Tools for automated cell identification and cell lineage tracking

该项目是Broad研究所、Amnis、英国癌症研究所和斯旺西纳米健康中心之间的合作项目。休假将允许Paul Rees教授访问Broad研究所和Amnis，以开发用于测量细胞谱系的新工具，并在大细胞群中自动识别特定细胞类型。通过与英国癌症研究所、Broad研究所附属医院和英国的一系列合作者合作，这些工具将根据临床医生的需求量身定制。然而，其目的是促进两个美国合作伙伴——斯旺西大学（Swansea）和伦敦研究所（CRUK）——之间的长期合作。因此，我们制定了一项长期的研究计划，该计划将由保罗·里斯的访问推动，并由斯旺西团队其他成员未来计划的访问来维持，并为美国合作伙伴访问斯旺西提供条件。一个有机体的功能是由一个细胞群的进化决定的，这些细胞群都是从一个祖细胞进化而来的。一个细胞群从一个祖细胞进化而来的谱系（关系树）经常被用来衡量该细胞群或生物体的健康状况。确定谱系最合适的方法是使用显微镜（亮场）拍摄细胞群的延时图像。跟踪细胞运动，确定有丝分裂事件，并注意到亲本和子细胞之间的适当关系。这可以手动完成，或者最近研究人员正在开发自动细胞跟踪算法。然而，这个过程是计算密集型的，如果细胞移出焦点或如果细胞边界与背景相比对比度低，则会失败。在这个项目中，我们的想法是简单地使用荧光核内体作为细胞的替代标记，因此通过简单地跟踪核内体，我们就可以跟踪细胞。许多临床和研究应用依赖于鉴定具有大量细胞群的特定细胞类型。这类应用的最佳技术之一是流式细胞术，其中细胞流过激光并检测散射光。这样就可以测量细胞的大小和结构以及标记细胞结构和功能的荧光。在斯旺西，我们使用最近开发的成像流式细胞仪，这是一种混合系统，通过在令人兴奋的激光器和相机之间的流体中流动细胞，可以以非常高的速度对细胞群中的每个细胞进行成像。这是一个理想的平台，通过图像分析来识别特定的细胞类型，而不是荧光标记或散射信号的强度，这些信号不提供空间信息，是对细胞特性的更模糊的间接测量。然而，成像细胞术的本质意味着任何测量都需要用户有效地处理大量图像以检测所需细胞的特征。这通常是使用细胞仪提供的基本图像处理工具手动完成的，并且检查每个单独的图像以检查目标细胞，这对于通常超过10^6的细胞群来说是令人难以置信的耗时。作为第二个项目，我们的目标是开发一种工具，它使用进化算法（或遗传编程）和机器学习来确定哪种图像处理算法（和算法组合）最好地区分目标细胞和非目标细胞。所使用的算法将与Amnis目前免费提供的IDEAS imagestream软件兼容（该软件目前只允许简单的用户驱动屏蔽过滤器来评估细胞），从而允许将先进的机器学习和进化算法纳入IDEAS平台。

This project is a collaboration between the Broad Institute, Amnis, Cancer Research UK and the Centre for Nanohealth in Swansea. The sabbatical will allow Professor Paul Rees to visit both the Broad Institute and Amnis in order to develop new tools which will be used to measure cell lineages and to automate the identification of specific cell types in large cell populations. These tools will be tailored to the needs of clinicians by collaborating with Cancer Research UK, Broad Institute affiliated hospitals and a range of collaborators in the UK. However the aim is to foster long term collaboration between the two US partners, the group at Swansea and CRUK, London Research Institute. Therefore we have developed a long term research programme which will be instigated by Paul Rees's visit and sustained by future planned visits for the other members of the team in Swansea and provision for the US partners to visit the Swansea.The function of an organism is determined by the evolution of a cell population all descended from a single progenitor cell. The lineage (relationship tree) of a cell population evolving from one progenitor cell is often used as a measure of that cell population or organism's health. The most appropriate method of determining lineage is to take time lapse images using microscopy (bright field) of the cell population. The cell movement is tracked and mitosis events identified and the appropriate relationship between parent and daughter cells noted. This can be done manually or recently researchers are developing automated cell tracking algorithms. However this process is computationally intensive and fails if the cells move out of focus or if the cell boundary has a low contrast compared with the background. For this project our idea is to simply use the florescent endosomes as a surrogate marker for the cell so by simply tracking the endosomes we track the cell. Many clinical and research applications rely on the identification of a particular cell type with a large cell population. One of the best techniques for this type of application is flow cytometry where cells flow past a laser and the scattered light is detected. This allows the cell size and structure to be measured together with the fluorescence from markers which can label cell structure and function. At Swansea we use the recently developed imaging flow cytometer which is a hybrid system that enables each individual cell within a cell population to be imaged at very high speeds by flowing the cells in a fluid between an exciting laser and a camera. This is an ideal platform identifying specific cell types by image analysis rather than the intensity of a fluorescent marker or scatter signal which provides no spatial information and is a more ambiguous indirect measure of the cell property. However the very nature of imaging cytometry means any measurement requires the user to effectively process the vast number of images to detect the traits of cells required. This is usually done manually using the basic image processing tools supplied with the cytometer and each idividual image is inspected to check for target cells which is incredibly time consuming with cell populations often in excess of 10^6. As a second project we aim to develop a tool which uses both evolutionary algorithms (or genetic programming) and machine learning to determine which image processing algorithms (and combinations of algorithms) best distinguish between the target cells and non target cells. The algorithms used will be compatible with the current IDEAS imagestream software provided freely by Amnis (which currently only allows simple user driven masking filters to assess cells) to allow the inclusion of advance machine learning and evolutionary algorithms into the IDEAS platform.