Multi-Task MR Simulation for Abdominal Radiation Treatment Planning

放射治疗计划的准确性在很大程度上影响外照射的效果。 放射治疗(EBRT)。个体化的RTP从一个“模拟”开始，在这个模拟中，患者处于治疗的姿势 通常使用计算机断层扫描(CT)来确定治疗目标和危险器官(OAR)。 当软组织对比度不足以支持基于CT的RTP中准确的靶点和OAR勾画时， 保守的大治疗边际被用来避免几何错失。粗暴的处理防止 在不超过周围正常组织耐受性的情况下向肿瘤提供足够的辐射剂量。 磁共振(MR)可以作为CT的模拟平台，补充CT以改善软组织 很显眼。然而，如此复杂、昂贵和繁琐的多模式RTP工作流伴随着不可避免的 系统性MR-CT联合配准误差限制了其在EBRT中的应用，特别是在腹部部位 因此解剖结构具有很高的流动性。在过去的几年里，人们对MR的整合产生了浓厚的兴趣 单独进入RTP，甚至进入治疗工作流程(即MR引导的放射治疗，MRGRT)。腹部姿势 MR模拟面临的关键挑战。目前的磁共振成像序列不太适合产生无运动 图像和解析呼吸运动。腹部RTP的MR数据处理还不发达。等高线绘制 目标和桨的使用通常依赖于耗时且容易发生变化的人工、繁琐的程序。 在这个方案中，我们将大幅改进MR采集和自动多器官分割，因此 MR作为一种模拟方式的潜力可以在腹部EBRT中得到充分释放。三个具体目标 将会完成。在目标1中，我们将开发并验证一个独立的多任务MR(MT-MR)序列 致力于腹部磁共振仿真。在目标2中，我们将开发一种基于MT-MR仿真的多器官自动... 分割工具。在目标3中，我们将优化基于深度学习的剂量预测模型，并评估 基于MT-MR的RTP工作流程在自适应立体定向体部放射治疗计划中的效果 胰腺癌患者。该项目的成功完成将极大地促进临床采用 腹部RTP的MR模拟，将提高治疗精度和预后。此外， 开发的技术将为未来的研究打开大门，旨在优化癌症诊断和治疗 放射治疗。

The accuracy of radiation treatment planning (RTP) heavily influences the effectiveness of external beam radiotherapy (EBRT). Individualized RTP begins with a “simulation”, in which the patient in a treatment position is commonly scanned using computed tomography (CT) to define the treatment target and organs at risk (OARs). When soft-tissue contrast is inadequate to support accurate target and OAR delineation in CT based RTP, conservatively large treatment margins are used to avoid a geometric miss. The crude treatment prevents delivering sufficient radiation dose to the tumor without exceeding the tolerance of surrounding normal tissues. Magnetic resonance (MR) can be used as a simulation platform complementary to CT for improved soft-tissue conspicuity. Yet, such a complicated, costly and tedious multi-modal RTP workflow along with unavoidable systematic MR-CT co-registration errors has limited its applications in EBRT, especially at the abdominal site whereby anatomies are highly mobile. Over the past few years, there is a keen interest in the integration of MR alone into RTP and even the therapy workflow (i.e. MR-guided radiotherapy, MRgRT). The abdomen poses critical challenges to MR simulation. Current MR imaging sequences are suboptimal to produce motion-free images and resolve respiratory motion. MR data processing for abdominal RTP is underdeveloped. Contouring of target and OARs typically relies on manual, tedious procedures that are time-consuming and variation-prone. In this proposal, we will substantially improve the MR acquisition and automated multi-organ segmentation, so the potential of MR as a simulation modality can be fully unleashed for abdominal EBRT. Three specific aims will be completed. In Aim 1, we will develop and validate a standalone multi-task MR (MT-MR) sequence dedicated to abdominal MR simulation. In Aim 2, we will develop an MT-MR simulation based multi-organ auto- segmentation tool. In Aim 3, we will optimize a deep learning-based dose prediction model and assess the effectiveness of the MT-MR based RTP workflow in adaptive stereotactic body radiotherapy planning of pancreatic cancer patients. Successful completion of the project will significantly promote the clinical adoption of MR simulation for abdominal RTP, which will improve treatment precision and outcomes. Moreover, the developed techniques will open the door to future studies aiming at optimizations in both cancer diagnosis and radiotherapy.