Results on Proton Tomography Prima – RDH – IRPT Collaboration C. Civinini1, M. Bruzzi1,2, N. Randazzo3, M. Rovituso4, M. Scaringella1, V. Sipala5,6, F. Tommasino4 1INFN - Florence, Florence, Italy 2Physics and Astronomy Department, University of Florence, Florence, Italy 3INFN - Catania, Catania, Italy 4INFN - TIFPA, Trento, Italy 5INFN - Laboratori Nazionali del Sud, Catania, Italy 6Chemistry and Pharmacy Department, University of Sassari, Sassari, Italy 7Physics Department, University of Trento, Trento, Italy 13° Trento workshop on Advanced Silicon Radiation Detector Max-Planck-Institut für Physik, München 19-21 February 2018

Why proton Computed Tomography? Hadron therapy exploits the sharp shape of the Bragg peak to precisely irradiate a tumor. But to define a threatment plan To potentially reduce inaccuracies in hadron teraphy we need: 1) Direct measurement of the 3D proton stopping power (SP) maps; 2) Patient positioning and treatment in one go. Treatment planning presently uses proton stopping power maps extracted from x-ray CTs resulting in errors on Bragg peak position up to a few millimeters. B. Schaffner and E. Pedroni Phys. Med. Biol. 43 (1998) 1579–1592 February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

How to perform a proton Tomography To mitigate the effect of the multiple Coulomb scattering we need an event-by-event measurement: 1) Tracker to measure the proton Most Likely Path (MLP)  Silicon microstrip detectors (~70 mm point resolution); 2) Calorimeter to assign an energy loss to each proton track  YAG:Ce scintillating calorimeter (~2% energy resolution @ 200 MeV); 3) Image reconstruction  Most Likely Path + Algebraic algorithms running on GPUs. Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München February 21st 2018

Tracking with multiple scattering Measurements: entry and exit positions and angles Proton true trajectory Measurements: entry position and angle L’  straight line with confidence limits L’ L’’ L  straight line with confidence limits L L’’ curved trajectory with Norrower confidence limits Measurements: entry and Exit position and angle + Most Likely Path (MLP) calculation February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Most Likely Path in a pCT geometry Starting from D.C. Williams Phys. Med. Biol. 49 (2004) and R.W. Shulte at al. Med. Phys. 35 (11) (2008) 5 cm of air have been inserted in front and behind the 20cm H2O phantom MLP example with 200MeV kinetic energy protons in 20cm of water: Entry: Y(0) = 0.2cm Y’(0) = -10mrad Exit: Y(20) = -0.1cm Y’(20) = +10mrad Silicon microstrip detectors: 320mm thick 200mm strip pitch MLP error envelope plus contributions from detector position measurement error (~ pitch/√12) and MCS inside the silicon sensors  The sensor thickness contribution affects only the MLP error at the edge of the phantom s ~ 150-250mm 200MeV in 90MeV out Errore sulla posizione del mlp errore fino a 800um nel centro a bordo fantoccio e’ dell’ordine di 200-300um February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Algebraic Reconstruction Techniques The tomographic reconstruction problem is to solve, for Sj (stopping power value in pixel j), the following set of equations: N = number of pixels; M number of protons In our case: N = (512x512x64)=16777216 voxels M = 54.3*106 events (in 40 angles) 𝑤 11 𝑆 1 + 𝑤 12 𝑆 2 + …+ 𝑤 1𝑁 𝑆 𝑁 = 𝑝 1 𝑤 21 𝑆 1 + 𝑤 22 𝑆 2 + …+ 𝑤 2𝑁 𝑆 𝑁 = 𝑝 2 ... 