presentation slides - DIR-Lab
Transcription
presentation slides - DIR-Lab
Hierarchical Particle Transport in Radiation Therapy Applications Dragan Mirkovic, Ph.D. Department of Radiation Physics D. Mirkovic, Rice University Seminar, Spring 2014 Motivation • Cancer represents a huge socioeconomical burden in US and the World • Cancer treatment – Surgery – Chemotherapy – Radiation therapy • Treatment outcomes have improved but are still far away form the cure D. Mirkovic, Rice University Seminar, Spring 2014 Motivation (cont.) D. Mirkovic, Rice University Seminar, Spring 2014 Motivation (cont.) D. Mirkovic, Rice University Seminar, Spring 2014 Introduction • Radiation therapy (RT) has proven to be an effective form of treatment in many cancer types • The main goal of RT is to maximize tumor control probability (TCP) while minimizing the normal tissue complications probability (NTCP) – Most of the biological effects of radiation are stochastic – Biological effects are proportional to the dose • 1st approximation D. Mirkovic, Rice University Seminar, Spring 2014 Introduction • The 1st approximation of the problem – Maximize the dose to the tumor while minimizing the dose to normal tissue – Approximate the all other dependencies using a single fudge factor called RBE • Most of the technical innovations so far have been directed towards improving the dose conformity in treatment planning – More conformal modalities: protons and heavy ions D. Mirkovic, Rice University Seminar, Spring 2014 Difficulties • Problem is very complex – Cancer is a cell disease – Biological effects of radiation span multiple time and space scales – Accurate solutions based on 1st principles would be very time consuming if not impossible – Need good models and accurate approximations D. Mirkovic, Rice University Seminar, Spring 2014 Modeling mantra • “Essentially, all models are wrong, but some are useful.” – George E.P. Box, 1987* *Box, G. E. P., and Draper, N. R., (1987), IN: Empirical Model Building and Response Surfaces, p. 424, John Wiley & Sons, New York, NY. D. Mirkovic, Rice University Seminar, Spring 2014 Problem complexity • A human body of 70 kg is comprised of: • ~6.7⋅1027 atoms and ~1014 cells • Activity: • ~1025-1026 molecular reactions per day in human body • ~10,000 DNA single strand breaks/base loss per cell per day • Blood and small intestinal cells produced per day: ~1010-1011 • A human cell is 20µm in diameter and consists of • ~65% water, 20% proteins, 12% lipids, ~1% RNA and 0.1– 3% DNA • No. of genes in the human body: ~20–25,000 D. Mirkovic, Rice University Seminar, Spring 2014 Biological effects of radiation Physics Chemistry Biology D. Mirkovic, Rice University Seminar, Spring 2014 Introduction proton therapy • Proton therapy provides a superior dose distribution – Better local control of tumors – Better sparing of normal tissue • It has taken a long time for protons to enter the mainstream of radiation therapy – The idea has been around for more then 50 years • Why? – Technology is complex, expensive and there are many challenges – Clinical benefits have not yet been proven in randomized clinical trials (except for ocular treatment) • How to bring protons closer to the mainstream of RT – Development, research, and implementation D. Mirkovic, Rice University Seminar, Spring 2014 Proton beam physics • Proton treatment offers many advantages over classical photon therapy – Finite range - no exit dose – Sharp lateral penumbra – Bragg peak - localized dose deposition – Proton dose distributions can be made to conform tightly to irregular target shapes in all three dimensions – Clinically, the radiobiology of proton beams is almost identical to that of photon beams – Lower normal tissue dose for the same therapeutic dose D. Mirkovic, Rice University Seminar, Spring 2014 Physics