Shirin A. Enger Université Laval

Geant4 in Medical Radiation Physics: One Physicist's Perspective on Why This is the Monte Carlo Code to Choose

1st New Mexico Workshop on Monte Carlo for Particle Therapy Treatment Planning Albuquerque, New Mexico 16-18 May 2011

Outline

•  General introduction of Geant4 •  Basic concepts and User classes

•  Geant4 in my research

•  Geant4 in medical radiation physics •  PTSsim

•  Topas

•  Gamos

•  Geant4-DNA

GEANT4: GEometry ANd Tracking 4

•  successor of GEANT3

•  Object-Oriented environment

•  C++

•  Data and functions tied together Easier to manage

complexities

•  Application areas:

•  High Energy Physics

•  Nuclear Physics

•  Space applications

•  Medical Physics

Mandatory user classes

G4VUserDetectorConstruction

G4VUserPhysicsList G4VUserPrimaryGeneratorAction

Optional user classes G4UserRunAction

G4UserEventAction

G4UserStackingAction G4UserTrackingAction G4UserSteppingAction

User must provide her own classes derived from the following three mandatory abstract classes and register them to the G4RunManager

The user can customize the default functionality of GEANT4 simulation by making her own classes derived from the following 5 optional user classes.

G4RunManager : controls all of state changes •  manages the event loop(s) within a run •  Responsible for managing initialization procedures

Geant4 does not provide the main().

• In main():

•  Construct G4RunManager

•  Set user mandatory classes to RunManager

•  Set optional user action classes to RunManager

•  Define VisManager, (G)UI session …

Detector construction (G4VUserDetectorConstruction) •  Three layers

•  G4VSolid •  G4LogicalVlolume •  G4VPhysicalVolume

G4Box

G4Tubs

G4VSolid G4VPhysicalVolume

G4Material

G4VSensitiveDetector

G4PVPlacement

G4PVParameterised

G4VisAttributes

G4LogicalVolume

•  G4VSolid: •  shape •  size

CSG (Constructed Solid Geometry )

Solids

BREP (Boundary REPresented)

Solids

Boolean Solids

Predefined shapes Defined via description of boundaries Boolean combinations of CSGs and BREPs

G4Box,G4Cylinder, G4Sphere, G4Cone, G4Polycone,

G4Polyhedra,…

G4BREPSolidPolycone, G4BSplineSurface, Complex geometries using CAD programs

G4UnionSolid, G4SubtractionSolid,

…

3 types of shapes

•  G4LogicalVolume: •  Shape and dimension (G4VSolid) •  Material, sensitivity, visualization attributes •  Position of daughter volumes •  Magnetic field •  Shower parameterization

•  G4VPhysicalVolume: •  position •  Rotation

The user has the choice to place a physical volume: •  once in its mother volume (G4PVPlacement) •  Position the same logical volume many times through

Replicas and Parameterized Placements •  resulting in large savings in memory

In medical applications, such parameterized placement forms the most efficient way to built a patient CT geometry

The thousands of voxels that represent the patient geometry are represented as just a single logical volume (one CT voxel) placed thousands of times, varying only in material and location

•  Once the geometry has been built, particles are tracked through the geometry by “Navigator” classes

•  Navigators handles geometrical work •  determining the next traversed volume •  obtaining distance to this volume

•  Geant4 includes several different navigators, some appropriate for any volume arrangement, others designed specifically for optimal performance in simple voxel geometries

•  Particle source (G4VUserPrimaryGeneratorAction) •  Source dimensions •  Energy spectrum •  Angular distribution

Primary vertices and primary particles are stored in

G4Event in advance to processing an event

MyPrimaryGenerator  (G4VUserPrimaryGeneratorAc5on)

Descrip5on  of  desired    primary  proper5es  

MyPar5cleGun  (G4VPrimaryGenerator)) Crea5on  of  ver5ces    And  primary  par5cles    

