Advanced developments of diffusion MRI tractography computational methods

结构连接体是现代人脑图谱和神经信息学的核心。阿尔茨海默氏症等神经退行性疾病、自闭症等发育障碍、脑肿瘤和脑震荡等创伤性脑损伤影响着加拿大很大一部分人口。目前尚不清楚这些疾病是如何改变、退化和破坏白质中的纤维连接的。扩散磁共振成像(MRI)是唯一一种能够成像白质的神经结构并更好地了解大脑是如何连接的非侵入性技术。因此，扩散磁共振成像是一种在宏观尺度上测量连接体的工具，即在毫米尺度上测量短距离和长距离连接。 绘制白质连接体图的一个关键部分是从dMRI纤维束成像计算的测量值。纤维束造影术是重建白质纤维束的计算机化过程。遗憾的是，目前的纤维束成像技术不能控制算法产生的无效连接，不能重建现有纤维束的全部范围，也不是一种定量技术。到今天为止，众所周知的是，使用跟踪照相术发表的作品缺乏准确性，但大多被忽视了，因为没有公共数据库来验证技术。越来越多的证据表明，追踪图错误可能会导致错误的连通性解释。 我的研究计划的主要目标是提出一场扩散磁共振“光路成像革命”。射线成像必须作为一个困难的不适定反问题来解决，它需要所有可能的外部信息，以期以高质量、高精度和高重复性来解决该问题。第一个短期目标是提出一个标准化和开放的数据库，用于评价成功的制图术发展。第二个短期目标是将先进的多维(MD)数学和物理学融入到光谱分析过程中。中期目标是提出一个计算框架，将解剖学知识、显微结构信息和跟踪过程中的多模式信息结合在一起。另一个中长期目标是开发机器学习技术，通过结合过去的成功和失败信息，学习从错误中学习，学习什么应该跟踪，哪里不应该跟踪。 在连接组学应用的时代，神经科学和医学迫切需要解决脑束成像和结构脑成像的根本局限性。拥有新的和定量的光谱分析工具的潜在价值是非常高的。这将对加拿大和全世界的神经科学产生重大影响，因为它是人脑图谱和连接组项目的核心。它将有助于更好地描述广泛、大规模的神经元网络，这些网络是几种最复杂的人类神经疾病的基础。

The structural connectome is at the heart of modern human brain mapping and neuroinformatics. Neurodegenerative diseases such as Alzheimer's disease, developmental disorders such as autism, brain tumors and traumatic brain injuries such as concussions affect a large proportion of the Canadian population. It remains unknown how fiber connections in the white matter are altered, degenerated and damaged by these disorders. Diffusion magnetic resonance imaging (MRI) is the only non-invasive technique able to image the neural architecture of the white matter and better understand how the brain is wired. Diffusion MRI is thus a tool to measure the connectome at the macroscale, i.e. for short and long distances connections at the millimeter scale. A crucial part of mapping the white matter connectome are measurements computed from dMRI tractography. Tractography is the computerized process of reconstructing white matter fiber bundles. Unfortunately, current tractography techniques do not control for invalid connections produced by algorithms, do not reconstruct the full extent of existing tracts, and is not a quantitative technique. As of today, the lack of accuracy in published works using tractography is well known but has mostly been overlooked because there are no public databases to validate techniques. There is mounting evidence that tractography errors can lead to wrong connectivity interpretations. The main objective of my research program is to propose a diffusion MRI "tractography revolution". Tractography must be solved as a difficult ill-posed inverse problem, which needs all the external information possible to hope to solve the problem with high quality, accuracy and reproducibility. A first short-term objective is to propose a standardized & open database for tractography development evaluation of success. A second short-term objective is to incorporate advanced multi-dimensional (MD) mathematics and physics into the tractography process. A mid-term objective is to propose a computational framework that incorporates anatomical knowledge, microstructure information and multi-modality information inside the tracking process. Another mid to long-term objective is to develop machine learning techniques that learn from mistakes, learn what to track and where not to track, by incorporating the past information of successes and failures. At the era of connectomics applications, it is urgent for neuroscience and medicine to address the fundamental limitations of tractography and structural brain mapping. The potential value of having new and quantitative tractography tools is extremely high. This will have a significant impact on neurosciences in Canada and worldwide because it is central to human brain mapping and connectome projects. It will help better characterize the widespread, large-scale neuronal networks that underlie several of the most complex human neurological disorders.