Robust IMPT with automated beam orientation and scanning spot optimization

点击翻译按钮获取中文摘要

Proton beams have emerged as an appealing new modality for cancer therapy. With continuing clinical adoption and technical advances in the past two decades, intensity modulated proton therapy (IMPT) using scanning pencil beams has been established as the desired delivery method to fully take advantage of proton physics. Thus far, IMPT optimization has mainly focused on modulating the scanning spots with manually selected beam angles. At the same time, for intensity modulated X-ray therapy (IMXT), researchers including the PI's group have demonstrated that superior dosimetry can be attained with integrated beam orientation optimization (BOO). Nevertheless, the benefit of BOO has not extended to IMPT due to the paramount computational challenges of solving the integrated BOO and scanning spot optimization (SSO) problem, which by itself is a higher-dimensional problem than the fluence map optimization problem in IMXT. Currently, IMPT BOO is considered a combinatorial problem that is not mathematically tractable with increasing problem size. Despite the computational challenge, compared with IMXT, BOO is more important for IMPT for the following reasons. First, the optimal number and orientations of beams for IMPT have not been known. While the BOO problem in X-ray therapy is often circumvented in practice by using single or multiple arc beams, the same technique applied to IMPT would increase the volumes of normal tissue being irradiated by the entrance dose and would therefore start losing its low dose sparing advantage. Furthermore, because IMPT beam time is restrictive, using many beams in a treatment fraction is operationally impractical. Subsequently, IMPT plan quality is heavily influenced by each of the few selected beams. Yet, manual IMPT beam orientation selection in the available non-coplanar solution space is unintuitive and ineffective. Second, IMPT plans are highly degenerate with different combinations of beams, spots and spot sparsity resulting in similar dosimetry but vastly different robustness to uncertainties. Existing worst-case optimization methods are a suboptimal compromise between the dosimetry and robustness. It is hypothesized that both the dosimetry and robustness will be significantly improved by integrating BOO in IMPT optimization. It is then hypothesized that the integrated optimization problem can be formulated as a group sparsity optimization problem with efficient solutions. To test these hypotheses, the following aims are proposed. Aim 1. Develop automated beam orientation and sparse spot optimization for IMPT. Aim 2. Develop fraction-variant IMPT. Aim 3. Incorporate sensitivity regularization (SenR) for robust beam orientation and scanning spot optimization. Aim 4. Validation of the integrated BOO, SSO and robustness optimization framework. The first three aims will be mainly performed at UCLA with the clinical and physics input from UPENN. The last aim will be mainly performed at UPENN. Depending on the feedback, UCLA will provide technical support to use and validate the proposed treatment planning system.