 |
|
||||
| Home | Login | |||||
| About | Editorial board | Articles
|
NSI Publications
|
Search | Instructions | Online Submission | Subscribe | Videos | Etcetera | Contact |
|
||||||||||||||||||||
|
|
|
Construction of Indian human brain atlas
Correspondence Address: Source of Support: None, Conflict of Interest: None DOI: 10.4103/0028-3886.253639
Keywords: Indian brain atlas, MRI, structure probability map
The significant difference in the shape and size of human brains across different races poses a great challenge for functional and structural comparison in neuroscience research.[1] Hence, spatial normalization is a critical preprocessing step in automated neuroimage analysis techniques such as voxel-based morphometry (VBM). Spatial normalization involves matching the brain of individual subjects to a standard coordinate framework using various registration methods.[2] Such a representation of the brain using a standard coordinate framework is known as a brain Atlas More Details. Among other things, a brain atlas facilitates comparison and/or combining of findings from different brain imaging modalities (functional or structural, e.g., functional MRI (fMRI), diffusion tensor imaging (DTI), T1, T2, fluid-attenuated inversion recovery (FLAIR), etc.), different brain states (healthy or diseased), and different subjects.[3] The Talairach and Tournoux atlas, also popularly known as the Talairach atlas, is the oldest and the most widely used brain atlas.[4] This atlas is a set of drawings created manually from the postmortem brain sections of a 60-year-old French woman. A digital atlas is typically created by registering the volumes, which is a process of spatially aligning images. In a groupwise registration method, a common coordinate system is identified for a group of images such that all images are at a minimum equal deformation distance from the coordinate system. An atlas is then created by warping these images into this coordinate system and taking their average. The Montreal Neurological Institute (MNI) and the International Consortium for Brain Mapping (ICBM) created the first digital human brain atlas using the brain MRI volume of 305 young normal Caucasian subjects (male[M]/female[F] = 239/66), in the year 1993.[5] This atlas was created using a 12-parameter, linear, groupwise registration method. In recent years, MNI and ICBM have released two more brain atlases, namely, MNI-152[6] and ICBM-452.[7] The MNI-152 is also widely used as a standard coordinate system in many neuroscience studies. Atlases have also been constructed from multiple scans of an individual subject, examples of which are Colin 27[8] and French atlas.[9] However, these brain atlases do not account for differences across phenotypic groups (e.g., age, gender, race, or disease conditions) as they are created using brain MRI of mostly Caucasian population. Of late, many studies have shown that there is a morphological difference across phenotypes. Specifically, this difference relative to the Caucasians has been reported for the East Asian populations [1],[10] as well as for the African-American ones.[11] These studies have motivated the creation of population-specific brain atlases, as spatial normalization of non-Caucasian brain to Caucasian atlas can result in measurement errors.[12] Initial efforts at such population-specific brain atlases resulted in the Chinese atlas, created using 56 right-handed male subjects (age range: 1.81-24.6 years)[1] and the Korean atlas, constructed using MRI and positron-emission tomography images of 78 (49 male and 29 female) right-handed subjects (age range: 19.4-44.6 years).[13] An atlas with 27 subjects (M/F = 17/10) has also been constructed to study population-dependent variations in brain morphology among people from India. However, subjects considered are largely males from a local region.[14]
The study, as approved by the Institute Review Board, involved collecting MRI volumes of young adults after obtaining an informed consent in writing. The work described has been carried out in accordance with the Code of Ethics of the World Medical Association. Subjects 100 young healthy adults (M/F = 50/50), aged 21–30 years, were recruited for the study. The volunteers were largely senior undergraduate and postgraduate students hailing from different geographical locations of India. The age distribution of the volunteers is shown in [Figure 1]. Each subject underwent a medical examination to rule out any psychological or neurological disorders. All images were examined by an experienced neuroradiologist to ascertain that no structural abnormalities were present.
Image acquisition MR images of these subjects were acquired at three different hospitals in Hyderabad. The subjects were divided into three groups with each group having approximately equal male and female subjects. The first group (17M/16F) was scanned using a Siemens 1.5-T MRI scanner with T1 MPRAGE sequence, second group (16M/17F) was scanned using a GE 1.5-T MRI scanner with T1 BRAVO sequence, while the last group (17M/17F) was scanned using a Phillips 1.5-T MRI scanner for T1 3D TFE sequence. All MRI volumes were acquired in the sagittal plane. The quality of the MRI volumes was assessed and approved by an experienced radiologist. Scanning parameters can be found in [Table 1].