G4Event  

Storage  of  primaries  for  later  tracking  

•  Physics list (G4VUserPhysicsList) •  Needed particles to be transported •  Physics processes assigned to the particles

•  electromagnetic, hadronic, and decay categories

•  Production treshholds (applied to the world or region by region)

Process: •  Decides when and where an interaction will occur

using GetPhysicalInteractionLength (GPIL) method •  Needs cross sections

•  Generates the final state using DoIt method •  Needs interaction model

•  Three types of GPIL, DoIt methods •  PostStep, AlongStep, AtRest

Physics Processes Provided by Geant4 •  Electromagnetic

•  standard – complete set of processes covering charged particles and gammas energy range 1 keV to ~PeV

•  low energy – specialized routines for e-, γ, charged hadrons

•  more atomic shell structure details •  some processes valid down to 250 eV or below •  others not valid above a few GeV

•  optical photon – only for long wavelength photons (x-rays, UV, visible)

•  processes for reflection/refraction, absorption, wavelength shifting, Rayleigh scattering

•  Weak physics •  decay of subatomic particles •  radioactive decay of nuclei

•  Hadronic physics •  pure hadronic processes valid from 0 to ~TeV

•  elastic •  inelastic •  capture •  Fission

•  electro- and gamma-nuclear valid from 10 MeV to ~TeV

•  In Geant4 all particles are tracked down to zero kinetic energy unless a tracking cut is specified

•  There is however a production cut or production threshold below which no secondary are generated

•  This threshold is a range cut-off (distance) to avoid the dependence on particle type and material and is internally converted to an energy cut for individual materials

•  If the user proposes a range corresponding to smaller energy than 250 eV, it is ignored

Optional classes: •  G4UserRunAction

•  Define histograms, Analyze the run, Store histograms •  G4UserEventAction

•  Event selection, Output event information •  G4UserStackingAction

•  Scoring, kill / suspend / postpone the track •  G4UserTrackingAction

•  Decide trajectory should be stored or not, Create user-defined trajectory, Delete unnecessary trajectory

•  G4UserSteppingAction •  Scoring, kill / suspend / postpone the track

Run in Geant4 is a collection of events which are produced under identical conditions

•  Run begins with the command “BeamOn”

•  At beginning of run: •  Geometry is optimized for navigation •  Cross sections are calculated according to materials in the

setup •  Cutoff values are defined

•  Event contains primary particles (from generator, particle gun, ...), which are pushed into a stack at beginning of processing

•  Each particle is popped from the stack and tracked during processing,

•  When the stack is empty, processing of the event is over

•  The class G4Event represents an event •  At the end of processing it has the following objects:

•  List of primary vertices and particles (the input) •  Hits collections •  Trajectory collections (optional) •  Digitizations collections (optional)

•  A track is a snapshot of a particle within its environment •  as the particle moves, the quantities in the snapshot change •  at any particular instance, a track has position, physical

quantities •  it is not a collection of steps

•  Track object lifetime •  created by a generator or physics process (e.g. decay) •  deleted when it:

•  leaves world volume •  disappears (particle decays or is absorbed) •  goes to zero energy and no “at rest” process is defined •  user kills it

•  No track object survives the end of an event (not persistent) •  User must take action to store track record in trajectory

•  The step is the basic unit of simulation •  Has two points (pre-step, post-step) •  Contains the incremental particle information (energy

loss, elapsed time, etc.) •  Step size Is determined by

•  Volume boundaries •  Physics processes

•  Each point contains volume and material information

Geant4 in my research •  I have used Geant4 since 2005/2006 •  Why Do I prefare to work with Geant4? •  Examples of Geant4 in my research: 2006: •  Monte Carlo calculations of thermal neutron

capture in gadolinium: A comparison of GEANT4 and MCNP with measurements

•  Aim: To compare MCNP and GEANT4 with experimental results and select the suitable code for gadolinium neutron capture applications

Enger SA, af Rosenschold PM, Rezaei A, Lundqvist H. Monte Carlo calculations of thermal neutron capture in gadolinium: a comparison of GEANT4 and MCNP with measurements. Med Phys. 2006 Feb;33(2):337-41.

