Beam orientation optimization for intensity modulated radiation therapy using adaptivel2,1-minimization

Xun Jia, Chunhua Men, Yifei Lou, Steve JiangPhysics 2011 · DOI · arXiv

Abstract

Beam orientation optimization (BOO) is a key component in the process of intensity modulated radiation therapy treatment planning. It determines to what degree one can achieve a good treatment plan in the subsequent plan optimization process. In this paper, we have developed a BOO algorithm via adaptive $l_{2,1}$-minimization. Specifically, we introduce a sparsity objective function term into our model which contains weighting factors for each beam angle adaptively adjusted during the optimization process. Such an objective function favors a small number of beam angles. By optimizing a total objective function consisting of a dosimetric term and the sparsity term, we are able to identify unimportant beam angles and gradually remove them without largely sacrificing the dosimetric objective. In one typical prostate case, the convergence property of our algorithm, as well as how beam angles are selected during the optimization process, is demonstrated. Fluence map optimization (FMO) is then performed based on the optimized beam angles. The resulting plan quality is presented and is found to be better than that of equiangular beam orientations. We have further systematically validated our algorithm in the contexts of 5-9 coplanar beams for five prostate cases and one head and neck case. For each case, the final FMO objective function value is used to compare the optimized beam orientations with the equiangular ones. It is found that, in the majority of cases tested, our BOO algorithm leads to beam configurations which attain lower FMO objective function values than those of corresponding equiangular cases, indicating the effectiveness of our BOO algorithm. Superior plan qualities are also demonstrated by comparing DVH curves between BOO plans and equiangular plans.

Introduction

In Intensity Modulated Radiation Therapy (IMRT), the goal is to deliver a therapeutic dose of radiation to a tumor while minimizing exposure to surrounding healthy organs at risk (OARs). This is achieved by delivering radiation from multiple angles. The planning process involves two primary steps: Beam Orientation Optimization (BOO), which selects the delivery angles, and Fluence Map Optimization (FMO), which determines the intensity patterns for those beams. BOO is a complex combinatorial problem that has historically been addressed through trial-and-error or the use of equiangular beam setups. This study proposes a novel approach by reframing BOO as a continuous sparse optimization problem using adaptive $l_{2,1}$-minimization.

Methodology

The proposed algorithm utilizes the $l_{2,1}$ norm to enforce group sparsity. Unlike standard $l_1$ minimization, which might zero out individual beamlets, the $l_{2,1}$ norm treats all beamlets within a single angle as a group. This encourages the algorithm to eliminate entire beam angles that do not contribute significantly to the dosimetric goals.

A critical innovation is the adaptive reweighting mechanism. During the optimization process, weighting factors ($\mu_\theta$) for each candidate angle are dynamically adjusted. Angles that show promise in meeting dosimetric constraints are assigned lower penalties, while redundant or ineffective angles are penalized more heavily, effectively driving their weights toward zero.

Adaptive reweighting and sparsity selection: The relationship between beam weight norms ($||x^\theta||_2$) and adaptive penalty factors ($\mu_\theta$) across candidate angles.

The resulting convex optimization problem is solved using a computationally efficient forward-backward splitting algorithm, allowing for the integration of complex dosimetric constraints.

Results and Validation

The algorithm was validated using clinical data from five prostate cancer cases and one head-and-neck (HN) case. For the prostate cases, configurations of 5 to 9 coplanar beams were evaluated. The HN case provided a more complex anatomical challenge with multiple targets and OARs.

Key findings include:

  • Dosimetric Improvement: In a 7-beam prostate case, the BOO algorithm achieved a 14.2% reduction in the objective function value compared to the best equiangular plan.
  • OAR Sparing: Qualitative analysis via Dose-Volume Histogram (DVH) curves demonstrated superior sparing of critical organs, such as the rectum and bladder, without compromising tumor coverage.
  • Efficiency: The algorithm consistently identified "key angles" for specific sites. Notably, 6 optimized beams often outperformed 11 unoptimized equiangular beams, suggesting that optimized selection can reduce treatment times and potential patient motion errors.

Discussion and Conclusion

By treating beam selection as a sparsity problem rather than a combinatorial search, this adaptive $l_{2,1}$-minimization approach provides a robust framework for automated treatment planning. The method consistently outperforms standard equiangular configurations across various clinical scenarios. Future work will focus on extending the algorithm to non-coplanar beam geometries and improving computational speed through GPU-based acceleration to facilitate daily clinical use.

Summary updated

Abstract

Beam orientation optimization (BOO) is a key component in the process of intensity modulated radiation therapy treatment planning. It determines to what degree one can achieve a good treatment plan in the subsequent plan optimization process. In this paper, we have developed a BOO algorithm via adaptive l(2, 1)-minimization. Specifically, we introduce a sparsity objective function term into our model which contains weighting factors for each beam angle adaptively adjusted during the optimization process. Such an objective function favors a small number of beam angles. By optimizing a total objective function consisting of a dosimetric term and the sparsity term, we are able to identify unimportant beam angles and gradually remove them without largely sacrificing the dosimetric objective. In one typical prostate case, the convergence property of our algorithm, as well as how beam angles are selected during the optimization process, is demonstrated. Fluence map optimization (FMO) is then performed based on the optimized beam angles. The resulting plan quality is presented and is found to be better than that of equiangular beam orientations. We have further systematically validated our algorithm in the contexts of 5-9 coplanar beams for five prostate cases and one head and neck case. For each case, the final FMO objective function value is used to compare the optimized beam orientations with the equiangular ones. It is found that, in the majority of cases tested, our BOO algorithm leads to beam configurations which attain lower FMO objective function values than those of corresponding equiangular cases, indicating the effectiveness of our BOO algorithm. Superior plan qualities are also demonstrated by comparing DVH curves between BOO plans and equiangular plans.
BibTeX
@inproceedings{jia2011beam,
  author = {Xun Jia and Chunhua Men and Yifei Lou and Steve Jiang},
  title = {{Beam orientation optimization for intensity modulated radiation therapy using adaptive<i>l</i><sub>2,1</sub>-minimization}},
  booktitle = {Physics},
  year = {2011},
  doi = {10.1088/0031-9155/56/19/004},
}
Figure