A Novel Contour-based Machine Learning Tool for Reliable Brain Tumour Resection (ContourBrain)

胶质瘤是一种侵袭性脑瘤，存活率各不相同。在外科手术中，很难在降低复发风险和保护大脑功能之间取得平衡。这主要是因为人工勾画肿瘤是主观的、劳动强度大的，而且不同的从业者不同，导致不可靠的分割和高复发率。因此，迫切需要可靠的、自动化的肿瘤分割工具，以帮助外科医生在肿瘤控制和功能保留之间实现最佳平衡，减少医生花费的时间和资源，并为未来的分析提供量化数据。然而，目前的自动分割方法，包括深度学习技术，可能会受到使用确定性边界来划定肿瘤浸润区的限制，这在高度不确定的情况下可能是有问题的。因此，本项目的目标是开发一种新的统计机器学习方法，利用部分标记的临床信息来做出更有信息量和更可靠的脑肿瘤切除术前决策。这一新方法有望为外科医生提供更多信息和细微差别的指导，增强他们准确计划手术的能力，在保留功能和可靠性的同时降低肿瘤复发的风险。预计这一新方法将推广到其他类型的基于MRI的癌症诊断，并有可能显著推进人工智能推动的肿瘤切除，并改善患者的预后。这项研究有两个受益者：(I)一个由英国和国际临床外科医生组成的大型社区，他们以传统方式进行脑瘤切除。该项目的成果将有助于肿瘤切除的术前决策，并大幅提高成千上万患者术后的生活质量，从而产生重大的社会经济影响。(Ii)致力于基于核磁共振的肿瘤研究的英国和国际临床学者/专业人员的大型社区。本项目产生的新颖的统计机器学习工具和思想将更广泛地适用于其他类型的基于MRI的癌症诊断和勾画。这将有助于进一步研究基于图像的肿瘤手术的可靠人工智能技术。已经精心设计了一些活动，以便有效地与这项研究的受益者接触。这些活动包括与临床学者共同制作和验证知识，在领先的学术期刊/会议上发布结果，在GitHub上宣传最新的项目进展和共享开源软件，以及与基于MRI的肿瘤外科领域的现场专家和国家学术和非学术利益相关者举行研讨会。

Glioma is a type of aggressive brain tumor that has varying survival rates. In surgical operations, it can be difficult to strike a balance between reducing the risk of recurrence and preserving brain function. This is mainly due to the fact that manual tumor delineation is subjective, labor-intensive, and varies among practitioners, leading to unreliable segmentation and high recurrence rates. Therefore, there is an urgent need for reliable and automated tumor segmentation tools to assist surgeons in achieving an optimal balance between cancer control and functional preservation, reducing the time and resources spent by doctors, and providing quantitative data for future analysis. However, current automated segmentation approaches, including deep learning techniques, can be limited by the use of a deterministic boundary to delineate the tumor-infiltrating area, which can be problematic in cases with high uncertainty.Therefore, the aim of this project is to develop a novel statistical machine learning approach that utilises partially labelled clinical information for more informative and accountable pre-surgery decision making in brain tumour resection. This new method is expected to provide more informative and nuanced guidance to surgeons, enhancing their ability to plan the surgery accurately, reducing the risk of tumour recurrence while preserving function and reliability. The new approach is expected to be generalised to other types of MRI-based cancer diagnostics and have the potential to significantly advance AI powered tumour resection and improve patient outcomes.The research has two streams of beneficiaries: (i) A large community of UK and international clinical surgeons that conduct brain tumor resection in traditional ways. The outcomes of this project would assist the pre-operative decision making for tumor resection, and substantially improve thousands of patients' quality of life after surgery, therefore achieve significant socioeconomic impact. (ii) A large community of UK and international clinical academics/professionals who work on MRI-based tumor research. The novel statistical machine learning tool and idea generated by this project will be more widely applicable to other types of MRI-based cancer diagnostics and delineations. This will assist further investigation of accountable AI techniques for image-based tumor surgery. A number of activities have been carefully designed to effectively engage with beneficiaries of this research. These activities include co-production and validation of knowledge with clinical academics, publishing of the results in leading academic journals/conferences, publicize up-to-date project advances and share open-source software on GitHub, and a workshop with field specialists and national academic and non-academic stakeholders in MRI-based tumor surgery.