Koehler 1972 D. Mirkovic, Rice University Seminar, Spring 2014 Radiation beam quality Ar Si C Pions 250 kV X-rays 60Co 22 MeV X-rays RBE LET Ne He H Quality of dose distribution Based on Fowler (1981) D. Mirkovic, Rice University Seminar, Spring 2014 Energy loss of protons • Interaction with atomic electrons – – – – – Almost no deflection Small energy loss in individual collision Continuous slowing down approximation Bethe-Bloch theory (1930-33) Stopping power: S/ρ = 1/ρ[dE/dx] ~ 1/v2 – Results in a narrow Bragg peak “broadened” by two effects: 1. The incident beam has a narrow energy spread (not monoenergetic) 2. Range straggling caused by statistical differences in energy losses in individual proton paths. D. Mirkovic, Rice University Seminar, Spring 2014 Energy deposition • Energy range: 72 – 221 MeV • Monte Carlo Simulations using MCNPX • Compared to measurements • Useful for scanning beam comissioning D. Mirkovic, Rice University Seminar, Spring 2014 Multiple Coulomb Scattering • Protons are deflected frequently in the electric field of the nuclei – Small deflection angles • Beam broadening can be approximated by a Gaussian distribution • Theory: – Moliere, Bethe, Scott (1940-63) D. Mirkovic, Rice University Seminar, Spring 2014 Nuclear interactions of protons • A certain fraction of protons undergo nuclear interactions, mainly on 16O • Nuclear interactions lead to secondary particles and thus to local and non-local (neutron) dose deposition Pedroni et al PMB 2005, 2007 D. Mirkovic, Rice University Seminar, Spring 2014 Beam delivery systems • Passive scattering – Uses double scattering system with range modulation to produce distal edge conforming dose in the patient – Needs patient specific aperture and range compensator for each beam – Neutrons are produced in the first and second scatterers, range modulation wheel, aperture, range compensator, and in the patient • Dynamic scanning – Uses two magnets to scan a narrow proton beams of different energies to cover the target volume – Scanning beam has almost no neutrons generated outside of the patient D. Mirkovic, Rice University Seminar, Spring 2014 Passive scattering nozzle RMW Brass aperture Range Compensator Smith, AAPM 2008 D. Mirkovic, Rice University Seminar, Spring 2014 Pencil Beam Scanning Nozzle • Performance: • Range 4–36 g/cm2 • Adjustability 0.1 g/ cm2 • Max. field size 30 x 30 cm • Beam size in air 6 – 10 mm σ • SAD = 2.7 m D. Mirkovic, Rice University Seminar, Spring 2014 MDACC Proton Center 3 Rotating Gantries 1 Fixed Port 1 Eye Port 1 Experimental Port Accelerator System (slow cycle synchrotron) D. Mirkovic, Rice University Seminar, Spring 2014 Synchrotron D. Mirkovic, Rice University Seminar, Spring 2014 Beam transport line D. Mirkovic, Rice University Seminar, Spring 2014 Gantry factory tests 13 m diameter 190 tons SAD ! 2.7 m D. Mirkovic, Rice University Seminar, Spring 2014 Treatment room D. Mirkovic, Rice University Seminar, Spring 2014 Mathematical description • Charged particle transport through a prescribed background medium – Described by a linear kinetic or transport equation • Simple particle balance in phase space (Liouville’s theorem) – Assume continuous, isotropic, stationary medium – May need separate solutions on various length and time scales D. Mirkovic, Rice University Seminar, Spring 2014 Space and Time Scales D. Mirkovic, Rice University Seminar, Spring 2014 Particle interactions (r, v, t)2 Particle (r, v, t)1 θ Medium • Interactions described with a linear integral operator whose kernel indicates the probability of scattering from the state (r, v, t)1 to the state (r, v, t)2 • It is often more practical to use (r, E, Ω, t) instead of (r, v, t)1 D. Mirkovic, Rice University Seminar, Spring 2014 The Transport Equation • Let f (r, E, Ω, t) denote the density of particles in phase