Proton tracking for medical imaging and dosimetry J.Taylor, P.Allport, G.Casse For the PRaVDA Consortium 1
Background and motivation - What is the PRaVDA experiment? - Why are we using Monte Carlo? GEANT4 Simulation design and useful outputs - Simulating efficiency - Simulating tracking and reconstruction - Simulating effects from electronics and detector alignment Summary Talk Overview 2
x - outputs x - outputs x - outputs x - outputs PRaVDA System Overview Each 150um thick silicon detector has 2048 strips, 1024 read out on each side of the detector. Each station of strip modules has three detectors crossed at 60 o in an (x,u,v) configuration to resolve ambiguities at high particle rate. 3
Operating Modes Treatment Mode (High Current) Field size: 5cm collimated treatment beam Energy: 60-191 MeV Flux: ~10 7 protons/cm 2 /s Use Strip Tracker to.. Check beam profile reconstruct 1D & 2D histograms Measure dose Requirements: - Proton counting - 1D histograms - 2D beam profile Patient Imaging Mode (Low Current) Field size: 10cm (max.) Energy: 191 MeV Flux: ~10 5 protons/cm 2 /s Use Strip Tracker to.. Track individual protons in (x,u,v) layers Use Range Telescope to.. Measure positions and energies of each proton Requirements: - Accurate Tracking - High Efficiency 4
Simulated de/dx distributions 5
Simulating the Efficiency of the (x,u,v) configuration How efficient is the tracker at distinguishing hits in a multi-hit environment in both Treatment and Patient Imaging mode? For a strip detector with orthogonal strips and N hits, there are: N 2 - N Ghost-hits or ambiguities generated 6
x - outputs x - outputs Simulating the Efficiency of the (x,u,v) configuration Tilt planes of strips at a (stereo) angle w.r.t other planes to prevent strip from one plane crossing all strips in the next plane. u v Tracker module design x Each station has strips crossed at 60 o to one another in (x,u,v) configuration allowing higher particle rate. 7
Simulated Efficiency in Treatment Mode Pacman Input Distribution Output Distribution Output Distribution 60 protons/frame 120 protons/frame Determine efficiency of (x,u,v) configuration in treatment mode using simple Monte Carlo A concave distribution is a good test of how well we can reconstruct the beam profile without introducing artifacts High Efficiency The detector is highly efficient for patient imaging with >99% hits unambiguous at well over 10 times planned ithemba pct mode rates 8
Input distributions for GEANT4 ithemba beamline above used to generate input distributions for GEANT4 simulations of silicon strip tracker 125 MeV beam Angular distribution of 9.8 mrad Field size of 8.5 x 8.5 cm 22-04-15 UCL Monte Carlo Meeting, 22-04-15 9
GEANT4 Simulation of tracker Phantom Phantom Calorimeter Range Compensator GEANT4 Simulation of the tracker uses the ithemba beamline simulation on the previous slide as an input. The tracker simulation consists of: There are 8 planes before and 8 planes after the phantom: - 6 silicon planes arranged into two (x,u,v) modules (shown in red) - 2 air planes used to provide truth information (shown in white) Each plane uses G4VDigitizerModule to store sensitive detector information into strips Need to take data from simulation in the same way as experiment (thresholds, resolution, noise etc) 22-04-15 UCL Monte Carlo Meeting, 22-04-15 10
Simulating geometry effects Average uncertainty as function of phantom spacing Calorimeter Phantom at isocentre, latest spacing from mechanical design: Distance between tracker layers now 12mm Distance between tracker stations 41mm Distance between distal and proximal parts of tracker ie gap for phantom (110 mm) Truth planes 1mm after/before detectors and 1mm before/after phantom surface 22-04-15 UCL Monte Carlo Meeting, 22-04-15 11
Mechanical offsets and alignments θ In the first case, a misalignment in x/y doesn t cause a problem for the tracking, just a loss of events. In the second case not placing opposite corners of the detector properly causes a rotational misalignment that could impair the resolution when tracking We believe that opposite sensor corners can be placed level to within 200um across the length of the sensor this is the same as having a rotational misalignment of ~2mrad. 22-04-15 UCL Monte Carlo Meeting, 22-04-15 12
Tracking data from GEANT4 u1 u2 u3 u4 Range compensator 22-04-15 UCL Monte Carlo Meeting, 22-04-15 13
