Modelling and Accounting for Delineation Uncertainties and Anatomical Changes in Proton Beam Therapy using Machine Learning

1)研究背景的简要描述，包括潜在影响该项目将利用最先进的机器学习方法来模拟危险器官（OAR）描绘的不确定性以及接受质子治疗的头颈癌患者的日常解剖变化。将针对这两种不确定性来源开发不同的模型。基于深度学习的方法将用于同时生成高质量的危险器官自动描绘，并估计和量化描绘中不同类型的不确定性（训练数据中的不确定性，例如手动描绘的可变性，模型参数的不确定性）。将研究统计形状/变形模型和基于深度学习的方法，利用在治疗过程中获得的患者的常规成像来模拟OAR日常变化的可变性。这些预测模型可用于预测新患者在质子治疗过程中桨叶形状和位置的不确定性，并最终指导稳健的质子治疗计划，以考虑这些变化。2)目的和目标该项目旨在使用最先进的机器学习方法来建模和估计影响质子治疗治疗的桨叶描绘和解剖变化的不确定性，并制定对这些不确定性具有鲁棒性的质子治疗计划。这分为以下几个目标：从CBCT扫描中产生适合剂量计算和分析的图像和描绘2。模型在患者群体中描述和解剖变化的可变性，并使用模型来估计新个体描述和解剖变化的不确定性3。生成适合于评估质子治疗计划对特定个体估计不确定性的稳健性的图像和轮廓4。使用更新的不确定性估计来预测是否/何时需要重新规划，并生成适合用于重新规划的图像和轮廓3)研究方法的新颖性这项工作将利用最先进的基于深度学习的方法来进行轮廓传播、合成CT生成和不确定性估计，这些方法已经为不同的应用而开发。并将以新颖的方式调整和结合这些方法，用于CBCT剂量计算的应用。然后将开发新的方法来模拟变异性和估计描绘和解剖变化的不确定性，并制定对估计的不确定性具有鲁棒性的质子治疗计划。4)与EPSRC的战略和研究领域保持一致本项目与以下EPSRC医疗技术重大挑战保持一致：物理干预的前沿；优化治疗。它还将建立以下交叉研究能力：新颖的计算和数学科学；任何公司或合作者参与本项目是与UCLH质子治疗团队合作的。不涉及工业合作者

1) Brief description of the context of the research including potential impactThe project will utilise state-of-the-art machine learning methods to model the uncertainties in the organ at risk (OAR) delineations and day to day anatomical changes to patients receiving proton therapy for head and neck cancer treatments. Different models will be developed for these two sources of uncertainty. Deep-learning based approaches will be used to simultaneously generate high-quality automatic delineations of organs at risk and to estimate and quantify the different types of uncertainty in the delineations (uncertainties in the training data, e.g. variability in the manual delineations, uncertainty in the model parameters). Statistical shape/deformation models and deep-learning based approaches will be investigated to model the variability in the day to day changes in OAR using routine imaging from patients acquired during the course of treatment. These predictive models can then be used to forecast uncertainty in shape and position of OAR for a new patient over the course of proton therapy and ultimately guide robust proton treatment planning to take account of these changes.2) Aims and ObjectivesThe project aims to use state-of-the-art machine learning methods to model and estimate the uncertainties in OAR delineations and anatomical changes that affect proton therapy treatments, and to develop proton therapy treatment plans that are robust to these uncertainties. This is split into the following objectives:1. Produce images and delineations from CBCT scans that are suitable for dose calculations and analysis2. Model variability in delineations and anatomical changes across the population of patients and use models to estimate uncertainty in delineations and anatomical changes for new individuals3. Produce images and contours suitable for assessing the robustness of proton therapy plans to the estimated uncertainties for a specific individual4. Update estimates of uncertainty for an individual as new CBCT images become available for that individualUse updated estimates of uncertainty to predict if/when a re-plan will be required and to produce images and contours suitable for use in generating the re-plan3) Novelty of Research MethodologyThis work will utilise state-of-the-art deep-learning based approaches to contour propagation, synthetic CT generation, and uncertainty estimation that have been developed for different applications, and will adapt and combine these approaches in new and novel ways for the application of CBCT dose calculations. New and novel approaches will then be developed for modelling the variability and estimating the uncertainty in the delineations and anatomical changes, and for producing proton therapy plans that are robust to the estimated uncertainties.4) Alignment to EPSRC's strategies and research areasThis project is aligned with the following EPSRC Healthcare Technologies Grand Challenges: Frontiers of Physical Intervention; and Optimising Treatment. It also will build the following Cross-cutting research capabilities: Novel computational and mathematical sciences; Novel Imaging Technologies5) Any companies or collaborators involvedThis project is a collaboration with the Proton Therapy team at UCLH. No industrial collaborators are involved