space, such that dn = f (r, E, Ω, t)drdEdΩ • Is the number of particles at time t in a six dimensional infinitesimal volume drdEdΩ . • Particle interact by absorption and scattering specified by the respective cross sections, sa and ss such that in traveling a distance ds, a particle has a probability of absorption pa and scattering ps given by ps = σ s (r, E, t)ds pa = σ a (r, E, t)ds D. Mirkovic, Rice University Seminar, Spring 2014 The Transport Equation • Assume isotropic medium – sa and ss are not functions of W • Absorption removes particle from the beam • Scattering changes the particle energy and direction – Define the scattering kernel as σ s (r, E, E !, Ω! ⋅ Ω, t) , such that in traveling a distance ds, the probability ps’s of a particle with energy E’ and direction W’ before the collision, scattering to an energy E and direction W after the collision is given by ps!s = σ s (r, E, E !, Ω! ⋅ Ω, t)dsdEdΩ – W’ . W is the cosine of the scattering angle D. Mirkovic, Rice University Seminar, Spring 2014 The Transport Equation D. Mirkovic, Rice University Seminar, Spring 2014 Numerical approximations • The problem has 7 dimensions • Numerical methods – based on discretization using FEM and FD are very rare – Monte Carlo very popular • Simple modeling of geometry and physics • Caveat: may need a large number of events in order to achieve desired numerical uncertainty D. Mirkovic, Rice University Seminar, Spring 2014 Multiple scales • Macroscopic dose calculations – Domain size in ~0.5 m – Patient discretization in mm (1 – 5 mm voxel size) – Millions of voxels • Microscopic effects captured using track structure simulations and RBE models – Domain size in microns – Resolution in nm D. Mirkovic, Rice University Seminar, Spring 2014 Monte Carlo for Protons • Motivation: – Proton treatment planning is usually done using a simplified, pencil beam (PB) algorithm for dose calculation – PB is inaccurate in some cases • High heterogeneities • Distal edge degradation – Neutrons are neglected completely – Are the errors and approximations clinically significant? • Analysis of treatment failures and excessive toxicities – Can we improve the treatment? D. Mirkovic, Rice University Seminar, Spring 2014 Monte Carlo Implementation • Use in-house made Monte Carlo Treatment planning systems: – MCPRTP (Newhauser et al.), MC2 (Mirkovic & Titt) • Monte Carlo (MC) calculations provide superior accuracy • MC requires a detailed, patient specific geometrical model • Use DICOM files exported from TPS to build the input for MC code D. Mirkovic, Rice University Seminar, Spring 2014 MC model for passive scattering nozzle c e d a b f (a) range-modulator wheel, (b) scattering foils, (c) range-shifting plates, (d) block (aperture), (e) range compensator, and (f) voxelized phantom. Taddei P, Mirkovic D, Fontenot JD, Giebeler A, Zheng Y, Titt U, Woo S, Kornguth, D and Newhauser WD. Estimated risk from stray neutron dose for proton craniospinal irradiation to a pediatric patient represented by a voxelized phantom. PMB (In press). D. Mirkovic, Rice University Seminar, Spring 2014 Biological effects of protons • Biological effects of ionizing radiation are consequence of a complex sequence of events – Unreparable DNA damage is the most important • Modeled using relative biological effectiveness (RBE) – Depends on many factors D. Mirkovic, Rice University Seminar, Spring 2014 DNA Damage Low HighLET LET D. Mirkovic, Rice University Seminar, Spring 2014 Radiobiological models • Need to predict a complex sequence of molecular, cellular, and tissue responses to initial radiation damage – Relative biological effectiveness (RBE) models Radia+on RBE Dose • • • • • energy linear energy transfer (LET) dose per treatment frac+on +ssue type clinical endpoint, … Effect Variable RBE for protons • Constant value of 1.1 