An interactive deep-learning method to semi-automatically segment abdominal organs to support stereotactic MR guided online adaptive radiotherapy (SMART) for abdominal cancers

摘要 立体定向MRI引导的在线自适应放射治疗（SMART）是治疗 胰腺和其他上腹部癌症。SMART允许精确交付升级的 处方剂量的腹部肿瘤的目标，同时避免并发症的辐射 对肿瘤靶点周围的移动的胃肠道（GI）器官的毒性。在临床 SMART的工作流程，危险GI orangs（OAR）的手动分割是最重要的 这是最重要的步骤，也是劳动密集型的步骤。手动分割需要10分钟 平均值，但范围为5至22分钟。缓慢而昂贵的手动分割步骤直接 降低了在线SMART的可访问性和可负担性，并间接降低了 SMART的有效性归因于患者的分次内身体和器官运动。在这 研究，我们将开发一个基于深度学习的互动和半自动程序， 准确、快速地对GI OAR进行分段，使SMART更加高效、经济实惠。

Abstract Stereotactic MRI-guided online adaptive radiotherapy (SMART) is an effective treatment for the pancreas and other upper abdominal cancers. SMART allows precise delivery of escalated prescription dose to the abdominal tumor targets while avoiding the complications of radiation toxicity to the mobile gastrointestinal (GI) organs surrounding the tumor target. In the clinical workflow of SMART, manual segmentation of the GI orangs at risk (OARs) is one of the most important but also the most labor-intensive steps. Manual segmentation takes 10 minutes on average but ranges from 5 to 22 minutes. The slow and costly manual segmentation step directly decreases the accessibility and affordability of online SMART and indirectly reduces the effectiveness of SMART due to intra-fractional body and organ movement of the patients. In this study, we will develop a deep-learning based interactive and semi-automatic procedure to accurately and quickly segment the GI OARs to make SMART more efficient and affordable.