Tracking data from GEANT4 TrackerGeometry Find (x,u,v) strips over threshold Find the crossing of these strips Store hits in list and process/clean up Reconstruct a position for the event Subtract reconstructed positions from truth values 22-04-15 UCL Monte Carlo Meeting, 22-04-15 14
Tracking data from GEANT4 22-04-15 UCL Monte Carlo Meeting, 22-04-15 15
Simulated tracking resolution, u1 u2 planes x y xy x y xy Ellipses indicate 1,2, & 3 sigma values for reconstructed position truth position 22-04-15 UCL Monte Carlo Meeting, 22-04-15 16
Simulated tracking resolution, u3 u4 planes x y xy x y xy Ellipses indicate 1,2, & 3 sigma values for reconstructed position truth position 22-04-15 UCL Monte Carlo Meeting, 22-04-15 17
Simulating the forward tracking The uncertainty on the projected track depends on its length, ie the distance from the reconstructed point in the detectors to the place where the proton enters the phantom surface. Forward tracking Backward tracking 125 MeV protons Phantom Phantom ~60 MeV protons Tracking uncertainty is greatest at the phantom edges where the projected tracks are longest, forward tracking before the phantom gives a resolution at the edge of ~300um and the backwards tracking ~600um for a proton beam of 125 MeV and a spread of 9.8 mrad 22-04-15 UCL Monte Carlo Meeting, 22-04-15 18
Adding Charge sharing into GEANT4 Can be caused by.. Charge diffusion Crosstalk Delta rays Large incoming angle P. Koppenburg (IPHE) Charge diffusion expected to be the dominating process 22-04-15 UCL Monte Carlo Meeting, 22-04-15 19
Charge sharing in GEANT4 The diffusion width σx for a charge carrier created at a distance Δz from the strip is: Where Θ is the thickness of the silicon, V the bias voltage and q the charge of the particle. σx ~ 6um for protons in 150um Si. The fraction of charge on any strip then becomes: Which is a quantity known as the error function. Integration carried out for all primary events in detector co-ordinates 22-04-15 UCL Monte Carlo Meeting, 22-04-15 20
GEANT4 validation Experimental equipment 1cm x 1cm x 150um silicon strip detector with 128 channels and 80um strip pitch Sensor readout with ALiBaVa motherboard. Should yield similar results to PRaVDA strip detector (150um thickness and 90.8um pitch) What we wanted to measure Pulse height the no. of electrons generated by a proton for high and low energies Cluster width the no. of strips that fire for high and low energies Both important quantities for design choices in PRaVDA strip ASIC (RHEA) 21
GEANT4 validation 22
Simulating the electronics dead strips Dead/very noisy channels from the sensor and the ASICs. ~1% (10 strips/detector) with current setup, may improve as calibration routines improve and less channels are turned off by DAQ. We want to see how the tracker performs when a certain fraction of strips is missing in each detector plane Strips over threshold Strips over threshold 23
Trackr Efficiency (useable protons) Simulating the electronics dead strips 70.00% Efficiency vs dead strips fraction 60.00% 50.00% 40.00% 30.00% 20.00% 10.00% 0.00% 0.00% 5.00% 10.00% 15.00% 20.00% 25.00% Dead strips/detector The no. of useable protons falls off rapidly with a small fractional increase in dead strips/detector due to all valid events requiring a hit in every plane. Algorithms can designed using GEANT4 data to try and recover events with missing information in some of the detectors and to see how this effects the performance of the tracker 24
Simulating the electronics noise 25
Simulating the electronics noise 26
Simulating the electronics 27
Simulating the electronics 28
Summary Lots of nice things can be done with Monte Carlo data before the real detectors are made! Simulation together with preliminary measurements has been used to inform design of a new silicon sensor, ASIC, hybrid and DAQ Simulation results used to inform and optimise mechanical design and geometry of tracker Tracking software and analysis algorithms have been designed with simulated data from GEANT4 which has been validated with measurements Simulated data now being used for CT reconstruction - using realistic mechanical offsets and geometry as well as realistic defects/noise in the electronics 22-04-15 UCL Monte Carlo Meeting, 22-04-15 29
Thanks for Listening! And many thanks to: P.Allport, G.Casse, N.A.Smith, I.Tsurin and.. 30