Detector simulations for in-beam PET with FLUKA. Francesco Pennazio Università di Torino and INFN, TORINO

Transcription

1 Detector simulations for in-beam PET with FLUKA Francesco Pennazio Università di Torino and INFN, TORINO

2 Outline Why MC simulations in HadronTherapy monitoring? The role of detector simulation in in-beam PET Detector construction and simulation validation Simulated image reconstruction Fluka detector simulation source.f Geometry mgdraw.f Post-processing Results Conclusions See Elisa Fiorina presentation for project introduction 1

3 Why MC simulations in HadronTherapy monitoring? monitored quantity (PET image, charged tracks, prompt gamma) needs reference to be compared with dose in patient is not directly related to any of them b + activity distribution 11 C, 15 O, 10 C, J Pawelke et al., Proceedings IBIBAM, , Heidelberg prompt secondary particles emission L. Piersanti et al. Phys. Med. Biol

4 Why simulate the in-beam PET detector? (1) Usual MC detector development and preliminary performance assessment: Preliminary validation Detection rate Coincidence rate Geometrical acceptance

5 Why simulate the in-beam PET detector? (1) Usual MC detector development and preliminary performance assessment: Preliminary validation Detection rate Coincidence rate Geometrical acceptance!!! DATA FOR PRELIMINARY DETECTOR VERSION!!! (few channels, actual results are fairy better)

6 Why simulate the in-beam PET detector? (2) Dual-head scanner -> LORs have only limited angular distribution Severe artifacts in reconstructed image along the vertical coordinate (y) although the resolution is still pretty good on the x-z plane How to compare activity scored in simulation and experimental image? bw simulated pencil beam image color target activity

7 Why simulate the in-beam PET detector? (2) Dual-head scanner -> LORs have only limited angular distribution Severe artifacts in reconstructed image along the vertical coordinate (y) although the resolution is still pretty good on the x-z plane How to compare activity scored in simulation and experimental image? Try to correct artifacts in exp. image recovering somehow the missing information (TOF?) Simulate PET detectors and digitization Look for coincidences Reconstruct the simulated image with the same algorithm used for data!

8 Why simulate the in-beam PET detector? (2) Dual-head scanner -> LORs have only limited angular distribution Severe artifacts in reconstructed image along the vertical coordinate (y) although the resolution is still pretty good on the x-z plane Simulate PET detectors and digitization Look for coincidences Reconstruct the simulated image with the same algorithm used for data! simulation data Missing projections artifact reproduced

9 Why simulate the in-beam PET detector? (2) Dual-head scanner -> LORs have only limited angular distribution Severe artifacts in reconstructed image along the vertical coordinate (y) although the resolution is still pretty good on the x-z plane Simulate PET detectors and digitization Look for coincidences Reconstruct the simulated image with the same algorithm used for data!!!! DATA FOR PRELIMINARY DETECTOR VERSION!!! (few channels, actual results are fairy better) simulation data Missing projections artifact reproduced

10 In-beam PET Università di Torino Isotopes production is a poor signal all the statistics must be simulated. STEP 1: Beam simulation Time-tagged activity scoring STEP 2: PET simulation Data analysis & image reconstruction About 1/100 of primary hadrons Annihilation time and position Isotope production map: 11 C (t1/2= s) 15 O (t1/2=122.4s) 10 C (t1/2=19.290s) All positrons are simulated. Detector simulation. Same as real data: Line Of Response (LOR) list extraction Image reconstruction (MLEM algorithm, 5 iterations) The temporal structure of the beam delivery is simulated. Isotope decays are simulated. time The 4D reconstructed image depends on the acquisition time. E. Fiorina FLUKA User Workshop, Paris, May 12 th 2016

11 FLUKA detector simulation (1): source Custom source.f routine to generate positrons: read activation (i.e. radioactive isotopes position) from step1 ->x,y,z,t 0 SMBNAI to generate energy according to beta+ emission spectrum if positron is below EM transport threshold it is arbitrary sampled in a small cube around x,y,z Particle age set to t 0 +t, with t extracted according to isotope t increase statistics to simulate the whole treatment typically 100x positrons emitted for each isotope (controlled from card SOURCE) Discard positrons emitted after the end of acquisition time (controlled from card SOURCE) 4

12 FLUKA detector simulation (2): geometry Active part of the INSIDE in-beam PET scanner is reproduced in geometry 16x16 independent channels LFS+MPPC detector is reproduced use of LATTICE to replicate the block detector head distance configurable via pre-processor directive define 4