𝑤 𝑀1 𝑆 1 + 𝑤 𝑀2 𝑆 2 + …+ 𝑤 𝑀𝑁 𝑆 𝑁 = 𝑝 𝑀 Where: 𝑝 𝑖 ≡− 𝐸 𝑖𝑛 𝐸 𝑜𝑢𝑡 [ 𝑆 𝜌 𝐻 2 𝑂 ] 𝐸 𝐸 0 𝑑𝐸 𝑖=1,…,𝑀 February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Algebraic Reconstruction Techniques The system could be solved using an iterative formula: Sk+1= Sk + lk 𝑖∈ 𝐴 𝑘 { (pi - <wi,Sk>) wi} / ǁwiǁ2 1) S. Kaczmarz Bulletin International de l'Académie Polonaise des Sciences et des Lettres. Classe des Sciences Mathématiques et Naturelles. Série A, Sciences Mathématiques, 35, pp. 355–357 2) Gordon, R; Bender, R; Herman, GT J. Theor. Biol. (1970) 29 (3): 471–81. Sk image vector at iteration k (stopping power) wi ith track length in each pixel (vector)  Tracker + MLP pi stopping power integral (number)  Calorimeter lk relaxing factor (constant value or 0 as ~k-1) S0 initial image: {0} or approx (i.e., from FBP reconstruction). ART implementations: Ak = {single event}: ART  too much ‘salt-pepper’ noise Ak = {full data set}: SART (simultaneous ART)  better noise + parallel implementation Ak = {1/n of the full data set}: BI-SART (n-Block iterative SART)  good noise performance + faster convergence + parallel implementation February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

INFN-Prima pCT apparatus Beam test July 2017 at Trento proton Therapy Center experimental beam line Proton energy: nominal 228 MeV (214 MeV at phantom) 5x20 cm2 field-of-view February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München Tracker 4 Tracker planes 4x2 silicon microstrip p-on-n sensors (HPK + FBK production), 200mm pitch, 320mm thick 6x8 front-end chips (discriminators - 32 channels each), single channel I2C programmable thresholds designed in collaboration with INFN-Cagliari 2 levels FPGA (data reduction + event building + serial transmission) 8 data + clk serial lines (1.6Gbs) Central DAQ board ML605 Xilinx Virtex 6 development board (1GB DDR3 memory + 1Gbs Ethernet interface + I2C + RS232)  3° level FPGA FMC Interface board (serial receivers + handshake + I2C+ spares) February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München Tracker plane X-side view February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München Tracker architecture 2° level FPGA 1° level FPGA Front-end chip INFN-Ca 3° level FPGA Interface board Silicon sensor Front-end chips 1° level FPGA February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München Calorimeter 2x7 YAG:Ce crystals 3x3x10 cm3 70 ns scintillating light decay time Hamamatsu 18x18 mm2 photodiodes (S3204) Analogue amplifier + shaper (1ms) NI-5751 ADC (14 bits - 5 MHz sampling) Main trigger generator + sync info Independent crystal trigger logic Neighborhood merging (4-6 elements) 7 bit event number to tracker for synchronism February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Calorimeter Photodiodes Calorimeter front face Calorimeter front-end board Calorimeter front face 2x7 YAG:Ce Crystals Array 3x3x10cm3 February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München Phantoms Two CIRS phantoms Electron density calibration phantom 18 cm diameter, 5 cm height: Water equivalent container + 9 plugs Anthropomorphic phantom Human head with titanium spine prosthesis and tungsten dental filling Motorized platform Inserted between plane 2 and 3 Phi and Z movements (Physik Instrumente, linear and step motors) RS232 remotely controlled February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Electron density phantom 1 2 3 4 5 6 7 8 8 lateral plugs: Liver 1.07 gcm-3 Lung exhale 0.50 gcm-3 Breast 0.99 gcm-3 Bone 1.53 gcm-3 Muscle 1.06 gcm-3 Bone 1.16 gcm-3 Adipose 0.96 gcm-3 Lung inhale 0.20 gcm-3 Central plug: 6 mm diameter Titanium core 4.51 gcm-3 February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Anthropomorphic phantom X-Rays radiographies Adult male model Dental filling in a molar (Tungsten) Spine prosthesis attached at C3-C5 vertebra (Titanum) Tomography region February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München Results Two sets of data for tomography reconstruction Electron density