Novel optimization framework for real-time automated radiation therapy

世界卫生组织估计，每年有超过800万人死于癌症，其中约70%在低收入和中等收入国家。放射治疗（RT）是一种使用X射线或粒子束杀死癌症患者体内特定细胞的方法，是帮助治疗癌症患者的最常用和最具成本效益的方法之一。据估计，超过50%的癌症患者可能在治疗过程中受益于接受RT，无论是单独还是与手术，化疗，激素治疗或免疫治疗相结合。然而，放射疗法治疗计划的递送是耗时的，涉及繁琐的治疗计划系统（TPS），在人员和基础设施方面都是昂贵的，并且可能受到不准确的计算模型的阻碍。这使得提供高质量和负担得起的治疗在全球范围内成为一项具有挑战性的任务，但也是一项对低收入和中等收入国家产生不利影响的任务。此外，虽然北美、欧洲、日本和澳大利亚的RT中心的可用性通常足以满足当前的需求，但在非洲（占估计需求的34%）和更广泛的亚太地区（61%），类似的覆盖率仍然很低。因此，全球大多数人口无法充分获得适当的癌症治疗。除非得到解决，否则这种情况预计将进一步恶化，因为预计未来十年中低收入国家的癌症发病率将大幅上升。因此，增加高质量癌症治疗的可用性，特别是RT，被认为是一个关键的全球和社会挑战。在此背景下，使放射治疗更广泛、更准确、更快、更经济，将在应对这一挑战中发挥重要作用。高质量放射治疗的提供依赖于准确的治疗计划系统，以针对不同癌症类型的光谱制定适当的放射治疗计划（TP）。优化这些TP可能需要大量的计算资源，并且通常是人员密集型的。在这个项目中，我们将开发一个基于先进的优化技术和远程超级计算的全自动治疗计划系统原型，从而解决在具有挑战性的远程环境中部署RT系统的重要困难。该系统将利用一系列先进的计算和机器学习技术，优化治疗规划系统的效率和稳健性，以期降低放射治疗中心的基础设施和人员成本，并在可能无法随时获得专业知识和大规模计算资源的情况下提供更大的灵活性。

The World Health Organization estimates that over 8 million people die of cancer every year, around 70% of them in low and middle income countries. Radiation therapy (RT), a process whereby x-ray or particle beams are used to kill specific cells in cancer patients, is one of the most commonly used and cost effective ways to help treat cancer patients. It is estimated that over 50% of all cancer patients may benefit from receiving RT during the course of their treatment, either on its own or in combination with surgery, chemotherapy, hormonal therapy, or immunotherapy. However, the delivery of radiation therapy treatment plans is time consuming, involves cumbersome treatment planning systems (TPS), is expensive both in terms of personnel and infrastructure, and can be hampered by inaccurate computational models. This makes the delivery of high-quality and affordable treatment a challenging task globally, but also one which disproportionally affects low and middle income countries. Further, while the availability of RT centres in North America, Europe, Japan and Australia is generally adequate to cover current needs, similar coverage remains poor in Africa (34% of estimated need covered) and in the wider Asia-Pacific region (61%). As a consequence, the majority of the global population does not have sufficient access to appropriate cancer treatment. Unless addressed, this situation is expected to worsen further given that cancer incidence rates are projected to grow significantly in low and middle income countries over the next decade. As such, increasing the availability of high-quality cancer treatment, and of RT in particular, is recognized as a key global and societal challenge. Within this context, making RT more widely available, increasingly accurate, faster, and more cost effective, will play an important part in addressing this challenge.The delivery of high-quality radiation therapy relies on accurate treatment planning systems to create appropriate radiation treatment plans (TP) across a spectrum of different cancer types. Optimizing these TP can require considerable computational resources and is often personnel-intensive. In this project we will develop a fully-automated treatment planning system prototype based on advanced optimization techniques and remote supercomputing, thus addressing an important difficulty in deploying RT systems in challenging and remote environments. The system will make use of a range of cutting-edge computational and machine learning techniques to optimize the efficiency and robustness of treatment planning systems, with the view to reduce infrastructure and personnel costs in radiation therapy centres, and to provide more flexibility for use where expertise and large-scale computational resources may not be readily available.