A Monte Carlo model of light propagation in tissue An introduction to Monte Carlo techniques ENGS168 Ashley Laughney November 13 th , 2009
Feb 23, 2016
A Monte Carlo model of light propagation in tissue
An introduction to Monte Carlo techniquesENGS168
Ashley LaughneyNovember 13th, 2009
Overview of Lecture• Introduction to the Monte Carlo Technique
– Stochastic modeling– Applications (with a focus on Radiation Transport)– Random sampling and Probability Distribution Functions
• Monte Carlo Treatment of the Radiation Transport Problem– Photon initialization– Generating the propagation distance– Internal reflection– Photon absorption– Photon termination– Photon scattering– Calculation of observable quantities
• Implementing Monte Carlo - Sample Code• Radiation treatment planning
Stochastic Modeling• Launch a photon (or particle)• Sample physical properties using random
variables (random sampling)• Let the photon evolve through the system• Keep track of all important parameters until the
photon dies or exits the problem space• Summarize and infer results after ENOUGH
photons interact.
Ref: Venkat Krishnaswamy
Applications (a few)• Modeling photon transport in tissue • Calculating dose distributions for radiation
therapy• Solving the problem of neutron diffusion in a
fissionable material• Calculating financial derivatives and evaluating
investment value and market behavior• Wireless network design• Numerical integration
Ref: Wikipedia
The radiation transport problem
Symbol Description UnitsN(r , sˆ) Number density of photons at a
point r, moving along sm-3Sr-1
Lυ Spectral Radiance Energy flow per unit normal area per unit solid angle per unit time per unit temporal frequency bandwidth
L Radiance, quantity used to describe propagation of photon power
Spectral radiance integrated over a narrow frequency range [υ ,υ + Δυ ]
Φ Fluence rate, indicates net radiant energy
Energy flow per unit area per unit time
Radiative Transfer Equation (RTE):
Diffusion Approximation:with diffusion coefficient, Ref: Ambrocio, Master Thesis,
http://appliedmath.ucmerced.edu/theses/ambrocio_2009.pdf
source
scattering
extinction
divergence
Tissue optical properties
Probability density functions (PDF)
• A probability distribution describes the range of possible values that a random variable can attain and the probability that the value of the random variable is within any subset of that range. – i.e., what is the chance of
getting a value for every possible outcome of the random variable
– Coin toss• A PDF is the functional
form of a probability distribution
Ref: Venkat Krishnaswamy
Beer’s Law - a probability distribution
Ref: Venkat Krishnaswamy
• The fraction of photons that will survive after a distance d’<d can be seen as a probability distribution over all d’.• How far will a photon travel in an absorbing medium without an interaction?
- 100% chance it will travel 0cm~40% chance it will travel 1cm
Cumulative distribution function (CDF)• The probability that a measurement yielding a
value of x will lie in the interval [0,x1] is given by the cumulative distribution function.
• Where, , represents the probability density distribution of a set size, x є [0, Inf] , that a photon takes between any two scattering events. Beer’s law as a cumulative distribution
* CDFs and Random Sampling*
• The CDF is always uniformly distributed on the interval [0,1].
• Sampling by inversion of the CDF1) Sample a random
number ξ from U[0,1]2) Equate ξ with the CDF,
F(x) = ξ3) Invert the CDF and
solve for xRef:
http://www.phy.ornl.gov/csep/CSEP/GIFFIGS/MCF12.GIF
Cumulative distribution functions associated with physical processes can be sampled using random numbers via direct inversion.
Ref: Jacques, Prahl, http://omlc.ogi.edu/software/mc/
• Computers can generate random numbers, , with a uniform PDF
• The associated CDF is given by
Pseudo random numbers
Flowchart for variable stepsize MC
Ref: Prahl, A Monte Carlo Model of Light Propagation in Tissue, SPIE (1989)
• Implicit Capture ~ each photon launched into the medium is thought to represent a photon packet, where each packet enters the medium carrying the photon weight (i.e. 1J)
1. Photon initialization• N photons are launched, each with a "photon
weight" initially set to 1 (computationally efficient)• Start coordinates for each photon are identical• Photon’s initial direction chosen via convolution
with the beam shape
Ref: Prahl, A Monte Carlo Model of Light Propagation in Tissue, SPIE (1989)Image: http://support.svi.nl/wikiimg/ft_1.png
Example of Convolution
2. Propagation distance• A fixed stepsize, Δs, must be small relative to the
average mean free path of a photon in tissue.
• It is more efficient to choose a different stepsize for each photon step; the PDF for the Δs follows Beer’s law.
• A function of a random variable, ξ : [0,1], that is distributed uniformly and yields a random variable with this distribution:
Ref: Prahl, A Monte Carlo Model of Light Propagation in Tissue, SPIE (1989)
3. Internal reflection• The probability that the photon is internally reflected is
determined by the Fresnel reflection coefficient
• A random number uniformly distributed between 0 and 1 is used to determine if the photon is reflected or transmitted. Internal reflection
• i.e. For a semi-infinite slab, the internally reflected position is updated by only changing the z-component of photon coordinates
Ref: Prahl, A Monte Carlo Model of Light Propagation in Tissue, SPIE (1989)
// angle of incidence on the boundary// angle of transmission given by Snell’s law
4. Photon absorption• After each propagation step, a fraction of the
photon packet is absorbed and the remainder is scattered.
