Segmentation and volumetric quantification of thalamic nuclei for assessing MS

描述（申请人提供）：丘脑在整合感觉信息以进一步处理基底神经节和皮层方面起着关键作用。长期以来，多发性硬化症（MS）被认为主要是一种白质疾病，最近的研究表明，认知能力下降与丘脑体积的关系比与白质磁共振成像（MRI）损伤负荷的关系更强。由于丘脑是由大脑内具有特定物理连接的核组成的，因此可能将MS引起的丘脑核的物理变化与特定的认知、行为或疾病亚型差异联系起来。该拨款建议开发一种自动化方法和相关软件工具，使用MRI进行丘脑核包裹。具体来说，我们提出：1)优化从解剖和扩散MRI中得到的丘脑特征的计算；2)开发一种集成的多核丘脑分割算法；3)人工圈定算法参数优化；4)使用现有的MRI数据库进行一项试点研究，该数据库包括99名正常对照和226名MS患者。这项工作建立在以前的方法，利用拓扑和连通性，以提高分割鲁棒性。主要创新是提供了一种协调的多目标方法，该方法将t1加权MRI的强度信息与扩散MRI获得的方向信息和连通性信息相结合。将主要扩散方向映射到一个五维空间，以便根据扩散方向对核进行聚类，并将该信息用于分块算法。应用于手动描绘的机器学习方法将用于学习特定于边界的属性，这些属性将用于执行联合分割方法。该算法将设计用于常规的3特斯拉临床MRI，并将使用高分辨率、高信噪比的7特斯拉MRI对15名受试者进行验证，同时进行3特斯拉扫描。该试点研究将对305名参与者进行822次扫描，并将检查算法的纵向稳定性，并将丘脑核（或核群）体积与各种临床指标（包括疾病亚型、疾病持续时间、视力和两种标准MS复合残疾评分）相关的横断面单变量统计分析。将对多个丘脑核体积进行探索性主成分分析，以寻找萎缩的模式及其与各种临床措施的关系。该算法

DESCRIPTION (provided by applicant): The thalamus plays a key role in integrating sensory information for further processing in the basal ganglia and cortex. In multiple sclerosis (MS), long thought to be primarily a white matter disease, it has recently been shown that cognitive decline is more strongly related to thalamic volume than to white matter magnetic resonance image (MRI) lesion load. Since the thalamus is made up of nuclei having specific physical connections within the brain, it may be possible to relate physical changes in thalamic nuclei caused by MS to specific cognitive, behavioral, or disease subtype differences. This grant proposes to develop an automated method and associated software tool to carry out thalamic nuclei parcellation using MRI. Specifically, it is proposed to: 1) optimize the computation of thalamic features from anatomical and diffusion MRI; 2) develop an integrated, multi-nuclear thalamus segmentation algorithm; 3) optimize the algorithm parameters using manual delineations; and 4) carry out a pilot study using an existing MRI database comprising 99 normal controls and 226 MS patients. The work builds on previous methods that exploit topology and connectivity in order to improve segmentation robustness. The primary innovation is to provide a coordinated multi-object approach that integrates intensity information from T1-weighted MRI with orientation information and connectivity information obtained from diffusion MRI. Primary diffusion directions will be mapped to a five- dimensional space in order to cluster nuclei by diffusion orientation and use this information in the parcellation algorithm. A machine learning approach applied to manual delineations will be used to learn boundary-specific properties that will be used to carry out a joint parcellation approach. The algorithm will be designed for conventional three tesla clinical MRI and will be validated using high-resolution, high signal-to-noise ratio seven tesla MRI on 15 subjects scanned contemporaneously with their three tesla scans. The pilot study will use 822 scans of 305 participants, and will examine longitudinal stability of the algorithm and a cross-sectional univariate statistical analysis relatng thalamic nuclei (or nuclear groups) volumes to various clinical measures including disease subtype, disease duration, visual acuity, and two standard MS composite disability scores. An exploratory principal component analysis of multiple thalamic nuclear volumes will be carried out to look for patterns of atrophy and their relationships to various clinical measures. The algorithm will be made publicly available as open source code on the NITRC website so that the entire neuroscience community will be able to use the algorithm to study other diseases or modify and extend it for other applications.