Cabrera-Guerrero, Guillermo

Solving the Direct Aperture Optimisation Problem using Local Search Strategies

Leslie Pérez Cáceres, Ignacio Araya, Denisse Soto, and Guillermo Cabrera-Guerrero

Escuela de Ingeniería Informática, Pontificia Universidad Católica de Valparaíso, Valparaíso, Chile

In this work, we aim to solve the direct aperture optimisation problem (DAO) in radiation therapy for cancer treatment by means of two novel heuristic local search strategies. In the DAO problem, the goal is to find a set of deliverable apertures shapes and intensities so we can irradiate the tumor according to a medical prescription without producing any harm to the surrounding healthy tissues. Traditional inverse-planning approach starts by selecting a set of beam angles radiation will be delivered from (Beam Angle Optimisation problem). Then, the intensities for such angles are computed by optimising some “meaningful” objective function (Fluence Map Optimisation problem). Finally, the apertures shapes that will allow us to deliver the radiation computed during the previous step are generated (Multi-Leaf Collimator Sequencing Problem). Clearly, this sequential approach provokes that the number of apertures shapes and the time patients will be exposed to radiation depends, to a large extent, on both the beams selected when solving the BAO problem and the intensities computed when solving the FMO problem.

Unlike this sequential approach, in the DAO problem, constraints associated to the number of deliverable aperture shapes as well as physical constraints are taken into account during the fluence map optimisation process. Thus, we do not longer need any leaves sequencing procedure after solving the DAO problem. This integrated approach allows us to find an intensity map that not only optimise the radiation that is delivered to the patient but also minimises the number of deliverable apertures shapes.

To solve the DAO problem, we try two novel heuristic local search strategies on a prostate case and compare the obtained treatment plan to the one obtained using the traditional sequential approach. Results show that our algorithms are able to find treatment plans that are very competitive when considering the number of deliverable aperture shapes.