TPS Algorithms Flashcards
Different Types of Algorithms
• Segmentation
• Advanced margin
• Image Registration
• Inverse Plan Optimisation (specifically for IMRT or VMAT)
• Dose Calculation
Types of TPS and their corresponding Linac
• Monaco = Elekta TPS
• Eclipse = Varian TPS
• Pinnacle3 = Phillips
Values required for TPS Commissioning
- Output Factors
- Depth Dose Data
- A large proportion of data is measured on the central axis
Two Components of Dose
- Primary radiation (Dprim)
- Scattered radiation (Dscat)
Does Penumbra alter the accuracy of the dose calculation algorithm?
Not all algorithms can calculate dose accurately in the penumbra region
Different Classes of Algorithms
Factor-Based Algorithms
Model-Based Algorithms
Factor Based Algorithms
- Based on measured data
- Example: Clarkson’s Technique
Model Based Algorithm (Parts and Examples)
Consists of two parts:
- a part of the algorithm that models the beam, and provides a representation of the fluence distribution before the beam enters the patient
- a part that models the patient, usually based on a tomographic representation of the patient tissues
- Examples: Monte-Carlo, CCC, AAA, PB
What is the intent of algorithms?
To predict with as much accuracy as possible the dose delivered to any point in the patient
Speed and accuracy of most common algorithms
- PB (Least accurate, Fastest)
- Convolution (Intermediate)
- Monte Carlo (Most accurate, Slowest)
Examples of Photon Algorithms
PB, Convolution, Superposition, Monte Carlo, AcurosXB
What is a Convolution?
- Product of two functions to create a third
- Dose at any point can be calculated from the convolution of TERMA and kernel
- Dose is a product of TERMA and Kernel
Main Components of a Convolution
- energy imparted to the medium by the interactions of primary
photons, called TERMA - The energy deposited about a primary photon interaction site, the kernel.
Main Parts of a Kernel
- The primary kernel calculates the primary dose
- The scatter kernel calculates the first and multiple scatter doses.
What is a Scatter Kernel?
The summation of effects from scattering elements to calculate dose at a desired point
Scatter Kernel Shape
Tear-Drop Shape
What are some Variants of the Convolution Method?
- Super position or Convolution Superposition Algorithms
- Used by the TPS ‘CMS XiO’
- Variant of the Superposition-Convolution Algorithm = CCC
PB Algorithm
- Still clinically used for IMRT/VMAT optimisation (not dose calculation)
- Used in the first stage of IMRT/VMAT optimisation (finds initial set of control points)
- Faster speed of PB is preferable to CCC or MC
Pencil Beam Convolution Algorithms
- anisotropic analytical algorithm (AAA) used by Eclipse TPS is based on PB-Convolution technique
- AAA uses spatially variant convolution scatter kernels, from Monte Carlo simulation, to separate modelling for primary photons,
scattered photons, and contaminant electrons. - AAA is attractive option for routine clinical use because of its relatively short computation time and accuracy comparted to Monte Carlo
Electron Algorithms
- Dose calculations are more complicated than photons
- Side scattering of the electrons is more pronounced and has more influence on the dose distribution
Algorithm Examples
- Clarkson
Clarkson Algorithm
- Used on the CMS-XiO TPS
- Measurement based method
- Not accurate dose calculation
Pencil Beam model for Electrons
- Pinnacle3 uses PB for electron
- Has problems in regions of inhomogeniety
-Tends to underestimate the effects of sharp discontinuities in density
Monte-Carlo Dose Calculation Simulation Process
⇒ Start with an electron exiting from waveguide.
⇒ Follow it and its descendants through targets,
primary collimators, ion chambers etc.
⇒ Track it through patient-dependant structures (jaws,
MLC etc.).
⇒ Track it through the patient (as modelled from CT
data set).
Monte Carlo Simulation
- Is a stochastic integration method
- Monte Carlo calculated quantities are subject to statistical uncertainties
- One must simulate an infinite number of histories for a zero uncertainty
What are histories in the Monte-Carlo Simulation
-the number of particles generated by the source model
Statistical Uncertainty
Statistical Uncertainty (SU) is proportional to 1/√N
What happens to the calculation time if you reduce the statistical uncertainty
Reducing the SU, calc time is increased
Dose Distribution from Monte Carlo Simulation
⇒ Any MC-calculated dose distribution will be a noisy representation of the true dose distribution
Dose Distribution from Monte Carlo Simulation
⇒ Any MC-calculated dose distribution will be a noisy representation of the true dose distribution
What does the SU in MC Computed Dose affect in the treatment planning process?
- Isodose representations- appearance (noise)
- DVH accuracy diminishes
- Maximum and minimum dose in a volume affected
- Dose metrics such as TCP, NTCP, EUD are affected
- Cost functions used for treatment plan optimization are
also affected.
What would be the effect of increasing statistical uncertainty (SU) beyond 10% per calculation?
Further from 1, the isodose lines will display with more noise (increasing will change the visualisation of isodose line) (preferred is <2%)
Benefits of PB
Fast
Accounts for scatter
Limitations of PB
Doesn’t account for increased lateral scatter which occurs when the beam interacts with air.
Poor representation of inhomogeneity effects
-Assumes monoenergetic beams
-Struggle to deal with irregular contours
Benefits of CONV/SUP-CONV
+ Accounts for inhomogeneities
+ Accounts for spectrum of energies
+ Accounts for beam modifiers
Limitations of CONV/SUP-CONV
-Relatively Slow
-Difficult to commission
Benefits of MC
+ Extremely accurate
+ Inherently accounts for patient and beam parameters
Limitations of MC
- Demanding of computing power
- Slow compared to other Algorithms
- Difficult to commission
Acuros XB algorithm
- advanced dose calc algorithm on Eclipse TPS
Simulates an infinite number of particles, and systematic errors are introduced by discretisation in space, angle and energy
What is SVD algorithm
The singular value decomposition
- used in pinnacle and raystation
Separates the lateral and depth direction of a pencil beam distribution in water therby saves memory and increases calc speed
Model based algorithms use clinically
Eclipse - AAA, Acuros, EMC
Monaco - EMC, CC, PB
Pinnacle - CCC, SVD, PB
Raystation - CC, VMC, SVD
Oncentra MP - VMC, CCC
XiO - Superposition, clarkson, FFF