Data preprocessing Every MRI volume was preprocessed using a standard pipeline consisting of N4-Bias field correction [15] and denoising with nonlocal mean filtering.[16] As the data were collected from three different scanner manufacturers with different acquisition parameters, intensity range of all the volumes varied significantly. Intensity standardization was done using a histogram-matching-based method [17] as it is also an internal part of the advanced normalization tools (ANTs) package used for the Atlas construction (http://stnava.github.io/ANTs/). All the volumes were automatically aligned with respect to their anterior commissure (AC)–posterior commissure (PC) line using an algorithm proposed by Ardekani et al.,[18] available with automatic resistration and toolbox (ART) package (https://www.nitrc.org/projects/art/). This was further corrected by a senior radiologist using the medical image processing, analysis, and visualization (MIPAV) tool (http://mipav.cit.nih.gov/). Atlas construction An atlas was constructed with all the 100 T1 MRI volumes by a nonrigid groupwise registration method [19] using the ANT's toolbox. This method aids in the derivation of an optimal atlas which is unbiased with respect to both shape and appearance in the diffeomorphic space. The atlas derived from 100 volumes will henceforth be referred to as the IBA100 atlas. Two gender-specific atlases (IBA50M for male and IBA50F for female) were also constructed using the same pipeline. Structure map creation The 100 MRI volumes used for atlas creation were automatically segmented/marked for 7 structure pairs using label transfer:[20] thalamus, putamen, pallidum, hippocampus, amygdala, caudate nucleus, and accumbens area. These 3D markings were verified and corrected by a medical expert in the three views. The corrected volumes were cross-checked by another medical expert. The experts used ITK-SNAP tool (an interactive software application that permits navigation of 3-D medical imag edseveloped by researchers at University of Pennsylvania and Utah, USA), for correcting the markings of seven structure pairs. The markings were first transferred to the atlas space and the coregistered structure markings combined to generate a probability map for each structure.
The constructed atlases and structure maps are given in [Figure 2] and [Figure 3], respectively. They were analyzed at both global and local structure levels.
Global brain feature comparison across different population atlases The constructed atlas was validated against atlases available for other populations.[1],[3],[4],[5],[6],[7],[13],[21],[22] The global brain features such as length, width, height, and volume of whole brain were measured on the atlas and compared. These were defined on the cerebrum as follows: length: distance between extreme anterior and posterior points; width: distance between extreme right (R) and left (L) points; and height: distance between extreme superior (S) and inferior (I) points. These measurements were performed by a senior neuroradiologist. Further measurements made were the width-to-length (W/L), height-to-length (H/L), and height-to-width (H/W) ratio: the distance between AC and PC of the brain. We automatically segmented brain MRI volume in tissue probability class of gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) using FAST [23] algorithm of FSL toolbox.[7] The total brain volume was computed as the net volume of all the voxels for which sum of the three tissue probabilities was more than 0.5. The tabulated measurements in [Table 2] indicate that the brain size of western population (MNI305, MNI152, and ICBM452) is significantly higher compared to the eastern population (Chinese2020 and Korean96).
While the Indian brain is closer to the Chinese2020 brain in length, it differs in terms of width and height, whereas the difference in all dimensions is low when compared to the Korean96. Thus, the brain size varies even among the eastern population. Based on the tabulated results for IBA100 atlas, it can be observed that the Indian brain, on an average, is smaller in height, width, and volume compared to MNI152,[6] Chinese2020,[22] and Korean96[21] atlases. Gender-specific atlas analysis Brain size and shape have been reported to vary across gender.[21],[24] In order to verify if this holds for the Indian population, a gender-based separation and analysis was done of the 100 subjects for the whole brain as well as GM, WM, and CSF volumes. The global brain-size difference across gender was assessed using an independent t-test (two-tailed). Results are listed in [Table 3].
Based on these results, we observe that on average, the male brain is larger than female brain in all the three dimensions, namely, length, width, and height. Consequently, the male brain also has a higher total volume as well as higher CSF, GM, and WM volumes relative to the female brain. However, W/L, H/L, and H/W ratios are comparable for both genders. These findings are consistent with reports on other populations. Structure level brain feature comparison with other structure maps Next, we compared IBA100 with two different population atlases at the structure level, namely, LPBA40[25] and Chinese2020.[22] This comparison is motivated by the fact that atlases are popular in automated structure segmentation [26],[27] and selection of the structure atlas could be important. The Chinese2020 atlas is labeled by the automated anatomical labeling (AAL) atlas,[28] where 45 anatomical volumes of interest in each hemisphere are marked, and the LPBA40 has manual markings for 56 structures. IBA100 has manual markings for six structures in two hemispheres. Two of these structure pairs, namely, hippocampus and putamen, are common with other two population atlases and hence their average volumes (measured in the respective atlas space) were compared. These averages are tabulated as percentage of the total brain volume in [Table 4]. The size of structures of Indian population is more similar to that in Chinese population. Overall, however, there is a significant difference in the structure volumes across the three atlases. Thus, variation across population appears to also hold at the structure level.
