Wellcome Trust Centre for Neuroimaging, UCL Institute of Neurology, University College London, London, United Kingdom.
PLoS One. 2013 May 17;8(5):e63842. doi: 10.1371/journal.pone.0063842. Print 2013.
Brodmann's 100-year-old summary map has been widely used for cortical localization in neuroscience. There is a pressing need to update this map using non-invasive, high-resolution and reproducible data, in a way that captures individual variability. We demonstrate here that standard HARDI data has sufficiently diverse directional variation among grey matter regions to inform parcellation into distinct functional regions, and that this variation is reproducible across scans. This characterization of the signal variation as non-random and reproducible is the critical condition for successful cortical parcellation using HARDI data. This paper is a first step towards an individual cortex-wide map of grey matter microstructure, The gray/white matter and pial boundaries were identified on the high-resolution structural MRI images. Two HARDI data sets were collected from each individual and aligned with the corresponding structural image. At each vertex point on the surface tessellation, the diffusion-weighted signal was extracted from each image in the HARDI data set at a point, half way between gray/white matter and pial boundaries. We then derived several features of the HARDI profile with respect to the local cortical normal direction, as well as several fully orientationally invariant features. These features were taken as a fingerprint of the underlying grey matter tissue, and used to distinguish separate cortical areas. A support-vector machine classifier, trained on three distinct areas in repeat 1 achieved 80-82% correct classification of the same three areas in the unseen data from repeat 2 in three volunteers. Though gray matter anisotropy has been mostly overlooked hitherto, this approach may eventually form the foundation of a new cortical parcellation method in living humans. Our approach allows for further studies on the consistency of HARDI based parcellation across subjects and comparison with independent microstructural measures such as ex-vivo histology.
布罗德曼 100 年前的综述图在神经科学中被广泛用于皮质定位。现在迫切需要使用非侵入性、高分辨率和可重复的数据来更新该图谱，以捕捉个体差异。我们在此证明，标准 HARDI 数据在灰质区域之间具有足够多样化的方向变化，可以将其划分为不同的功能区域，并且这种变化在扫描之间是可重复的。这种将信号变化描述为非随机和可重复的特征是使用 HARDI 数据进行成功皮质分割的关键条件。本文是朝着个体皮质全灰质微观结构图谱迈出的第一步。高分辨率结构 MRI 图像确定了灰质/白质和软脑膜边界。从每个个体收集了两个 HARDI 数据集，并与相应的结构图像对齐。在表面剖分的每个顶点处，从 HARDI 数据集中的每个图像在灰质/白质和软脑膜边界之间的中点提取扩散加权信号。然后，我们根据局部皮质法向方向推导出 HARDI 轮廓的几个特征，以及几个完全各向同性不变的特征。这些特征被视为潜在灰质组织的指纹，并用于区分不同的皮质区域。在三个志愿者的重复 1 中的三个不同区域上训练的支持向量机分类器，在重复 2 中的未见过数据中对相同的三个区域实现了 80-82%的正确分类。尽管迄今为止灰质各向异性大多被忽视，但这种方法最终可能成为活体人类新皮质分割方法的基础。我们的方法允许进一步研究 HARDI 分割在不同受试者之间的一致性，并与独立的微观结构测量（如离体组织学）进行比较。