• Or generate photon absorption (weight) according to randomly generated step size and Beer’s law
a = the single particle albedo (fraction scattered)// New weight assigned to surviving
photon packet
Ref: Prahl, A Monte Carlo Model of Light Propagation in Tissue, SPIE (1989)
5. Photon termination• Propagating a photon packet with minimal weight
yields little information. How is the packet terminated?– Roulette is used to terminate a photon when its weight
falls below some minimum– The roulette gives a photon of weight w, one chance in
m, of surviving with weight mw. Otherwise, w 0– Unbiased elimination, conservation of energy
others
mifmww
0/1
Ref: Prahl, A Monte Carlo Model of Light Propagation in Tissue, SPIE (1989)
6. Photon scattering• The PDF for the scattered cosine of the deflection
angle (cosθ) in tissue is characterized by the Henyey-Greenstein phase function.
• The azimuth angle is uniformly distributed between [0,2π], and may be generated by multiplying a random number ξ:[0,1] by 2π
• The photon is scattered at an angle (θ, )
* For isotropic scattering,
*Assumes phase function has no azimuth dependence
Ref: Prahl, A Monte Carlo Model of Light Propagation in Tissue, SPIE (1989)
7. Observable quantities• Analytic Solution to RTE
fluence rate resulting from photons launched at a single point corresponds to the Green’s function for the medium.
• Monte Carlo Solution to RTE defines grid over solution space and scores physical quantities (reflection, transmission, absorption = energy deposited) at each grid element as the program traces N photons.
Ref: Venkat Krishnaswamy
Abbreviated review
Implementing Monte Carlo: A steady state example (“mc321.c”)• The following slides walk through a steady-state Monte Carlo
simulation by Steve Jacques and Scott Prahl at the Oregon Medical Laser Center. All code discussed is available online: http://omlc.ogi.edu/software/mc/
• Problem Definition:– Photons are launched from an isotropic point source of unit power, P = 1W into
an infinite, homogeneous medium with no boundaries– The medium has the optical properties of absorption, scattering and anisotropy– N Photons are launched, each initialized with weight, W = 1.– Solution space is divided into an array of bins (position is defined by distance r
from source), 3 options – spherical shell, cylindrical shell, planar shell– Each bin accumulates photon weights deposited due to absorption by N photons
Array containing accumulated weight of absorbed photons
Concentration of Photons
Map of fluence
Implementing Monte Carlo: Definitions of variables and arrays
Implementing Monte Carlo: User input
Implementing Monte Carlo: Launching photons
Implementing Monte Carlo: Moving photons ~ HOP
Beer’s Law
Implementing Monte Carlo: Moving photons ~ DROP
Implementing Monte Carlo: Moving photons ~ SPIN
Implementing Monte Carlo: Moving photons ~ SPIN
Implementing Monte Carlo: Moving photons ~ CHECK ROULETTE
Implementing Monte Carlo: Output bin arrays as fluence rate
Implementing Monte Carlo: Example Output
Ref: Jacques, Prahl, http://omlc.ogi.edu/software/mc/
Ref: Venkat Krishnaswamy, https://www.llnl.gov/str/Moses.html
• Motivation:
– Mortality caused by (1) providing too little radiation to the tumor for cure, or (2) providing too much radiation to nearby healthy tissue.
• Need: Improved radiation therapy planning• Ionizing and non-ionizing radiation can be used
– Photon therapy accounts for 90% of all radiation treatment in US
– Photons, electrons, neutrons, heavy charged particles (protons)• Dose distribution is key parameter of interest in
treatment planning
Advance Application: Radiation Treatment Planning
Diagnosed with life-threatening forms of cancer annually
Receive radiation treatment
Are considered curable
Die anyway
Current dose estimation techniques• Generate 3D electron-density map of body using
stack of CTs– Model the body as a homogeneous bucket of
water• One way to estimate dose in tissue is to use a
water phantom– use ionization chambers/chamber arrays to
detect dose distribution– currently used in clinical treatment planning– complicated experiments– heterogeneities hard to model
Ref: Venkat Krishnaswamy
Monte Carlo-based treatment planning• Voxelize medium of interest using CT/MR patient
images– Compute solution space geometry– Assign material data to each voxel (from
atomic and nuclear-interaction databases)• Launch radiation particles one at a time and let
them evolve• Store accumulated dose per voxel for N radiation
particles• After following many particle histories, an
accurate estimation of dose is obtained.
Ref:Venkat Krishnaswamy
PEREGRINE3D Monte Carlo Treatment Planing
Ref:https://www.llnl.gov/str/Moses.html
PEREGRINE Defining the radiation source and patient
Ref: https://www.llnl.gov/str/Moses.html
•3D Transport mesh of patient generated from stack of CT images• Radiation Source- upper portion of accelerator does not vary between treatments, but the lower portion is modified by collimators, blocks and wedges to customize patient treatment.- PEREGRINE library accounts for modification in lower half of accelerator
PEREGRINE Calculating Dose
Ref: https://www.llnl.gov/str/Moses.html, Venkat
Five-field treatment for a lung tumor; 6 MV photon beam
Seven-field conformal boost to the prostate; 18MV photon beam
Predicted dose build up for treatment of a brain tumor
Suggested ReadingLiterature: ED Cashwell, CJ Everett, "A Practical Manual on the Monte Carlo Method for Random
Walk Problems,“ Pergammon Press, New York, 1959.BC Wilson, G Adams, A Monte Carlo model for the absorption and flux distributions of
light in tissue, Med. Phys. 10:824-830, 1983.L Wang, SL Jacques, L Zheng, MCML - Monte Carlo modeling of light transport in multi-
layered tissues, Computer Methods and Programs in Biomedicine 47:131-146, 1995.L Wang, SL Jacques, "Monte Carlo Modeling of Light Transport in Multi-layered Tissues in
Standard C,” 1992-1998. Download as 177-page manual in pdf format.
Source Code:http://mcnp-green.lanl.gov/index.html Radiation MC, state of the arthttp://omlc.ogi.edu/software/mc/ Photon MC, MCML and othershttp://omlc.ogi.edu/software/polarization Photon MCML (Vector)