Decision Theory
M Phillips, M Kim, I Kalet, C Holdsworth, Juergen Meyer, Wade Smith
Given a range of tradeoffs, which one do you make? Our work in multiobjective optimizationleads naturally to the question of decision making. Our basic approach has been to model the decision making environment using Bayesian networks and influence diagrams since they provide appropriate calculus for the propagation of uncertainty that is the hallmark of clinical decision making. We have constructed models for radiation therapy of prostate cancer and head & neck cancer, and are exploring other areas as well, such as cervical cancer. In determining relative ranks of different outcomes, we have studied the application of utility theory and Markov models in calculating quality adjusted life expectancy. We see the development of these models as being applicable to many areas of radiation oncology including IMRT plan selection, the evaluation of new technology and the design of clinical trials.
Markov Decision Processes in the treatment of Brain Metastases
M Phillips, M Kim, J Rockhill
In her thesis work, Dr Minsun Kim applied Markov Decision Processes (MDP) to fractionation issues in radiation therapy. MDPs are the method of choice for optimizing decisions involving stochastic problems over time. This makes them a good method for the increasingly common problem of deciding on the best method for treating brain metastases. Treatments include surgery, whole brain RT and stereotactic radiosurgery. Historically, such patients were given a single treatment, but improved systemic therapies mean that patients are living long enough to need multiple treatments. We hope to provide a useful tool for clinicians who need to make these critical decisions.
Monte Carlo Damage Simulation (MCDS)
RD Stewart
The MCDS is a Monte Carlo model to generate nucleotide-level maps of the multitude of clustered DNA lesions, including double strand breaks (DSB), formed by ionizing charged particles with an atomic number less than 26 (i.e., 56Fe). The MCDS has the capability to simulate cluster induction for arbitrary mixtures of charged particles with the same or different kinetic energies. The effects of oxygen concentration on cluster induction can also be simulated (0 to 100% O2). Although not required for the simulation of damage induction, the MCDS reports information such as the charged particle stopping power in water, CSDA range, absorbed dose per unit fluence, frequency-mean specific energy, energy imparted per radiation event, and the lineal energy. For additional information, see http://faculty.washington.edu/trawets/mcds/
Monte Carlo Excision Repair (MCER) Model
RD Stewart
Base Excision Repair (BER) is the primary DNA repair mechanism for the removal of clustered DNA lesions other than the double strand break (DSB). In the MCER, the Monte Carlo Damage Simulation MCDS is used to simulate the induction of clustered DNA lesions. Then, key steps in the excision repair of the lesions forming a cluster are simulated. The MCER tabulates repair outcomes, such as the probability a cluster is correctly repair (no change in DNA base sequence), repaired with a mutation (at least one base substitution) or converted to a DSB. The MCER also tabulates information potentially related to the kinetic of cluster repair. For additional information, see http://faculty.washington.edu/trawets/mcer/
Multiobjective optimization
M Phillips, M Kim, C Holdsworth
Optimization algorithms are used in radiation oncology to calculate IMRT plans. We are exploring the class of algorithms known as “multiobjective optimization” since there are multiple, often-competing, objectives that we strive to achieve. These algorithms are much more time-consuming than the conventional algorithms used in commercial treatment planning programs. We have been studying the use of genetic optimization algorithms in order to exploit their ability to optimize any form of objective function. We are also investigating how the form of the objectives determines the type type of solutions obtained. Overall, it is our goal to provide the clinician with a range of treatment plans that embody the range of tradeoffs inherent in the case.