In this paper, we presented the construction, validation and utilization of a new Chinese brain template using high resolution 3.0T T1 structural MR images. The within- and between-slice spatial resolution was chosen to be 0.47×0.47mm and 0.7mm, respectively, to achieve a good spatial resolution for more detailed structural information. The data included a total of 63 Chinese male volunteers ranged in age from 20 to 30yr (mean age=24.49±1.76yr). The Chinese_56 atlas was created using 56 volumes and it was validated on the remaining 7 subjects.
Dissimilarities of genetics and environmental exposures between different populations lead to differences in brain structure and function. Areas of functional differences between Chinese and Caucasian groups have been identified by a rapidly growing body of imaging studies (Kuo et al., 2001
; Kuo et al., 2003
; Tan et al., 2001a
; Tan et al., 2001b
; Tan et al., 2003
; Tan et al., 2000
). Not all regions reported as having functional differences were seen to have anatomical difference. However, each area where anatomical differences were observed has been selectively implicated in Chinese language processing by one or more studies. Further, the areas detected as being anatomically different between the groups have shown robust and highly reproducible functional differences (Kochunov et al., 2003
). To explore the anatomical differences between Chinese and Caucasian brains, we selected two comparable samples (35 subjects for each group) from the Chinese and Caucasian populations that are matched for gender and age. Although global brain shape and size can not provide detailed structural information throughout the human brain, these measures are important for comparing different brains. Analysis of these morphometric measurements indicated that the mean values of length, width, height and AC-PC line distance were significantly different (p
<0.01) between the Chinese brain and Caucasian brain. Thus, if Caucasian-based brain atlases are employed as reference templates in Oriental neuroimaging studies, some bias, processing errors or localized differences may be observed that are driven by the intrinsic differences between these two cohorts, and not caused by the underlying process investigated in the studies.
We used the LONI BrainParser software (Tu et al., 2008
) to measure the regional volumes of 56 brain structures for all 70 subjects. For each subject we obtained 56 contiguously labeled structures in the subject native delineation space. After analyzing the volumes of all the 56 structures, we found that some brain structures were significantly different (p
<0.01) between the Chinese and Caucasian brains. For instance, some structures of the Chinese brains (e.g., the middle orbitofrontal gyrus, the gyrus rectus, the left superior parietal gyrus, the superior temporal gyrus, the middle temporal gyrus, the inferior temporal gyrus, the left parahippocampal gyrus, the lingual gyrus, the left cingulated gyrus, the putamen) are larger than their counterparts in the Caucasian brains. Whereas some structures (e.g., the right superior frontal gyrus, the right precentral gyrus, the lateral orbitofrontal gyrus, the right postcentral gyrus, the right superior parietal gyrus, the right angular gyrus, the precuneus, the right superior occipital gyrus, the middle occipital gyrus, the right parahippocampal gyrus, the insular cortex, the caudate) are smaller compared to the corresponding regions in Caucasian brains. We also found the hemispheric asymmetries of the brain structures in both of the two population groups. This confirmed previous brain-asymmetry studies (Kuo et al., 2001
; Kuo et al., 2003
; Tan et al., 2001a
; Tan et al., 2001b
; Tan et al., 2003
; Tan et al., 2000
). These results demonstrate the need for lateral and population-specific data processing and analysis (including atlas-based spatial normalization) in modern computational neuroimaging studies of brain structure and function.
