|Home | About | Journals | Submit | Contact Us | Français|
Diffusion tensor imaging (DTI) is a powerful and noninvasive imaging method for characterizing tissue microstructure and white matter organization in the brain. While it has been applied extensively in research studies of the human brain, DTI studies of non-human primates have been performed only recently. The growing application of DTI in rhesus monkey studies would significantly benefit from a standardized framework to compare findings across different studies. A very common strategy for image analysis is to spatially normalize (co-register) the individual scans to a representative template space. This paper presents the development of a DTI brain template, UWRMAC-DTI271, for adolescent Rhesus Macaque (Macaca mulatta) monkeys. The template was generated from 271 rhesus monkeys, collected as part of a unique brain imaging genetics study. It is the largest number of animals ever used to generate a computational brain template, which enables the generation of a template that has high image quality and accounts for variability in the species. The quality of the template is further ensured with the use of DTI-TK, a well-tested and high-performance DTI spatial normalization method in human studies. We demonstrated its efficacy in monkey studies for the first time by comparing it to other commonly used scalar-methods for DTI normalization. It is anticipated that this template will play an important role in facilitating cross-site voxelwise DTI analyses in rhesus macaques. Such analyses are crucial in investigating the role of white matter structure in brain function, development, and other psychopathological disorders for which there are well-validated non-human primate models.
The Rhesus Macaque (Macaca mulatta) is one of the most widely studied non-human primate species. This species has strong similarities to humans in terms of physiology and anatomy. It is the second non-human primate species to have its full genome sequenced (Gibbs et al., 2007), making it attractive for genetic studies. The rhesus has been widely used to study neuroanatomy, neurobiology, neurological and psychiatric illnesses, behavior and social interactions. In particular, the well-developed prefrontal cortex of the rhesus monkey makes it a unique model to study the brain systems implicated in psychopathology (Kalin and Shelton, 2003; Nelson and Winslow, 2009). Moreover, humans and rhesus monkeys share a prolonged period of brain development during childhood and adolescence, which is a critical period for the study of developmental psychopathology. For these reasons, the rhesus monkey is an ideal candidate for studies of brain and behavior relationships across development. Accordingly, noninvasive measurements of the rhesus monkey brain using magnetic resonance imaging (MRI) have been used for nearly 25 years (Chiro et al., 1985). Monkey brain atlases based upon post-mortem histological slices have been developed and made available to the research community (Martin and Bowden, 1996; Paxinos et al., 2009). Only recently has a computational rhesus brain template based upon T1-weighted structural MRI scans been described and made available (McLaren et al., 2009). However, T1-weighted templates alone do not provide accurate localization of white matter regions.
Diffusion tensor imaging (DTI) (Basser et al., 1994) is an exquisitely sensitive method to non-invasively map and characterize the microstructural properties and macroscopic organization of brain white matter (WM) tissues (Jones et al., 1999; Mori et al., 2002). Measurements of water diffusion are highly sensitive to and modulated by changes in brain tissue microstructural elements such as myelin, cellular density, cellular membranes, the cytoskeleton and mitochondria. In WM, which consists of packed axon fibers, the diffusion of water is anisotropic i.e. directionally dependent because the movement of water molecules perpendicular to the axon fibers is more hindered than in the parallel direction. The diffusion tensor is a positive-definite, rank-2 tensor matrix that describes the covariance of water diffusion in the three orthogonal Cartesian directions. From the diffusion tensor one can obtain maps of the diffusion tensor trace, eigenvalues, anisotropy and orientation (Basser and Pierpaoli, 1996). The fractional anisotropy (FA), the most commonly used measure of diffusion anisotropy, is a normalized standard deviation of the eigenvalues that ranges between 0 and 1. The higher the value the more organized (in a primary direction) the WM in that voxel. The trace of the diffusion tensor (TR) is the sum of the three eigenvalues and indicates the total amount of diffusivity in that voxel. It is inversely related to the microstructural density. The axial diffusivity (also known as longitudinal or parallel diffusivity) is the first eigenvalue of the tensor. This measure indicates the amount of diffusion in the primary direction of the WM organization. The radial diffusivity (also known as the transverse or perpendicular diffusivity) is the mean of the second and third eigenvalues. The diffusivities are indicative of the restrictedness of the tissue. These DTI measures have been used to characterize differences in brain microstructure for a broad spectrum of disease processes (e.g., demyelination, edema, inflammation, neoplasia), injury, disorders, brain development and aging, and response to therapy (see Alexander et al., 2007 for a review). The directional diffusion information has been used to estimate and reconstruct the trajectories of WM fiber bundles using tractography (Conturo et al., 1999; Mori et al., 1999; Hofer et al., 2008).
