AbstractsComputer Science


Intensity modulated radiation therapy (IMRT) is a relatively new treatment technique used to treat different kinds of cancer. Each beam delivering radiation to the patient is split into smaller beamlets whose intensities can be set individually. This increased resolution makes it possible to achieve better conformance with the target structure (tumour) while sparing critical structures. The inverse problem of IMRT concerns to how to assign intensities to the beamlets in order to achieve a satisfactory dose distribution in the patient. Based on an established model of IMRT and authentic patient data provided by Oslo University Hospital a realistic scenario for benchmarking algorithms is created. The main goal of this thesis is to extend and modify the relaxation method for linear inequalities in order to obtain a method with better performance when applied to inverse problems resulting from a linear model of IMRT. A preprocessing technique exploiting dose influence data is presented, and the performance of the relaxation method in combination with the preprocessing technique is compared to that of the original relaxation method and the algorithms provided by CPLEX. In addition a simplified version of the inverse problem resulting from a simplified model of radiation treatment is studied, and a graph bashed algorithm for solving such problems is presented. Finally, preliminary numerical results regarding the capabilities of CPLEX to take advantage of a solution to the inverse problem obtained using the relaxation method combined with the preprocessing technique when solving a more sophisticated linear programming approach to the inverse problem are presented.