Our findings show that global and regional anatomical brain measurements are significantly different between the Chinese and Caucasian populations. Thus, widely popular Caucasian atlas templates (Evans AC, 1993
; Mazziotta et al., 2001
; Talairach J, 1988
; Toga and Thompson, 2001
) may not provide an optimal reference framework for processing brain images from the Chinese population. This implies that appropriate population-specific atlases (e.g., ethnic, gender, age, or disease) of average brain anatomy need to be employed in neuroimaging studies of well-stratified cohorts. To address this need for Oriental populations, we developed an average brain atlas specific to the Chinese population. As shown in , the Chinese brain template is about 168.77mm in length, 144.39mm in width, 110.64mm in height and AC-PC distance is 26.25mm. Compared to the widely-used ICBM152 brain template, the Chinese brain template is relatively shorter but wider, and its height is notably smaller. Among the 3 dimensional ratios, only the width/length ratio of the Chinese brain template is greater than that of the ICBM152 counterpart. This implies that the Chinese brain template is smaller but flatter. In the previous reports, the length, width and height of the Korean standard man brain template were estimated to be 16.50cm, 14.30cm and 12.10cm, respectively (Lee et al., 2005
), and Japanese hemispheres were relatively shorter but wider than European hemispheres (Zilles et al., 2001
). The shape and size measurements of the new Chinese brain atlas support these previous studies of regional brain differences between Asian and Occidental populations. In addition, the new Chinese brain template is composed of high quality data originated from subjects scanned using 3.0T MRI scanner, while the ICBM152 was constructed using the data originated from 1.5T MRI scanner. As a result, the resolution of the Chinese brain template is 0.47×0.47×0.47 mm3
, which is much higher than ICBM152 with the resolution of 1×1×1 mm3
. Thus, the Chinese brain template may capture more detailed and precise regional-based anatomical information about Oriental brains. The accuracy of the registration to the Chinese brain atlas was evaluated and the achieved results were compared with analogous co-registration results based on the ICBM152 target. For spatial normalization of individual Chinese brains, higher deformations were required to align these subjects into the ICBM152 template, compared to overall lower deformations necessary to register Chinese brains to the new Chinese_56 atlas. This suggests that the Chinese brain atlas is more precise for 12-parameter registration of Chinese cohorts into a common stereotaxic space.
The Chinese brain template (Chinese_56) is based on 56 subjects, while the ICBM 152 is based on 152 subjects chosen from a database with more than 7000 ICBM subjects. It is not known how many subjects are needed to build an optimal average brain template. But all the subjects used in the construction of Chinese_56 template were randomly selected young men aged from 20yr to 30yr. As age and gender are important factors in the delineation of brain structures and functions, and population-specific atlases are necessary for modern computational neuroimging, the sample size of 56 is appropriate to observe global and local patterns of group anatomical difference and to construct a brain template for Chinese populations. Cohort-specific brain templates are important for multi-subject structural or functional brain studies. Although this study focused only on ethnicity-specific brain atlases, it demonstrates that further research on other phenotypic characteristics such as gender, age, and disease should be taken into account for optimal and powerful analyses of regional brain morphometry. Such group-specific templates may replace the static atlases that are currently provided as default with many research tools. A population-specific brain atlas may increase the accuracy of the results by improving the statistical power and decreasing type I and type II error rates. For example, Oriental neuroimaging studies may replace the ICBM Caucasian brain atlas by the Chinese_56 template, as there are structural differences between these two populations (Kim et al., 2005
) . Therefore, it is necessary to make and distribute novel cohort-specific brain templates of well-stratified populations in the future. We will c
ontinue to collect and augment our Chinese database using different sub-populations (i.e., ethnic, gender, age, or disease), extend the Chinese brain atlas and establish smaller customized phenotypic brain templates for different sub-groups. There are also some potential limitations of the Chinese_56 atlasing framework. For instance, the Chinese_56 template is constructed using 3.0T MRI scans, whereas the 1.5T ICBM152 template and some future Oriental subjects studies using the Chinese_56 atlas may use 1.5T MRI acquisition protocols. These differences in the strength of the magnetic fields may introduce important variations between the imaging characteristics of the data and the Chinese_56 template (e.g., field effects, tissue intensity distributions, image contrasts, etc.). In the construction of the Chinese_56 atlas we averaged the set of 56 initially-aligned brains. This may cause anatomical detail to be lost, due to the reslicing interpolation. To mitigate this limitation we used sinc interpolation, which introduces the smallest (aliasing) artifacts. The final registration step employed a polynomial warp of 5th
degree. The decision of the complexity of the final registration step may also affect the final atlas, as 5th
degree may not be the optimal complexity for this population. An alternative is to geometrically average the warping fields, instead of the resliced volumes, and avoid the anatomical intensity averaging of the interpolation step.