•  Thermal neutron scattering from nuclei within chemically bound atoms was missing in Geant4

•  Only free gas model was availabe

•  At thermal neutron energies, atomic translational motion as well as vibration and rotation of the chemically bound atoms affect the neutron scattering cross section and the energy and angular distribution of secondary neutrons.

•  The energy loss or gain of incident neutrons can be different from interactions with nuclei in unbound atoms.

•  Status of Low energy neutron transportation package of Geant4(below 20MeV) at that time: •  Neutron High Precision models and cross sections •  Data driven models •  Geant4 neutron data library(G4NDL) were derived based

on the following evaluated data libraries Brond- 2.1, CENDL2.2, EFF-3, ENDF/B-VI.0, 1, 4, FENDL/E2.0, JEF2.2, JENDL-FF, JENDL-3.1,2.

•  Elastic, Inelastic, Capture and Fission models and cross sections were prepared

•  On-the-fly Doppler broadening of cross sections were carried out

•  Only individual Maxwellian motion of the target nucleus was taken into account (Free Gas Model)

Double differential scattering cross section for thermal neutrons:

•  E and E‘ are the incident and secondary neutron energies • Ω is the scattering angle • σb is the bound scattering cross section for the material •  kT is the temperature in eV •  S(α,β) is the thermal scattering law

•  A is the ratio of the mass of the scattering atom • M to the neutron mass.

A neutron facility at Studsvik was used in experiments

Epithermal neutron spectrum from region III were used in the calculations

A point at 3 cm depth in the PMMA phantom on the beam centerline was chosen to be the reference point.

• The fluence was measured and calculated in x, y and z positions.

• The MCNP calculated fluence agreed well with measurements.

Conclusion:

Cross section for thermal neutron scattering from chemically bound atoms is necessary for reliable thermal neutron capture dosimetric studies. GEANT4 is not reliable for neutron capture calculations as long as these cross sections are missing.

•  A couple of months later: •  Dr. Tatsumi koi put the S(a,b) cross section in

Geant4 and made this comparison with our results:

Dr Tatsumi’s Conclusions: •  We successfully developed a Geant4 version of the

S(α,β) model of thermal neutron scattering from nuclei within chemically bound atoms.

•  Validation against the BNCT clinical beam data was done and good agreement with the measurement and previous calculation was seen.

•  This model and cross section will be included in the next release of Geant4. It will then be more suitable for Neutron Capture Therapy applications.

•  Low energy electromagnetic and hadronic physics packages and cross sections of Geant4 has been under continuous maintenance and development, and mostly driven by the requirements from users medical and space users

•  We have used electromagnetic package of Geant4 in different applications for many years and have noticed the improvments of the models ans cross sections for each version

•  Next example: Geant4 EM

•  Cross-fire doses from ß-emitting radionuclides in targeted radiotherapy. A theoretical study based on experimentally measured tumor characteristics.

•  AIM: Estimation of inter-cluster cross-fire radiation dose from different β-emitting radionuclides in a breast cancer model

Enger SA, Hartman T, Carlsson J, Lundqvist H. Cross-fire doses from ß-emitting radionuclides in targeted radiotherapy. A theoretical study based on experimentally measured tumor characteristics. Phys. Med. Biol. 2008 April; 53:1909-1920.

Typical immunohistochemical HER2 stainings. A section from a breast tumor lymph node metastasis (a) and a section from a primary breast tumor (b) are shown. The tumor cell clusters, containing tumor cells with brown-stained cell membranes within both the lymph node metastases and the primary tumor, can be easily identified.

