectives within the project – CCI Image Reconstruction Status quo Plans • Python •

Objectives within the project – CCI Image Reconstruction Status quo Plans • Python • Migration to massive parallel programming • Num. Py + Sci. Py • Workstation & HPC cluster • Selective (module) Wishes • Interface to high-level programming • Python integration • Successive • Open. CL vs. CUDA • Permanent Parallel drop-in replacement ? (beam time, treatment room) • Multi-GPGPU • GPGPU-alternatives • …. Are ‘general purpose’ implementations reasonable? Member of the Helmholtz Association Sebastian Schöne • Radiation Physics • HZDR

Objectives within the project – CCI Image Reconstruction HPC @ HZDR HPC @ TU

Objectives within the project – CCI Image Reconstruction HPC @ HZDR HPC @ TU Dresden • 2 HPC clusters • Multiple HPC cluster • Small GPGPU cluster • S 1070 • C 2070 • S 2050 • C 2070 • … • Lectures on massive parallel programming • CUDA Research Center • Awarded CUDA Center of Excellence Member of the Helmholtz Association Sebastian Schöne • Radiation Physics • HZDR

Member of the Helmholtz Association Sebastian Schöne • Radiation Physics • HZDR

Member of the Helmholtz Association Sebastian Schöne • Radiation Physics • HZDR