Perrot Matthieu, Rivière Denis, Tucholka Alan, Mangin Jean-François
CEA, Neurospin, LNAO, Saclay, France.
Inf Process Med Imaging. 2009;21:176-87. doi: 10.1007/978-3-642-02498-6_15.
In this paper, we study the recognition of about 60 sulcal structures over a new T1 MRI database of 62 subjects. It continues our previous work [7] and more specifically extends the localization model of sulci (SPAM). This model is sensitive to the chosen common space during the group study. Thus, we focus the current work on refining this space using registration techniques. Nevertheless, we also benefit from the sulcuswise localization variability knowledge to constrain the normalization. So, we propose a consistent Bayesian framework to jointly identify and register sulci, with two complementary normalization techniques and their detailed integration in the model: a global rigid transformation followed by a piecewise rigid-one, sulcus after sulcus. Thereby, we have improved the sulci labeling quality to a global recognition rate of 86%, and moreover obtained a basic but robust registration technique.
在本文中，我们在一个包含62名受试者的新T1磁共振成像（MRI）数据库上研究了约60个脑沟结构的识别。它延续了我们之前的工作[7]，更具体地扩展了脑沟定位模型（SPAM）。该模型在群体研究中对所选的公共空间敏感。因此，我们将当前工作重点放在使用配准技术优化这个空间上。然而，我们也受益于脑沟定位变异性知识来约束归一化。所以，我们提出了一个一致的贝叶斯框架，通过两种互补的归一化技术及其在模型中的详细整合，来联合识别和配准脑沟：先进行全局刚性变换，然后逐个脑沟地进行分段刚性变换。由此，我们将脑沟标记质量提高到了86%的全局识别率，并且还获得了一种基本但稳健的配准技术。