(a) (b)

•  Sevenβ-emitting radionuclides •  Homogeneous radioactivity distribution •  Spheres with 15, 25, 50 µm radius

The Livermore model

•  The Livermore model utilises the evaluated libraries (EPDL97)

•  extends the range of accuracy for the simulation of EM interactions down to 250 eV

•  It is designed for any application that requires higher accuracy of electrons, hadrons and ion tracking without magnetic field

•  For multiple scattering a step function is used to compute the mean energy loss per step

•  The lowest cut with the Livermore physics is 250 eV, which correspond to a range of ~ 16 nm in water

Scandidos Delta4

Geant4 calculations in brachytherapy •  Brachytherapy is a cancer treatment modality where

encapsulated radiation source are placed directly into or near the tumor

•  Brachytherapy sources: •  needles, tubes, seeds, wires and pellets

•  Accurate knowledge of dose distribution around these sources is necessary in order to provide a more solid basis for clinical strategy

Overlapping of the radiation sources with the patient geometry has been a challenge

In our group we do post-implant dosimetry for patients with 125I permanent seed implants

•  a computed tomography (CT) scan is performed on the patient about 4 weeks after implantation with slice thickness of 2.5 mm

•  The seeds are detected on the CT images and X-ray films for accurate seed counts

•  The organs are contoured by the treating physician on the CT dataset

•  A detailed post-implant dose distribution is obtained with a dose calculation engine in our case Monte Carlo caculations with Geant4

Until recently: •  Patient CT voxels that

included the radiation source was subtracted

•  The source was placed into the resulting voids

•  The remaining space of the voids was filled with water

The code was unable to take advantage of the faster and more memory efficient voxel-aware geometry navigators (that require a simple voxel geometry)

Several years ago, Geant4 introduced a concept called "Parallel Geometry”

•  Allowing the user to define additional geometrical hierarchies that overlay the standard "Mass Geometry World” •  These other worlds have no mass

•  Do not affect the fundamental physics processes •  They are still seen by all navigators

•  Boundaries seen in this parallel geometry can be used to control: •  shower parameterizations •  geometrical event biasing •  scoring

•  Only “standard world” can contain mass •  Parallel worlds could not simplify the most difficult

Geometry overlay issue in brachytherapy: •  that seeds and applicators must overlay a CT based patient geometry

Parallel Mass Geometry (PMG) and Brachytherapy •  Regular world used for CT based patient

reconstruction •  Parallel world used for any other device :

•  Seeds •  Applicators (with GDML)

•  Avoid any form of overlapping •  Could also be extended to detector simulation in in-

vivo dosimetry

•  Both worlds have mass: •  The standard world (CT based patient) •  parallel world (radiation sources such as seeds

and applicators) •  For any area where mass is present in the parallel world,

this parallel mass is used. •  Elsewhere, the mass of the standard world is used

Memory efficient and fast voxel navigators and prametrizations can now be used

Geant4 Applications in medical physics:

PTSsim •  Software framework for particle therapy facility •  Class library for implementing a geometry

model of hadrontherapy facilities are provided •  Beam lines at 3 facilities in Japan are

implemented •  consists of modules and data files •  The user can configure the irradiation system

and the geometrical parameters of beam modules by using user UI

•  The modules are categorized to a beam •  module, a scoring module, and a DICOM

module •  The configuration parameters of beam modules

are described in ASCII data files •  allow the configuration to be changed easily

without recompiling.

Model nozzle Import patient CT

Transport in an all particle code Score dose, fluence, etc.

