Optimization methods for spatiotemporally fractionated radiotherapy planning
Abstract
The goal in radiotherapy, as one of the main modalities for cancer treatment, is to deliver
sufficient radiation dose to the tumor region to eradicate the disease while sparing the
surrounding healthy tissues to the largest extent possible. To achieve this goal, radiotherapy
plans for individual cancer patients are designed to deliver the desired spatial dose distribution
to the patient. The radiotherapy plan will be then used on a daily basis to deliver a daily
fraction of the prescribed radiation dose over the course of treatment. However, there is
biological evidence suggesting that additional therapeutic gain may be achieved if we allow for
temporal variation in the radiotherapy plan. This treatment paradigm is known as
spatiotemporal fractionation, which is the topic of this dissertation. In this research, we first
develop mathematical models and solution methods to show the existence of the potential
therapeutic benefit of spatiotemporal plans over conventional plans for stylized cancer cases.
Then, we focus on the computational challenges of optimizing spatiotemporally fractionated
plans for clinical cancer cases. The main challenge is the non-convex and large-scale nature that
arises in the proposed optimization models. We develop customized solution methods that use
sequential quadratic/linear programming in a constraint generation framework. We also provide
optimality bounds on the found feasible solutions by solving Lagrangian relaxation using a
column generation approach. We also extend our spatiotemporal planning approach to
explicitly account for a major source of uncertainty that may compromise the potential
therapeutic benefit achievable from spatiotemporal fractionation. Specifically, we incorporate
the uncertainty associated with inter- and intra-patient variability in the radio-biological
parameter values. To account for such uncertainties, we propose a robust optimization
approach, which optimizes over a range of possible realization of the uncertain parameters. We
test the computational performance of the proposed solution methods using stylized as well as
de-identified clinical cancer cases.
Description
Thesis (Ph.D.)-- Wichita State University, College of Engineering, Dept. of Industrial, Systems and Manufacturing Engineering