Developing knowledge models to enable rapid learning in radiation therapy

摘要 本提案旨在开发一种能够为放射治疗提供证据的快速学习系统 为医生提供以患者为基础的治疗指导。快速学习的医疗保健是由 医学研究所将医疗保健提供转变为一种产生和应用的方式 可能需要为每个癌症患者提供最佳护理所需的证据“。 放射治疗(RT)是一种将复杂的放射传输设备应用于 提供高度适形的剂量分布，并将对危险器官的损害降至最低。因为 技术的复杂性，对辐射影响的不完全了解，以及患者和 患者情况下，通过学习使用目前的RT可以显著提高RT的有效性 最理想的技术。 在过去的几年里，我们小组和其他一些研究小组已经开发出调强放射治疗剂量 使用常规临床计划数据的预测和计划模型，在学习中产生了令人鼓舞的结果 规划知识，提高规划质量。这些努力代表着在第一方面取得的初步成功 快速学习系统。然而，这些现有的努力主要集中在几个主要的癌症部位和 仅限于“学习”方面。在扩展模型、翻译模型方面做了大量的进一步工作 临床实践中，需要闭合循环以进行持续学习，才能真正实现快速学习 放射疗法。本提案旨在制定一套全面和综合的模式和 能够以下列具体目标快速学习放射治疗的方法：(1)开发和 加强调强放射治疗计划模型，以涵盖所有主要癌症部位和治疗方案；(2)翻译 将模型转化为临床实践，以提供最佳的针对患者的RT计划并实现连续 通过增量学习对模型进行改进；(3)对知识模型进行验证和评估 快速学习框架的性能和价值。 虽然这个项目将集中于快速学习放射治疗的规划方面，但我们预计 可以扩展相同的框架以纳入结果数据，并最终实现完整的快速 以更低的成本不断提高癌症护理质量的学习框架。

Abstract The present proposal aims to develop a rapid learning system for radiation therapy that can provide evidence based, patient-specific treatment guidance for physicians. Rapid-learning health care is a vision proposed by Institute of Medicine to transform the health care delivery into one which generates and applies “as rapidly as possible the evidence needed to deliver the best care for each cancer patient”. Radiation therapy (RT) is a cancer treatment modality that applies complex radiation delivery equipment to deliver highly conformal dose distribution with minimized damage to organs-at-risk (OARs). Because of the complexity of technologies, the incomplete understanding of radiation effects, and the variability of patients and patient conditions, significant improvements in RT effectiveness can come from learning to use the current RT technologies optimally. In the past a few years, our group and a number of other research groups have developed IMRT dose prediction and planning models using routine clinical plan data that produced encouraging results in learning planning knowledge and improving plan quality. These efforts represent early successes in the first aspect of a rapid learning system. However, these existing efforts have mostly focused on a few major cancer sites and are limited to the “learning” aspect. Substantial further work on expanding the models, translating the models into clinical practice, and closing the loop for continuous learning is required to truly enable rapid learning in radiation therapy. The present proposal aims to develop a comprehensive and integrated set of models and methods that will enable rapid learning in radiation therapy with the following specific aims: (1) Develop and enhance IMRT planning models to cover all major cancer sites and treatment scenarios; (2) Translate the models into clinical practice to provide best-achievable patient-specific RT planning and enable continuous improvement of the models via incremental learning; (3) Validate the knowledge models and assess the performance and value of the rapid learning framework. While this project will focus on rapid learning of the planning aspect of radiation therapy, we anticipate that the same framework can be extended to incorporate outcomes data and ultimately lead to a complete rapid learning framework that leads to continuously improved quality of cancer care at lower cost.