Tissue optics Monte Carlo
Project description
Light propagation with Monte Carlo in Python
This is an extremely simple objectoriented code in Python that simulates the propagation of light in scattering tissue. It is not just simple: it is outrageously simple and very slow (see below). However, it is extremely easy to understand and most importantly very simple modify. It is a Python implementation of Monte Carlo, as beautifully illustrated by the standard, tested, and loved MCML from Wang, Jacques and Zheng , itself based on Prahl and completely documented, explained, dissected by Jacques and Prahl. Everyone is doing Monte Carlo in tissue, and nothing would be possible without the work of these pionneers.
It may be slow, but speed is more than code performance: anyone with little to no experience can simulate something instantly instead of having to understand C, C++ or, god forbid, GPU code. Therefore, you can quickly modifiy everything in an afternoon and get your results in a few hours, instead of learning C (a few weeks?), learn to work with compiled code (a few days? libraries anyone?) and finally modify the C code written by someone else (days? weeks?). I think the overall speed to be concerned about is "the time it takes to get an answer", not necessarily "the time it takes to run 100,000 photons". Considering many calculations with high performance code (in C for instance) take a few minutes, it is fairly reasonable to imagine you could start a calculation in Python, run it overnight and get an answer the next day after a few hours of calculations. I think there is a need for such a solution, and you will find it here.
Therefore, the whole point is the following: this code is perfect for quickly prototyping a small calculation, and then determine if you need performance or not. For many things, you actually don't.
Getting started
Install with pip or get the code from GitHub. You can run the example code immediately:
pip install pytissueoptics upgrade python m pytissueoptics
You need Python 3, it will not work with Python 2. The example code will show you a graph of the energy deposited in the plane xz from a isotropic source at the origin:
Then it will display the logarithm (log10
) of the intensity as a fonction of distance along the x direction:
We can also display the intensity on surfaces:
Then, the idea would be to modify the code for your geometry (layers, boxes, cubes, spheres, etc...) and compute what you want.
What it can do
There are 6 main concepts (or Class
in objectoriented language) in this code:
Photon
: The photon is the main actor: it has a position, it propagates in a given direction. Its direction is changed when it scatters. It does not know anything about geometry or physical properties of tissue.Source
: A group of photons, such asIsotropicSource
,PencilSource
with specific properties. You provide the characteristics you want and it will give you a list of photons that responds to these criteria. This list of photons will give you the answer you want after it has propagated in theObject
of interest.Material
: The scattering properties and the methods to calculate the scattering angles are the responsibility of theMaterial
class. TheMaterial
also knows how to move the photon between two points (for instance, if there is birefringence, this is where you would put it although polarization is currently not implemented).Geometry
: A realworld geometry (Box
,Cube
,Sphere
,Layer
, etc...). The geometry has two important variables: aMaterial
(which will dictate its optical properties i.e. scattering and index) and aStats
object to keep track of physical values of interest. The material will provide the required functions to compute the scattering angles, and the photon will use these angles to compute its next position. TheStats
object will compute the relevant statistics.Stats
: An object to keep track of something you want. For now, it keeps track of volumetric quantities (i.e. the energy deposited in the tissue) and intensities through the surfaces delimiting geometries. Finally, very useful
Vector
,UnitVector
andSurface
helper classes with their subclasses are used to simplify any 3D computation with vectors, planes, surfaces, because they can be used like other values (they can be added, subtracted, normalized, etc...).
Limitations
There are many limitations, as this is mostly a teaching tool, but I use it for real calculations in my research:
 It only uses HenyeyGreenstein because it is sufficient most of the time.
 Documentation is sparse at best.
 You have probably noticed that the axes on the graphs are currently not labelled. Don't tell my students.
 Did I say it was slow? It is approximately 50x slower than the wellknown code MCML on the same machine. I know, and now I know that you know, but see Advantages below.
Advantages
However, there are advantages:
 It is extremely simple to understand.
 The code is very clear with only a few files in a single directory.
 It can be used for teaching tissue optics.
 It can be used for teaching objectoriented programming for those not familiar with it.
 It is fairly easy to modify for your own purpose. Many modifications do not even require to subclass.
 In addition, because it is very easy to parallelize a Monte Carlo calculations (all runs are independant), splitting the job onto several CPUs is a good option to gain a factor of close to 10 in perfromance on many computers. I have not done it yet.
The core of the code
The code is in fact so simple, here is the complete code that can create graphs similar to the ones above in 10 seconds on my computer:
from vector import * from material import * from photon import * from geometry import * # We choose a material with scattering properties mat = Material(mu_s=30, mu_a = 0.1, g = 0.8, index = 1.4) # We want stats: we must determine over what volume we want the energy stats = Stats(min = (2, 2, 2), max = (2, 2, 2), size = (41,41,41)) # We pick a geometry tissue = Layer(thickness=2, material=mat, stats=stats) #tissue = Box(size=(2,2,2), material=mat, stats=stats) #tissue = Sphere(radius=2, material=mat, stats=stats) # We pick a light source # The source needs to be inside the geometry (for now) source = PencilSource(position=Vector(0,0,0.001), direction=Vector(0,0,1), maxCount=1000) # We propagate the photons from the source inside the geometry tissue.propagateMany(source, graphs=True) # Report the results tissue.report()
The main function where the physics is hidden is Geometry.propagate()
. Geometry.propagateMany()
is a helper to call the function several times, and could possibly be parallelized:
class Geometry: [...] def propagate(self, photon): photon.transformToLocalCoordinates(self.origin) d = 0 while photon.isAlive and self.contains(photon.r): # Pick distance to scattering point if d <= 0: d = self.material.getScatteringDistance(photon) isIntersecting, distToPropagate, surface = self.intersection(photon.r, photon.ez, d) if not isIntersecting: # If the scattering point is still inside, we simply move # Default is simply photon.moveBy(d) but other things # would be here. Create a new material for other behaviour self.material.move(photon, d=distToPropagate) # Interact with volume: default is absorption only # Default is simply absorb energy. Create a Material # for other behaviour delta = self.material.interactWith(photon) self.scoreInVolume(photon, delta) # Scatter within volume theta, phi = self.material.getScatteringAngles(photon) photon.scatterBy(theta, phi) else: # If the photon crosses an interface, we move to the surface self.material.move(photon, d=distToPropagate) # Determine if reflected or not with Fresnel coefficients if self.isReflected(photon, surface): # reflect photon and keep propagating self.reflect(photon, surface) else: # transmit, score, and leave self.transmit(photon, surface) self.scoreWhenCrossing(photon, surface) break d = distToPropagate # And go again photon.roulette() # Because the code will not typically calculate millions of photons, it is # inexpensive to keep all the propagated photons. This allows users # to go through the list after the fact for a calculation of their choice self.scoreFinal(photon) photon.transformFromLocalCoordinates(self.origin) def propagateMany(self, source, showProgressEvery=100): startTime = time.time() N = source.maxCount for i, photon in enumerate(source): self.propagate(photon) self.showProgress(i, maxCount=N , steps=showProgressEvery) elapsed = time.time()  startTime print('{0:.1f} s for {2} photons, {1:.1f} ms per photon'.format(elapsed, elapsed/N*1000, N))
Note that this function is part of the Geometry
object and does not make any assumption on the details of the geometry, and relies on whatever material was provided to get the scattering angles.
How to go about modifying for your own purpose

