Academic Courses > DOS 523 > Planning Algorithms
DOS 523 - Treatment Planning for Medical Dosimetrists
Writing Prompt
There are a couple of algorithm sheets in the content section of the course. After looking them over, what computer algorithms does your treatment planning computer use? Discuss with the physicist and describe to your classmates which algorithms are used and why. What are the advantages and disadvantages your algorithm has? No references needed for this post.
Initial Post: Planning Algorithms
At my clinical site, we have just finished an upgrade to Varian Eclipse version 11.0.47.555. We use this system to perform the vast majority of our clinical planning, with the exception of stereotactic radiosurgery (SRS) and stereotactic radiotherapy (SRT), which are planned on BrainLab iPlan. Stereotactic Body Radiotherapy (SBRT) is planned on Eclipse. These systems have several dose calculation algorithms available for dose calculation. For photon planning, we have Acuros XB and the Anisotropic Analytical Algorithm (AAA) available in our systems. We use Acuros for IMRT and VMAT planning, and AAA for 2D and conformal planning. For electron planning, we have a choice between the Generalized Gaussian Pencil Beam (GGPB) algorithm and eMC, which is a Monte Carlo algorithm. For photon planning with iPlan, we use a Monte Carlo calculation algorithm.
AAA is a convolution algorithm that models the dose contributions from the primary beam, photon scatter from the treatment head, and electron scatter from the treatment head and air separately before combining them together for the final dose map. Dose kernels that have been modeled during commissioning are scaled in width and depth based on local tissue density to attempt to model lateral scatter contributions. This lateral scattering is important, especially when nearby tissue is a significantly different density than the point currently being measured.
Acuros XB models the actual tissues being traversed, which should in theory provide a better mapping of true dose distributions, especially in areas of inhomogeneity. Since CT scanners usually top out theor Hounsfield Unit (HU) scales around 3071 HU, materials that are high density are indistinguishable from each other, because they all read out as the same maximum value. Acuros XB allows users to explicitly state the material in each high density region to improve the accuracy of the calculation.
We do not typically use GGPB because pencil beam algorithms do not appropriately account for lateral scatter, and this can be very important, especially if we are trying to match field edges. Instead, we use the eMC algorithm, which is a Monte Carlo method. Monte Carlo algorithms are a class of computation methods that rely on modeling single particles (photons, electrons, protons, etc), and creating a simulated path for that particle from emission to absorption, including all matter interactions along the way, based on a probability model. The story of a single particle is a random set of events that can't tell the user much about what is going on, but when thousands of these simulated events are calculated, trends begin to emerge. The is computationally intense, but the accuracy is much better.
iPlan uses a Monte Carlo calculation model for its photon planning. We rely on this model's accuracy for our stereotactic planning, where sub-millimeter accuracy is important for the patient's plan outcome.
Grayden, Chicago
AAA is a convolution algorithm that models the dose contributions from the primary beam, photon scatter from the treatment head, and electron scatter from the treatment head and air separately before combining them together for the final dose map. Dose kernels that have been modeled during commissioning are scaled in width and depth based on local tissue density to attempt to model lateral scatter contributions. This lateral scattering is important, especially when nearby tissue is a significantly different density than the point currently being measured.
Acuros XB models the actual tissues being traversed, which should in theory provide a better mapping of true dose distributions, especially in areas of inhomogeneity. Since CT scanners usually top out theor Hounsfield Unit (HU) scales around 3071 HU, materials that are high density are indistinguishable from each other, because they all read out as the same maximum value. Acuros XB allows users to explicitly state the material in each high density region to improve the accuracy of the calculation.
We do not typically use GGPB because pencil beam algorithms do not appropriately account for lateral scatter, and this can be very important, especially if we are trying to match field edges. Instead, we use the eMC algorithm, which is a Monte Carlo method. Monte Carlo algorithms are a class of computation methods that rely on modeling single particles (photons, electrons, protons, etc), and creating a simulated path for that particle from emission to absorption, including all matter interactions along the way, based on a probability model. The story of a single particle is a random set of events that can't tell the user much about what is going on, but when thousands of these simulated events are calculated, trends begin to emerge. The is computationally intense, but the accuracy is much better.
iPlan uses a Monte Carlo calculation model for its photon planning. We rely on this model's accuracy for our stereotactic planning, where sub-millimeter accuracy is important for the patient's plan outcome.
Grayden, Chicago
Academic Courses > DOS 523 > Planning Algorithms
|
Written April 5, 2015
Second Semester, 3 Months into Internship |