Save/Replay Phase space Advanced graphics

Fully 4D: moving parts of nozzle, beam current modulation, time-varying fields for beam scanning, etc, patient motion

INRAD, February 2011 GAMOS 52

After almost ten years providing different utilities to make easier the use of Geant4 we created GAMOS with the objectives:

to provide a framework not only easy-to-use but also flexible, that may cover the needs of a wide range of Geant4 users

(a framework should not become an obstacle for an advanced user who wants to exploit all the functionality that Geant4 offers)

No C++ : comprehensive scripting language   Very wide range of input possibilities   Allow extraction of detailed information at any moment of the simulation

  Simulation is not a black box that outputs some results

Easily extendible with new user code   Provide a mechanism so that users can easily extend the framework

functionalities with their code -  Without the need to understand how GAMOS works internally

Optimised in CPU   Otimisation options of general purpose and also specific for each

application

GAMOS objectives

INRAD, February 2011 GAMOS 53

  Easy way to build accelerator geometries   Simple text format with special modules (MLC, range shifter, ...)

  Reading DICOM patient geometries

  Transform DICOM files to binary or text format   Insert objects in voxelised geometries

  Many primary generator distributions   Position, direction, energy and time

  Writing/reading phase space files   One or several planes in the same job

  Recycle particles that reach a plane

  Save extra info (regions particle traversed, regions particle created , regions particle interacted, particle origin postion)

Radiotherapy

INRAD, February 2011 GAMOS 54

  Dose deposition in phantoms   Fast Geant4 regular navigation is default   Dose with errors can be calculated   Different printing formats in each run

  Text of binary output   Dose histograms (X, Y, Z, XY, XZ, YZ, dose, dose-volume)

  CPU Optimisation   Automatic optimisation of cuts and user limits   Bremsstrahlung spitting   Flexible importance sampling

  Analysis and debugging tools

Radiotherapy Dose from brachytherapy seeds

Proton therapy treatment room

The Geant4-DNA project  55

  Initiated in 2001 by Dr Petteri Nieminen at the European Space Agency/ESTEC

  Main objective: to adapt the general purpose Geant4 Monte Carlo toolkit for the simulation of interactions of radiation with biological systems at the cellular and DNA level

  in order to predict early DNA damages (up to 1 microsecond after irradiation)

  in the context of future human space exploration missions, radiobiology & radiotherapy

  providing an open source access to the scientific community

  Phase 1 started in 2001

  Delivered work package reports and a user requirement document

  Phase 2 ongoing since 2004

  First physics models were added to Geant4 in late 2007

  Currently an on-going interdiciplinary activity of the Geant4 low energy electromagnetic Physics working group

  Coordinated by CNRS/IN2P3 since 2008

See Int. J. Model. Simul. Sci. Comput. 1 (2010) 157–178

How can Geant4-DNA model radiation biology ?

 56

Physics stage step-by-step modelling of

physical interactions of incoming & secondary ionising radiation with biological medium (liquid

water)

Physico-chemistry/chemistry stage •  Radical species production •  Diffusion •  Mutual interactions

Geometry stage DNA strands, chromatin fibres, chromosomes, whole cell nucleus, cells…

for the prediction of damages resulting from direct and indirect hits

•  Excited water molecules •  Ionised water molecules •  Solvated electrons

Biology stage DIRECT DNA damages

Biology stage INDIRECT DNA damages (dominant @ low LET)

t=0 t=10-15s t=10-6s

Physics models available in Geant4 9.4

 57

  Geant4-DNA physics models are applicable to liquid water, the main component of biological matter   Extension to DNA materials is in progress (A, T, G, C, sugar-phosphate)

  They can reach the very low energy domain (sub-eV limit) down to electron thermalization   Compatible with molecular description of interactions

  Sub-excitation electrons (below ~9 eV ) can undergo vibrational excitation, attachement and elastic scattering

  Purely discrete   Simulate all elementary interactions on an event-by-event basis

  No condensed history approximation

  Models can be purely analytical and/or use interpolated data tables   eg. computation of integral cross sections

  Since December 2009, they use the same software design as all electromagnetic models available in Geant4 (standard and low energy EM models)   Allows the combination of processes (see later)

Conclusion:

Geant4 is constantly evolving to meet medical physics needs

Acknowledgment

Joseph Perl Sebastien Incerti Pedro Arce Michel D’Amours