自動建構在無偏頗三維量度空間中的磁振造影腦模板

Abstract

在人腦分析研究中，腦模板對於提供一個標準座標空間是不可或缺的。藉由人腦空間正規化（spatial normalization），將腦部磁振造影（MRI）對應到腦模板的標準空間中，可對應出腦部功能與結構資訊。並且，將不同受試者的腦部磁振造影對應到同一標準空間中，才能進行合理的統計分析和比較。MNI305和ICBM152是目前被廣泛使用的腦模板，然而因為不同研究群（study group）的腦部結構差異性，例如不同種族、性別和年齡的腦部結構差異，利用非此研究群所建立的腦模板進行空間正規化會導致對位（registration）的不精準性，因此，自定義模板（customized template）對於特定研究群的腦部分析中是必須的。因此，我們提出了一個自動化建立腦模板的方法，基於不需人工定義重要結構（landmark）自動化的流程，本方法可用於建造標準腦模板或自定義腦模板，此外，所建立之腦模板與研究群為無偏頗的（unbiased）。

論文的主要內容是有關腦模板的建立方法。首先，我們利用一現有的工具－Non-uniform intensity Normalization (N3)來校正腦部磁振造影的影像亮度不均勻性（nonuniformity）。從研究群中挑選一個參考腦，此參考腦具有最小方差的非線性形變量度（nonlinear deformation magnitude）。接著我們利用參考腦與其他受試者的腦部磁振造影來計算一個屬於此研究群的無偏頗空間（unbiased space），最後，將研究群中所有腦經過對位到此無偏頗空間後作平均，以建立無偏頗的腦模板。此外，藉由使用MNI腦模板作為媒介，我們也提供了自所建立的腦模板自動化對應到Talairach空間的座標轉換。

我們的研究中，提出了一個自動化建構無偏頗腦模板的流程。在實驗中證明了，使用研究群所建立的腦模板確實能減少空間正規化所造成的形變量，無偏頗的腦模板與研究群之間的變異亦減少，此外也提升了對位的準確性。

In brain diagnosis, brain template coordinate system, which serves a standard stereotactic space, is indispensable to providing a common space for pathology detection in individuals or groups. Inter-subject brain comparison can be achieved by registering different MR images to the standard space. Furthermore, it also guides algorithms for knowledge-based image labeling by registering individual brain to the template space containing a set of anatomical and functional labels annotated at specific coordinates. There are many widely-used templates, such as MNI305 and ICBM152. However, normalizing brains to these templates arbitrarily may cause structure artifact due to large spatial distortion. Therefore, a customized brain template is necessary for structure brain analysis for specific study group. We proposed an automatic procedure to create the standard or study-specific brain templates. This non-manual and automatic process provides a convenient and efficient method to generate templates without manual landmark-definition artifact.

In this study, we develop associated algorithms to automatically construct MRI brain templates from a database containing brain MRI volumes of Taiwanese for both genders. First, we use Non-parametric Non-uniform intensity Normalization (N3) technique to correct the nonuniformity of image. We choose a brain volume, which is one subject of the image set and has the minimum variation of deformation magnitude to the other subjects, as the representative brain. Thirdly, we compute the unbiased space according to the representative brain and all other brain images. Finally, we normalize all images to the unbiased space and average them to generate the brain template. Otherwise, we provide the automatic transformation from the created template to Talairach coordinate system by using the MNI template as the bridge to Talairach space.

In this work, we proposed an automatic procedure of brain template construction. We demonstrate that the constructed study-specific brain templates can reduce the amount of spatial distortion of normalization and improve the registration accuracy.