The main goal of radiation therapy is to deliver a lethal dose of radiation to the targeted tumor while minimizing the radiation dose to the surrounding normal tissues and critical organs. Modern cancer therapy has benefited enormously from computer controlled treatment devices with increased precision and capability. However, this increased sophistication also creates new challenges for treatment planning. As the number of parameters in a treatment plan increases, the traditional computational approaches are no longer adequate to fully exploit the potential of the latest treatment devices. This is because at the heart of treatment planning is often a set of substantially non-trivial constrained geometric optimization problems. In this dissertation, we present a new optimization framework combining Particle Swarm Optimization (PSO) with deterministic optimization (e.g., least distance programming, non-negative least-square optimization, etc.). For our new PSO framework, we moved away from the classical view of a particle representing a potential solution of the optimization function; instead, we use the whole particle distribution as the solution. We modeled tumors, critical organs and other tissues as geometric volumes, whose surfaces have an associated potential function. The radiation source is modeled as kinetic particles subject to the forces from the potential functions from both the particles and the various geometric objects. The final configuration of the swarm of particles including their trajectories is the treatment plan. To demonstrate the potential of our new optimization paradigm, we have applied it to Gamma Knife radiosurgery and High-Dose Rate Brachytherapy (HDR) for prostate cancer. Mathematically, Gamma Knife radiosurgery is a ball-packing process whose goal is to 'pack' some spherical high dose volumes into a tumor volume, while brachytherapy is to find the trajectories of some spherical high dose volumes. Both problems are computationally intractable. Our new framework models the spherical high dose volume as kinetic particles and simulates the 'swarm' of these particles through a potential field created based on medical constraints and prescriptions. The resulting stable swarm, further refined by a deterministic optimization algorithm (e.g., non-negative least squares and least distance programming), is the final treatment plan. In the medical field, the adoption of new technologies take several years (sometimes even decades) due to the long trials they undergo, their high complexity, the lack of specific additional characteristics physicians demand or their high cost of infrastructure. As a consequence, Gamma Knife radiosurgery and HDR brachytherapy treatment planning are mainly performed as manual processes by physicians, despite the existence of algorithms that attempt to make them fully or partially automatic. Our experiments with challenging simulated and real clinical data have shown that our new framework significantly outperforms current clinical systems. In particular for Gamma Knife radiosurgery, our algorithm is able to produce high quality treatment plans that meet clinical standards. For HDR brachytherapy planning, our method can generate optimal (i.e., minimal and error tolerant) implant trajectories, which is a feature that no known algorithm has attempted to solve. Finally, we expect that due to the evidence shown in this dissertation, the simplicity of implementing our framework, and the ease of understanding the concepts of our approach, we will be able to widely impact the technologies currently being used not only in Gamma Knife radiosurgery and HDR brachytherapy, but also in other radiation therapy modalities.
Optimization, Radiation therapy, Gamma Knife radiosurgery, HDR brachytherapy
Level of Degree
Department of Computer Science
First Committee Member (Chair)
Second Committee Member
Third Committee Member
Riofrio, Daniel. "New Optimization Technique for Radiation Therapy Planning." (2015). http://digitalrepository.unm.edu/cs_etds/51