13 FLUKA detector simulation (3): mgdraw.f Active area corresponds to each scintillator crystal For each event and each detector element must be scored: Total energy deposited (if >0) Detector element number (identified by LT1TRK and MREG) Interaction time (ATRACK) (approximated by the first interaction time for the given event) one can also score more quantities, but (E, t) are the only physical observables measured by the DAQ system Entry points: mgdraw and endraw 4

14 Post-processing (E,t) quantities saved by mgdraw routine have infinite measurement precision How to take into account energy and time resolution of the detectors? Smear E,t in mgdraw before saving Develop a program to perform GATE-like digitisation: E,t smearing Event merging in the same crystal Coincidence finding with the same algorithm used for data 4

15 Results: image and profile reconstruction A B D Data Simulation beam PMMA phantom 2 slices 2.7x2.7 cm 2 (E 1 = 77 MeV, E 2 = 105 MeV) Acquisition time = 254 s PMMA phantom with hole 2 slices 2.7x2.7 cm 2 (E 1 = 77 MeV, E 2 = 105 MeV) Acquisition time = 155 s PMMA phantom Real treatment plan (E min = 76.6 MeV, E max = MeV) Acquisition time= 133 s

16 Results: image and profile reconstruction beam

17 Conclusions A two steps simulation of the 4D b + activity distribution generated during proton therapy treatments was implemented. In the first step, the beam delivery is simulated with a lower statistics by taking into account temporal information and the CNAO beam pipe features. The isotopes produced are scored and used as source for positrons generation in a second simulation step with a full statistics. In the second step, also, the detector geometry and figures of merit are taken into account. Simulated in-beam PET images show a good agreement with first measurements at the CNAO synchrotron facility.

18 Open issues 1) Use of the CT of the anthropomorphic phantom/patient to perform simulation. 2) Reduce as much as possible time calculation. Any suggestions? 3) Simulation uncertainty evaluation. Main contribution: delivery beam time distribution, spatial alignment, image reconstruction algorithm. How we can take into account statistical error given from simulation? 4) Biological washout? 5) Carbon ion therapy simulation.

19 Backup slides

20 Monitoring in HT: Operating principle Treatment plan Expected value calculation Treatment. The sooner the better ( real time?) Reliable system (<1mm uncertainty in Bragg peak position) Measurement Data acquisition Data analysis Comparison Result 5 F. Pennazio INFN CCR Workshop, March 17 th 2016

21 Monitoring in HT: in-beam PET PMMA phantom 2 slices 2.7x2.7 cm 2 (E 1 = 77 MeV, E 2 = 105 MeV) Acquisition time = 254 s, absolute comparison (no normalization applied) 6 F. Pennazio INFN CCR Workshop, March 17 th 2016

22 Preliminary results Bragg peak 77 MeV Bragg peak 105 MeV Expected difference 31.2 mm Measured difference (30.2 +/- 0.3) mm beam in p 11 F. Pennazio INFN CCR Workshop, March 17 th 2016

23 In-beam PET simulations STEP 1 (treatment): Beam simulation Time-tagged activity scoring STEP 2 (detector): PET simulation Data analysis & PET image reconstruction About 1/1000 of primary hadrons, custom simulation Annihilation time and position (gold standard) Isotope production map All positrons are simulated. Detector simulation (time and energy deposition) Same as real data: Line Of Response (LOR) list extraction Image reconstruction (MLEM algorithm, 5 iterations) LFS side In-beam PET heads simulated geometry LFS MPPC Simulation time: about 8 h (32-core) It is all about number of CPUs (low memory/disk requirements 16 F. Pennazio INFN CCR Workshop, March 17 th 2016

24 In-beam PET simulations: future challenges Goal: improve simulation speed Treatment is based on pencil-beams: easy parallelization Simulation simplification: bias, more aggressive energy cuts, simplified transport Use of new non-standard tools (planit?) GPU for image reconstruction Goal: improve simulation accuracy More accurate reproduction of beam structure (i.e. beam delivery time may not be exactly known a priori) Accurate bias and simulation of significant processes other than beta+ decay Ultimate Goal: fast detector response simulation On-line accurate monitoring for cyclotrons and fast duty cycle synchrotrons Easier deconvolution of data from background during beam delivery 17 F. Pennazio INFN CCR Workshop, March 17 th 2016