is used in clinical practice today • There is experimental evidence that the RBE is not constant, but depends on dose and on the LET – In vitro experiments confirm the dependence on LET • Blakely et al 1984, Belli et al 1993, Wouters et al 1996, Skarsgard 1998 • In vivo experiments show much smaller variations and confirm 1.1 for clinical use – most in vivo studies were done at low LET - entrance region or in mid SOBP • Gueulette et al 2001, Paganetti et al 2002, Mason and Gillin 2007. • Measured values have insufficient resolution of RBE dependence on LET – Many authors use physics based phenomenological models (Wilkens and Oelfke 2004, Frese 2009, Stewart 2009) D. Mirkovic, Rice University Seminar, Spring 2014 MC Dose D. Mirkovic, Rice University Seminar, Spring 2014 Eclipse Dose D. Mirkovic, Rice University Seminar, Spring 2014 MC LET D. Mirkovic, Rice University Seminar, Spring 2014 LET profile along the beam axis D. Mirkovic, Rice University Seminar, Spring 2014 RBE – LET Relationship Typical LET range for proton therapy RBE vs. LET. The data is from a number of experiments using a number of ions, energies and cell types. The shaded area shows the general trend of the data. (Blakely 1984) D. Mirkovic, Rice University Seminar, Spring 2014 Phenomenological models • Wilkens and Oelfke 2004 – Use LQ model for biological response S = exp(-aD-bD2) – Two survival curves • Reference (photon) and proton – RBE = ratio of doses for the same effect ap(L) = a0 +lL, bp(L)=bx, a0 = ax S [-] – Only ap depends on LET (LSM) 0 10 Increasing LET -5 10 0 D. Mirkovic, Rice University Seminar, Spring 2014 2 4 6 Dose [Gy] 8 10 In the limit of very small and very large doses the RBE is proportional to / respectively. x and ⇥/⇥x , Effect of l on proton RBE Frese, 2011 Figure 2.2: The influence ofRice the parameter of the LSM on the2014 proton RBE. A value D. Mirkovic, University⇤ Seminar, Spring obtained from in vitro experiments with V79 cells (⇤ = 0.02 µm/(keV · Gy)) is compared with a second value (⇤1.1 = 0.008 µm/(keV · Gy)) reflecting the clinical experienced average Clinical application • Use reference a and b parameters for different tissue types from literature – Example: lung TRP (Seppenwoolde 2003) a=0.0258 Gy-1 and b=0.0065 Gy-2 (a/b = 4) • The parameter l is not so easy to determine – Fit the value of l using experimental data – Compute the value of l such that the mean value of the RBE inside the target is 1.1 (Frese 2009) – Compute the value of l such that mid SOBP RBE matches values form literature – Here we use l = 0.008 mm keV-1 Gy-1 (Frese 2010) D. Mirkovic, Rice University Seminar, Spring 2014 RBE D. Mirkovic, Rice University Seminar, Spring 2014 RBE profile along the beam axis D. Mirkovic, Rice University Seminar, Spring 2014 Variable RBE dose D. Mirkovic, Rice University Seminar, Spring 2014 Constant vs. variable RBE dose D. Mirkovic, Rice University Seminar, Spring 2014 Dose difference D. Mirkovic, Rice University Seminar, Spring 2014 Collaboration areas • Fast and accurate solutions for radiation transport problems – Fast MC – Numerical PDE methods • Efficient geometry modeling using medical imaging data (CT or MRI) – Grid construction – Multiple length scales • Medical image processing D. Mirkovic, Rice University Seminar, Spring 2014 Credits and acknowledgments • Collaborators: Uwe Titt, Radhe Mohan, Pablo Yepes, Philip Taddei, Yuanshui Zheng, Annelise Giebeler, Jonas Fontenot, Wayne Newhauser, Gabriel Sawakuchi, Lei Dong, Luis Perles The University of Texas M.D. Anderson Cancer Center, 1515 Holcombe Blvd., Unit 94, Houston, TX, 77030, USA • This work was supported in part by – NCI P01 Award MGH + MDACC – Northern Illinois University through a subcontract of DOD contract W81XWH-08-1-0205 – Varian medical systems MRA D. Mirkovic, Rice University Seminar, Spring 2014