Atlas validation by linear registration Validation was done using linear registration as it compensated for any of the global differences, such as rotation, scaling, translation, affine, etc., between the volume from validation set and the atlas volume. This allowed us to compare global brain features of the validation set before and after registration to an atlas space. For this purpose, MRIs of 15 additional volunteers (8 males, 7 females) were collected. They were divided into three groups of five volunteers. Each group was scanned using either GE, Siemens, or Phillips scanner using parameters mentioned in [Table 1]. Brain MRI volumes Iv of these 15 volunteers (5 male, 24.50 2.67) were aligned to each of the 3 different atlases, namely, IBA100, Chinese2020, and MNI152, using a 12-parameter transformation as implemented in advanced normalization tools (ANTs). The registered result is denoted in general as Ivr. The global brain features in Iv and Ivr were then compared using a paired t-test to obtain three P values. These are denoted as P1, P2, and P3, for MNI152, Chinese2020, and IBA100, respectively. The global deformations required to register a brain to an atlas can thus be quantitatively evaluated. A similar validation method was recently used [22] where it was reported that comparatively smaller global deformation was required to register brains of Chinese subjects to the Chinese2020. In [Table 5], we can observe that P1 and P2 have lower values for all the global brain features. This indicates that the distributions of Iv and Ivr obtained with MNI152 or Chinese2020 are significantly different implying a high global deformation. This is not true when registration is to IBA100 as P3 has higher values for all the brain features, indicating marginal difference between Iv and Ivr. This shows that IBA100 atlas is near to the average Indian brain.
Structure map validation with expert marking The probability map, created by using the 100 coregistered volumes, is used to create a maximum likelihood structure map. The maximum probability structure map is registered with each of the 15 validation volumes, used in 3.4, and the labels are transferred to get the corresponding labels for the 15 validation volumes.[20] On the same 15 volumes, we got markings from experts for two structure pairs: hippocampus and putamen. These markings were done by one medical expert and cross-checked by a senior expert. We analyzed whether there was a good agreement between population-specific structure-map-based segmentation and the expert markings for the two structure pair segmentations. For comparison, we used dice coefficient and volume statistics of two structure pairs – hippocampus and putamen, and the details are given in [Table 6].
Our work is mainly focused on studying the importance of population-specific atlases. The result of this study indicates that there is a significant difference between global size of Indian (IBA100) and Caucasian brain (MNI152). This difference was noteworthy for all three global measures, that is, length, width, and height. Comparison of Indian atlas (IBA100) with Chinese (Chinese2020) and Korean (KNE96) shows that Indian brain is comparable in terms of length with both the atlases, while it was significantly smaller in terms of height and width. From the analysis based on the structure maps (IBA100, LPBA40, Chinese2020), we can conclude that significant difference extends to the structure level. IBA100 is more comparable to atlases of its geographical neighbors, namely, the Chinese (Chinese2020) and Korean (KNE96) than distant Caucasian population (MNI152). Validation results of IBA100 against MNI152, Chinese2020, and Korean96 indicated a few positive aspects when dealing with a new volume from the Indian population: less global deformation is required to register to IBA100 atlas, and there is more similarity in size at global and structural levels with IBA100 relative to other atlases. We have released the following products publicly via https://cvit.iiit.ac.in/projects/mip/IBA100/Home.php
Spatial mismatch and mislocalization between brain structures have been reported when MNI-152 atlas was used instead of population-specific atlas.[12] Voxel-based morphometry (VBM) and fMRI analysis require spatial normalization and hence should benefit from using IBA100, when the volume to be analyzed is from the Indian population. The study is limited by the number of subjects (100) but the findings are compelling enough to motivate future work on building an atlas with a much larger as well as more heterogeneous mix set of subjects with a diverse set of education levels. The latter is reported to affect the brain anatomy.[29] Similarly, deriving atlases for different age groups will also be of interest to check the effect of age-related factors on brain anatomy. Financial support and sponsorship Department of Science and Technology, Govt. of India, under Grant SR/CSRI/194/2013(G). Conflicts of interest There are no conflicts of interest.
[Figure 1], [Figure 2], [Figure 3]
[Table 1], [Table 2], [Table 3], [Table 4], [Table 5], [Table 6]
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|||||
