磁振造影腦模板之客製化建構

Abstract

腦部的磁振造影技術 (MRI)，已被廣泛的使用在研究人類的腦部結構上，腦 模板 (template) 提供一個共同對位 (registration) 的標準空間，做為分析腦 部結構和比較腦部結構差異的基準。然而，對於一個不適當的腦模板空間，可能 會造成受測者的磁振造影影像需要過大的空間形變，才能對位到腦模板上，對位 後的成像容易會產生較大的誤差，對於研究分析特定族群的腦部結構，如何建構 一個無偏頗 (unbiased) 的腦模板是必要的。

此研究的主要目的是發展一個標準的建構客製化腦模板 (customized brain template) 流程。首先，我們會運用影像分析，將所有的受測者影像，截 取出純腦 (brain-only) 結構影像和不同腦組織 (tissue) 影像。接著，我們會 選取一個參考影像當作起始的腦模板空間，藉由腦模板和受測者影像之間反覆性 的影像對位流程，逐步的優化該選取的起始腦模板影像，並得到一個代表性的影 像 (representative image)。最後，將所有的受測者影像對位到此代表性影像 亦為腦模板空間，在平均所有對位到該空間的影像後即可得到腦模板影像。

在此研究中，我們利用216 個正常受測者 (normal subject) 的影像，在台 灣建立一組腦模板稱為BTT216。在和ICBM152 腦模板的評比中，這216 個受測 者影像對位到不同的腦模板後的成像，BTT216 提供較高的影像相關性 (correlation)。在平均形變量 (magnitude of deformation) 的評比上，BTT216 亦提供較小的形變差異。除此之外，我們針對特定的研究族群 (study-specific) 建立該族群的腦模板，例如不同性別和年齡層的腦模板，在評比中亦顯示此特定 的腦模板影像相較於其他腦模板提供較好的對位正規化空間。

我們的研究中，建立一個標準的建構腦模板流程，並針對特定族群所建立的 客製化腦模板，提供一個更佳的對位空間。

The magnetic resonance imaging (MRI) of brain images has been widely used to reveal the structure of human brain. A template space or template image provided a standard space to analyse brain structure or compare the differences among subject groups. However, large spatial transformation due to improper template space may lead to large artifact of brain structures in the warped brain image. Therefore, how to construct an unbiased brain template suitable for specific studies is essential to structural brain analysis.

The main purpose of this study aims at the development of a standard procedure for constructing MRI customized brain template. At first, we applied image analysis to all subject images to segment T1 image into different brain tissues. Secondly, we selected a reference image as the initial template space and applied an iterative registration procedure, including affine and non-rigid transformation, to gradually refined the template space and obtained a representative image. Then we spatially normalized all subject images to this representative image and averaged the warped images to obtain brain template.

This study constructed brain templates in Taiwan from 216 normal subject images (BTT216), including brain-only, whole-brain, grey matter, white matter, and cerebrospinal fluid templates. In addition, we constructed the template images for study-specific subject images, such as gender templates and different age-group templates. The evaluation method showed higher image correlation of warped images in BTT216 template rather than ICBM152 template. The average magnitude of deformation field was also shown in lower variation between BTT216 and subject images. The results of correlation shown in high similarity between study-specific template image and warped subject images which is normalized into the study-specific template space.

The customized brain template provides a better common space for brain structure analysis. The proposed method could be used as a standard procedure for brain template construction.