Maybe you have a special light source?

Subclass
Source
with your own light source and compute the photon properties innewPhoton
according to your own rules. Use your source instead ofIsotropicSource
in the example above:class MySource(Source): def __init__(self, myProperty, position, maxCount): super(MySource, self).__init__(position, maxCount) self.myProperty = myProperty def newPhoton(self) > Photon: p = Photon() # Do your thing here with self.myProperty and modify p return p


Maybe you have a special material scattering model?

Subclass
Material
and override the methods forgetScatteringAngles()
. Use your material in your geometry instead ofMaterial
in the example above. You could use the photon direction, polarization, position, or even its wavelength to compute its scattering angles:class FunkyMaterial(Material): def __init__(self, myProperty, mu_s = 0, mu_a = 0, g = 0): super(MySource, self).__init__(mu_s, mu_a, g) self.myProperty = myProperty def getScatteringAngles(self, photon) > (float, float): # Do your thing here with self.myProperty and compute theta, phi # Use Photon if needed (position, direction, wavelength, etc..) return (theta, phi)


If
photon.keepPathHistory()
is called, it will keep track all positions during its lifetime. You can then compute whatever you want by rewriting that part of the code or with a hook function I will write at some point. 
Maybe you want to compute some funky stats? At each step,
scoreInVolume
is called with the photon and the drop in energy at that step. When leaving the geometry through a surface,scoreWhenCrossing
is called with the photon and the last position inside. 
Maybe your have a special geometry? Subclass
Geometry
and override thecontains
method to compute whether or not a given position is inside your object or not, andintersection
to compute the point on the surface of your object where the photon exits.
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Filename, size  File type  Python version  Upload date  Hashes 

Filename, size pytissueoptics1.0.2py3noneany.whl (19.3 kB)  File type Wheel  Python version py3  Upload date  Hashes View 
Filename, size pytissueoptics1.0.2.tar.gz (20.3 kB)  File type Source  Python version None  Upload date  Hashes View 
Hashes for pytissueoptics1.0.2py3noneany.whl
Algorithm  Hash digest  

SHA256  2fb67f61a70195bd66e7781b24641cfc8c68e1535caaf30092863437aa8bd026 

MD5  488ac5f31f65794e536c6266d5664529 

BLAKE2256  d70658bcf2b86af465136524533627ea854c266fce6cc54d8a55d8002d4ebd6a 