台灣人基準腦影像之建構

Abstract

中文摘要

基準腦對於功能性與結構性的腦部對應來說是必須的，因為基準腦提供了一個立體座標空間，並可在此座標空間中標記功能與結構區域。如此一來，我們便可以將人腦空間正規化（spatial normalization）到基準腦座標空間去，藉此對應出腦部功能與結構資訊。而且，利用統計方法以比較不同腦部結構與功能時，也必須在同樣的座標空間才能進行合理的比較。MNI305是目前廣泛被使用的基準腦，由Montreal Neurological Institute蒐集305個西方人的腦部磁振造影（MRI）所建立而成的。然而，不同種族之間具有相當大的腦部結構差異性，因此利用他族的基準腦進行空間正規化將會導致腦部結構的不合理變形。基於上述原因，一個本土、量身訂製化的基準腦對腦部結構分析而言是很重要的。

論文的主要內容是有關台灣人基準腦的建立方法。首先，我們提出中間矢狀切平面（mid-sagittal plane）自動對位技術。另外，在進行腦部對位時，我們選取兩個腦部的重要結構──前聯體（anterior commissure, AC)以及後聯體（postetrior anterior commissure, PC），並從頭部磁振造影影像資料庫中，挑選出一個參考腦，作為所有其他腦部影像的參考結構與大小。接著我們將資料庫中的所有腦部完成對位程序，也就是將中間矢狀切平面作為對位面，通過前聯體與後聯體的線（AC-PC line）作為對位軸，前聯體作為對位點，並縮放到參考腦的腦部大小，最後將經過對位程序的腦部作平均，以建立台灣人基準腦。

我們的研究證明了台灣人基準腦確實能夠減少在結構分析之中空間正規化所造成的不合理變形量。此外我們也經由比較後發現台灣人基準腦比西方人基準腦來得短且寬。

Brain template is essential to functional and structural brain mapping. It provides a standard stereotaxic space containing a set of anatomical and functional labels annotated at specific coordinates. An individual brain can be spatially normalized into this standard space to incorporate the annotation information. Furthermore, only in this same space that brain MR images can be compared to obtain statistical inference of structural discrepancy. MNI305 is a widely-used brain template, which was created by Montreal Neurological Institute from 305 brain MRI volumes of Western normal subjects. However, inter-ethnic difference of brain structure can be large. Normalizing the brain to a template of different race may cause structural artefact due to the large spatial distortion. Therefore, a customized brain template is necessary for structural brain analysis for Easter people.

In this work, we develop associated algorithms and construct a Taiwanese brain template from a database containing brain MRI volumes of Taiwanese for both genders. First, we propose an estimation technique that can automatically determine the mid-sagittal plane for each individual MRI. Then, the anatomical landmark, anterior commissure (AC), is selected as the origin point and another anatomical landmark, posterior commissure (PC), is selected such that the AC-PC line constitutes the mapping axis. We choose an individual MRI as the representative brain and register all other MRIs to the same stereotaxic coordinate space by aligning their mid-sagittal planes, origins, and mapping axes to those of the representative brain. Finally, the brain template is obtained by averaging all of the spatially normalized brain MRIs.

In this study, we also demonstrate that the constructed Taiwanese brain template can be used to reduce the amount of spatial normalization distortion when Taiwanese brain MRIs are involved in structural analysis. Another finding shows that Taiwanese brain template is shorter and wider than Western templates.