phantom 1.2x108 events in 40 angles Anthropomorphic phantom 2x108 events in 40 angles Two energy scan for calorimeter calibration 6 energies 5x106 events each (228, 202, 169, 143, 112, 83) MeV Two high statistics alignment run 107 events each at 228 MeV February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Electron density phantom tomography x pixel (x400mm) y pixel (x400mm) MeV/cm Radial artifacts influence the quality of the image This is a known behaviour of the algebraic reconstruction algorithms  working on that issue Nonetheless the plugs are (almost) visible and a quantitative analysis has been done  next slide February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Relative Stopping Power (RSP)@180 MeV The Titanium RSP measured value is underestimated by 13%: it has been removed from the plot (core too small, value influenced by track resolution) Hard Bone Lung Adipose, Water, Breast, Muscle, Liver Trabecular bone Relative Stopping Power (RSP) = Stopping Power/measured Water Stopping Power Errors: RSP meas. = 1.3% (from water measurements), RSP expected = 1% RSP estimation from: E. Schnell, S. Ahmad, and T. de la Fuente Herman AIP Conference Proceedings 1747, 110004 (2016) February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Anthropomorphic phantom tomography 64 axial slices voxels: 600x600x812mm3  0.3 mm3 Algebraic reconstruction 25° iteration 40 angles (9 deg. uniform spacing) About 5.4 x 107 events used First and last slices have low statistics (outside field of view) Movie starts from lower jaw ends at upper teeth (5.2 cm range) Much lower contrast with respect to the X-Ray images mainly because of the physics of the interactions: important Z dependence for X-Ray, density dependence for protons  BUT it is what is needed for hadron therapy February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Anthropomorphic phantom tomography February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Anthropomorphic phantom tomography Front (coronal) view X pixels [0.6mm] Lateral (sagittal) view Stopping power [MeV/cm] Z pixels [0.812mm] y pixels [0.6mm] Front (coronal) view X pixels [0.6mm] Lateral (sagittal) view y pixels [0.6mm] February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München Conclusions A silicon tracker / scintillator calorimeter, 5x20 cm2 field of view, Computed Tomography apparatus has been tested in a 228 MeV proton beam for pre-clinical studies; Tissue equivalent non-homogeneous phantoms have been used for tomographic data taking; Measured RSP values agree with the expected ones; Algebraic iterative reconstruction algorithms have been implemented to run on GPU; Three dimensional images with 16.8M voxels with size of 600x600x812mm (0.3 mm3) have been reconstructed; More tests during 2018 are foreseen. February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München Backup slides February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

proton Computed Tomography: principle Proton Radiotherapy → first proposed by R.R. Wilson in 1946 "Radiological Use of Fast Protons", Radiology, 47:487-491 (1946) Advantage : Highly conformational dose distribution: i) lower dose to healthy tissues in front of tumor; ii) healthy tissues beyond it are not damaged; Inaccuracies: Treatment planning presently performed by X-CT → expected errors typically of a few millimeters B. Schaffner and E. Pedroni Phys. Med. Biol. 43 (1998) 1579–1592 pCT Direct measure of the 3D stopping power maps using protons Precision improvement when positioning and treatment are made in one go February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Algebraic Reconstruction Techniques Iterative algorithm to reconstruct tomographic images (proton stopping power maps) from a set of single proton events Starting point (S(x,y,E) stopping power): −𝑑𝐸=𝑆 𝑥,𝑦,𝐸 𝑑𝑙 Introducing the mass stopping power S/r : − 𝑆 𝜌 𝑥,𝑦, 𝐸 0 𝑑𝐸= 𝑆 𝜌 𝑥,𝑦, 𝐸 0 𝑆 𝜌 𝑥,𝑦,𝐸 𝜌 𝑥,𝑦 𝑑𝑙 E0 being a reference energy at which the SP is calculated Dividing by S/r at energy E: − 𝑆 𝜌 𝑥,𝑦, 𝐸 0 𝑆 𝜌 𝑥,𝑦,𝐸 𝑑𝐸=𝑆 𝑥,𝑦, 𝐸 0 𝑑𝑙 February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Algebraic Reconstruction Techniques The left hand side doesn’t depend too much on the material composition (≤6*10-3) and could be replaced by the one measured for liquid water (NIST pstar tables - http://physics.nist.gov/PhysRefData/Star/Text/PSTAR.html): −[ 𝑆 𝜌 𝐻 2 𝑂 ] 𝐸 𝐸 0 𝑑𝐸=𝑆 𝑥,𝑦, 𝐸 0 𝑑𝑙 Where [ 𝑆 𝜌 𝐻 2 𝑂 ] 𝐸 𝐸 0 ≅[ 𝑆 𝜌 (𝑥,𝑦 ] 𝐸 𝐸 0 = 𝑆 𝜌 𝑥,𝑦, 𝐸 0 𝑆 𝜌 𝑥,𝑦,𝐸 . February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Algebraic Reconstruction Techniques Integrating along the proton path: − 𝐸 𝑖𝑛 𝐸 𝑜𝑢𝑡 [ 𝑆 𝜌 𝐻 2 𝑂 ] 𝐸 𝐸 0 𝑑𝐸= 𝑝𝑎𝑡ℎ 𝑆 𝑥,𝑦, 𝐸 0 𝑑𝑙 D. Wang, Med.Phys. 37 (8) (2010) 4138-4145. Ein is given by the accelerator, Eout by the calorimeter and the ‘path’ by the tracker (Most Likely Path) Subdividing the object into a set of pixels, for the ith proton: 𝑝 𝑖 ≡− 𝐸 𝑖𝑛 𝐸 𝑜𝑢𝑡 [ 𝑆 𝜌 𝐻 2 𝑂 ] 𝐸 𝐸 0 𝑑𝐸= 𝑗=1 𝑁 𝑤 𝑖𝑗 𝑆 𝑗 𝐸 0 𝑖=1,…,𝑀 Where wij is the path length of proton i inside the pixel j February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München Pixel 1 Pixel j p out 90 MeV Phantom: 20 cm of water Computational challenge: find the simplest (fastest) way to build the wij matrix (1015 elements, only 3x1010 ≠ 0 ) p in 200 MeV wij Pixel N February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München Motorized platform To be integrated in the DAQ to automatize data taking Seen from below Off-center axis (5 cm) to allow tomographies of large objects February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Relative Stopping Power (RSP)@180 MeV Tissue Measured SP (MeV/cm) RSP measured (H2O meas) RSP expected (Schnell et al.) RSP difference (%) Water1 4,628 0,991 1 0,85 Water2 4,744 1,016 -1,64 Water3 4,682 1,003 -0,31 Water4 4,616 0,989 1,10 Water mean 4,667 1,000 0,00 Lung inhale 0,889 0,190 0,192 Lung exhale 2,502 0,536 0,496 -8,08 Adipose 4,560 0,977 0,965 -1,25 Breast 4,741 0,99 -2,60 Muscle 4,952 1,061 1,056 -0,47 Liver 5,009 1,073 1,065 -0,76 Bone 200 HA 5,240 1,123 1,111 -1,04 Bone 800 HA 6,636 1,422 1,415 -0,48 Titanium 13,160 2,820 3,243 13,06  Possible overestimation (bias) on RSP from artefacts Relative Stopping Power (RSP) = Stopping Power / measured Water Stopping Power RSP estimation from: E. Schnell, S. Ahmad, and T. de la Fuente Herman AIP Conference Proceedings 1747, 110004 (2016) February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Large area system: radiography Calorimeter At ‘Centro Protonterapia’ Trento, Italy - October 2016 Intermediate step to full Tomography Two tracker planes and Calorimeter CIRS Antropomorfic phantom Tracker planes Two tracker planes to extrapolate the exiting proton trajectory to the calorimeter for better energy resolution and back to the head mean transverse plane for image reconstruction. Beam energy to the phantom : 180 MeV. About 107 protons per slice  90 mGy February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Large area system: radiography Stopping power integral map normalized to 180 MeV (same quantity used for Tomographic Reconstruction): Pixel size 1.5mm Beam energy 180 MeV No MLP Many thanks to TIFPA, IBA and ‘Centro Protonterapia’, Trento February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München

Large area system: radiography Residual energy map: Pixel size 1.5mm Beam energy 181 MeV Dose 90 mGy February 21st 2018 Carlo Civinini. INFN Firenze 'Trento Workshop' @ MPI-München