Decision Support System for Stereotactic Radiosurgery

立体定向放射外科决策支持系统

• 批准号: RGPIN-2019-04715
• 负责人: Ruschin, Mark
• 金额: $ 1.75万
• 依托单位: University of Toronto
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2021
• 资助国家: 加拿大
• 起止时间: 2021-01-01 至 2022-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=737810
• 关键词: Decision; Support; System; Stereotactic; Radiosurgery

The Gamma Knife (GK) system uses 192 small radiation beams that converge to deliver a high radiation dose within a brain tumour. This technique, called "stereotactic radiosurgery" (SRS) is an effective way of treating brain cancer. The point at which the beams intersect is called an "isocentre" and the treatment involves multiple isocentres to ensure the entire tumour receives a curative dose. Determination of the best isocentre positions and corresponding dwell times is a complex task, since there is a direct conflict between maximizing tumour dose and minimizing dose in surrounding healthy brain. Furthermore, as the number of patients requiring treatment is rapidly increasing due to an aging population and improved drug therapies, we need efficient computerized methods for automating this process. The present research program involves developing models of automation in order to obtain a deeper understanding of the complex interplay between the various objectives. The present program involves two main themes of research as follows. In the first theme, we will explore novel "inverse" automation models. In inverse methods we define the desired output and the computer infers the input. In SRS the desired output is a high radiation dose to the tumour and low dose to normal tissue and the input is the radiation dwell time at each isocentre. The mathematical formulation involves minimizing the numerical difference between the ideal and model-predicted radiation dose. Since the goals conflict (high tumour dose vs. low normal brain dose) and computers do not know what tradeoffs are acceptable, the user sets priorities in the algorithm via weighting factors. However since the relationship of such weight parameters to the solution is not known in advance, one needs to repeat the computation with different weight factors, and by doing so we generate a reasonable model of the problem. The mathematical models will ultimately feed back to end users to decide how weighting factors should be allocated, which will lead to completely automated planning in the future. In the second theme, our group will explore novel artificial intelligence (AI) based automation models. In AI, thousands of cases are used to define features such as tumour volume, and tumour histology. Novel mathematical approaches are then used to establish linkages between all of the features and the outcome, which can be a treatment plan quality metric. The result is a model that we can use to better understand the complex interplay between all of the features and outcome. It is possible to validate a given model by testing its predictive power for an unknown case, not included in the modeling itself. However the emphasis of the current program is in the engineering of the models to understand the problem. The successful completion of this program will lead to future clinically-oriented studies that can use such prediction models for informing plan quality outcome.

伽玛刀（GK）系统使用192个小辐射束，这些辐射束会聚在一起，在脑肿瘤内提供高辐射剂量。这种被称为“立体定向放射外科手术”（SRS）的技术是治疗脑癌的有效方法。光束相交的点被称为“等中心”，治疗涉及多个等中心，以确保整个肿瘤接受治疗剂量。确定最佳等中心位置和相应的停留时间是一项复杂的任务，因为在最大化肿瘤剂量和最小化周围健康大脑中的剂量之间存在直接冲突。此外，由于人口老龄化和药物治疗的改善，需要治疗的患者数量迅速增加，我们需要有效的计算机化方法来自动化这一过程。目前的研究计划涉及开发自动化模型，以便更深入地了解各种目标之间复杂的相互作用。本方案涉及以下两个主要研究主题。在第一个主题中，我们将探索新的“逆”自动化模型。在逆方法中，我们定义所需的输出，计算机推断输入。在SRS中，期望的输出是对肿瘤的高辐射剂量和对正常组织的低剂量，并且输入是在每个等中心的辐射停留时间。数学公式涉及最小化理想和模型预测的辐射剂量之间的数值差异。由于目标冲突（高肿瘤剂量与低正常大脑剂量），计算机不知道什么样的权衡是可接受的，用户通过加权因子在算法中设置优先级。然而，由于事先不知道这些权重参数与解的关系，因此需要使用不同的权重因子重复计算，并且通过这样做，我们生成了问题的合理模型。数学模型最终将反馈给最终用户，以决定如何分配权重因子，这将导致未来完全自动化的规划。在第二个主题中，我们的团队将探索基于人工智能（AI）的新型自动化模型。在人工智能中，成千上万的病例被用来定义肿瘤体积和肿瘤组织学等特征。然后使用新的数学方法来建立所有特征和结果之间的联系，这可以是治疗计划质量度量。结果是一个模型，我们可以用它来更好地理解所有功能和结果之间的复杂相互作用。可以通过测试其对未知情况的预测能力来验证给定模型，而不包括在建模本身中。然而，目前的计划的重点是在模型的工程，以了解问题。该计划的成功完成将导致未来的临床导向研究，可以使用这种预测模型来告知计划质量结果。