Robust Optimization in IMRT Robust Optimization in IMRT
Operations Research Center, Massachusetts Institute of Technology
∗ Electronic address: nohadani@mit.edu
Intensity Modulated Radiation Therapy (IMRT) is a key component in cancer treatment today. In this form of treatment, ionizing radiation are directed onto cancer cells from different directions, with the ob jective of destroying them. Unfortunately, healthy and non-cancerous cells are exposed to the destructive radiation as well, since cancerous tumors are often embedded within the patient’s body. Even though healthy cells can repair themselves, it is important to keep such unnecessary exposure to a minimum.
By varying the beam angles and modulations using multiple layers of collimators, a radiation oncologist can control the radiation dosage which is deposited into the respective organs. For the treatment planing, an optimization problem is solved with the ob jective of minimizing the total radiation received by the patient, while ensuring that the tumor is sub jected to the required level of radiation. Unfortunately, deviations from the desired treatment plan are common, due to unavoidable implementation errors in beam angles and modulations. In a simulation study using 10000 random errors as they realistically occur, we observe that an otherwise ”optimized” treatment plan in fact delivers an insufficient dosage to the tumor, 100% of the time.
In collaboration with Massachusetts General Hospital, we report the application of a novel robust optimization to the nonconvex IMRT planning problem in order to find treatment plans that perform better under implementation errors. Using the robust local search technique, we find a large number of treatment plans that are robust against errors of different magnitudes. By considering the pareto frontier of these designs, a medical treatment planer can assess the trade-off between the amount of undesirable radiation and the probability of delivering an insufficient dosage to the tumor.
Aside the practical application that enables treatment planers to find therapy plans that are resistant to errors in beam angles and intensity, our research showcases the relevance of robust optimization to complex health-care problems.
