High-Fidelity Radiotherapy Treatment Planning via Dimension-Free Zeroth-Order Algorithms

该奖项将通过改进癌症治疗中的放射疗法（放疗）方法，为国家的健康和福利做出贡献。 放射治疗长期以来一直被用作癌症治疗的一种流行模式;其有效性在于使用高能辐射来根除癌细胞，同时保留周围的正常组织。 放射治疗的基础是确定安全有效治疗计划的复杂优化算法。 由于问题的高维性以及个体对辐射剂量反应的不确定性，难以创建准确的治疗计划。该项目开发了改进算法的方法，这些算法指导将精确量的辐射输送到靶细胞。 研究结果将被纳入继续医学教育计划，以促进学者和医生之间的合作。为了吸引最近的高中毕业生，特别是那些来自代表性不足的社区，进入STEM专业，项目团队将参加佛罗里达大学的STEPUP推广计划。该项目旨在从根本上创建新的零阶算法范例，这些范例可证明能够减轻该研究计划将研究随机无梯度算法的变体，这些算法利用了稀疏性及其泛化等计算便利结构。该项目还将推导和分析算法，这些算法将联合收割机优化和深度学习方法结合起来，在不了解封闭形式公式的情况下解决问题。在理论上，这些算法的计算效率预计是几乎独立的问题的维数，对数项。这些算法将与蒙特卡罗模拟器集成，该模拟器被视为提供放射治疗结局准确建模的金标准。由此产生的新的治疗计划引擎预计将提高计划保真度，而不增加计算成本。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This award will contribute to the Nation's health and welfare by improving methods for radiation therapy (radiotherapy) in the treatment of cancer. Radiotherapy has long been used as a prevalent mode of cancer treatment; its effectiveness lies in using high-energy radiation to eradicate cancer cells while sparing the surrounding normal tissue. Underlying the delivery of radiotherapy are complex optimization algorithms that determine safe and effective treatment plans. The creation of accurate treatment plans is difficult due to the high-dimensionality of the problems as well as to uncertainties in individual response to radiation dosage. This project develops methods to improve algorithms that guide the delivery of precise amounts of radiation to target cells. The research results will be integrated into a continuing medical education program to facilitate collaborations between academics and medical practitioners. To attract recent high school graduates, especially those from under-represented communities, into STEM majors, the project team will participate in the STEPUP outreach program at the University of Florida. This project aims to create fundamentally new zeroth-order algorithmic paradigms that are provably capable of mitigating the The research plan will study variations of randomized gradient-free algorithms that exploit computation-facilitating structures such as sparsity and its generalizations. The project will also derive and analyze algorithms that combine optimization and deep learning methods in solving problems without the knowledge of closed-form formulations. In theory, the computational efficiency of these algorithms is expected to be almost independent of problem dimensionality, up to a logarithmic term. These algorithms will be integrated with the Monte Carlo simulators deemed the gold standard in providing accurate modeling of radiotherapy outcomes. The resulting new treatment planning engines are expected to improve plan fidelity without increasing the computational cost. Extensive experiments and comparisons of the methods will be conducted on realistic cancer treatment data.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.