Candidate: Kourosh Khedriliraviasl
Title: Geometry-based Constraint Generation for Large-scale Radiation Therapy Treatment Planning
Date: November 26, 2019
Time: 10:30 am
Place: CPH 3623
Supervisor(s): Mahmoudzadeh, Houra
Abstract:
Intensity-Modulated Radiation Therapy (IMRT) is a high precision radiotherapy with many beams that have different intensities to accurately irradiate a tumor/cancerous cells considering minimization of the dose to surrounding normal tissues.
Different optimization methods are developed to achieve optimal beam intensities considering different criteria in objective function and number of constraints. IMRT optimization models are large-scale in nature because large number of constraints and/or variables associated with tumor/cancerous cells can be defined which is very hard to solve and takes a huge amount of computational time. A novel constraint generation solution method for IMRT treatment planning is presented in this research. A Fluence Map Optimization (FMO) framework is used to formulate the optimization problem for a breast cancer treatment planning case study. In addition, a data processing method is developed to form clusters of similar voxels with the highest dose associated from beamlets. Because in this optimization problem there are large number of constraints related to each voxel, these data clusters will help in deciding which voxels to pick as set of new constraints for the sub-problems. This is for the purpose of reducing the number of constraints. The optimal beam intensities do not cause a violation of any other constraints in the main model. The results show that the objective function value from constraint generation model is same as the objective function value that is resulted from the full-size model implementation considering all the constraints. Also, the optimal beam intensities from these two models are comparable and they create similar fluence map pattern. This shows that the new model is able to maintain the same quality as large-scale model. The advantages of this new method are, reducing the number of constraints that should be considered for the optimization significantly, and maintaining solution quality which can help create faster optimization algorithms.