The application of DTI in rhesus monkey research is rapidly growing. DTI has been used in rhesus populations to study brain changes associated with aging (Makris et al., 2007; Bendlin et al., 2010), lesions (Shamy et al., 2010) and neuro-degeneration (Guo et al., 2011). Tractography methods have also been used to reconstruct estimates of white matter pathways (Schmahmann et al., 2007; Hofer et al., 2008; Liu et al., 2009). While several of these DTI studies in monkeys used voxel-based methods (Makris et al., 2007; Bendlin et al., 2010), to our best knowledge, DTI templates for rhesus monkeys are not publicly available. The creation of a standard space DTI template would facilitate the comparison and integration of research studies (Salimi-Khorshidia et al., 2009). To date, most published DTI studies in monkeys with voxel-based analysis have used intensity-based normalization with low to moderate dimensional registration methods (Guo et al., 2011; Makris et al., 2007; Bendlin et al., 2010; Willette et al., 2010; Shamy et al., 2010).
Voxel-based DTI studies in monkeys would benefit from the development of standardized DTI templates similar to human DTI templates (Mori et al., 2008; Zhang et al., 2011). A standardized DTI template may allow for use of tensor-based spatial normalization of the study subjects, which can improve the detection of WM differences in studying WM related pathology (Zhang et al., 2007b; Wang et al. 2011; Van Hecke et al. 2011). Tensor-based spatial normalization benefits from the rich information and heterogeneity of the DTI measurements with anatomy; however, the process of registering full tensor images is technically more complex than with scalar images. Full tensor spatial normalization requires reorientation of the diffusion tensor with the local deformation, while preserving the tensor shapes (Alexander et al., 2001). It has also been shown in human studies that the full diffusion tensor yields the most consistent spatial normalization across subjects (Park et al., 2003).
The purpose of this work was to develop, evaluate and optimize an approach for generating a population-averaged DTI brain template in nonhuman primates. These techniques were used to construct a high quality DTI brain template from 271 young Rhesus Macaques. We also present experimental evidence for using a high-dimensional, tensor-based, image registration technique in the population specific DTI template generation for non-human primates. Such a template can represent an unbiased mean of the large sample to describe the average structural shape, organization and diffusion properties of white matter in young rhesus population (Joshi et al. 2004; Zhang et al. 2007a). The DTI brain template may be used to facilitate whole brain (Smith et al., 2006) and tract-specific analyses (Yushkevich et al., 2008; Zhang et al., 2010). The development of a representative DTI template is also important for the generation of a detailed brain atlas. The template and tractography based reconstructions of five representative WM pathways are made publicly available at http://www.nitrc.org/projects/rmdtitemplate/.
The remainder of the paper is organized as follows: In section 2 we describe the image acquisition, pre-processing, and framework for DTI template construction using DTI-TK. Evaluation measures for comparison of normalization strategies are also described to demonstrate the efficacy of DTI-TK for non-human primates. Section 3 describes results of the comparison of normalization strategies and the details of the final DTI template. In section 4 the potential applications and limitations of our template are discussed.
The data were acquired as part of a unique brain imaging genetics study of young nonhuman primates. 271 young rhesus macaques in the age-range of 0.7370 to 4.2027 years with mean age of 2.4011±0.8795 years were scanned. All studies were performed using protocols approved by the University of Wisconsin Institutional Animal Care and Use Committee (IACUC). Before undergoing MRI acquisition, the monkeys were anesthetized with an intramuscular injection of ketamine (15 mg/kg). MRI scanning was performed using a GE SIGNA 3T scanner with a 16 cm diameter quadrature birdcage coil and the head was fixed in the sphinx position using a custom stereotaxic frame that fit inside the coil. DTI scanning was performed using a two-dimensional, echo-planar, diffusion-weighted, spin-echo sequence with the following parameters: repetition time=10s; echo time=77.2 ms; field of view=14 cm; matrix=128×128 (interpolated to 256×256 on the scanner); 2.5 mm thick contiguous slices; echo-planar echo spacing=800 μs. Diffusion-weighted imaging (b=1000s/mm2) was performed in 12 non-collinear directions with one non-diffusion weighted image and the acquisition was repeated six times and averaged. Images were acquired in the coronal plane through the entire monkey brain. A co-planar field map was also obtained using a gradient echo with images at two echo times: TE1=7 ms, TE2=10 ms.
The DWI volumes were eddy-current corrected using FSL (Smith et al., 2004). Echo-planar distortions were corrected using a field mapping procedure (Jezzard et al., 2005), which was based upon the fugue and prelude functions in FSL before performing a non-linear tensor estimation (Alexander and Barker, 2005). The brain tissue was carefully extracted for each subject using a rigorous semi-automatic protocol. We first evaluated two popular brain extraction methods (AFNI’s 3DSkullStrip, FSL’s BET) on a randomly selected subset of 40 volumes (Smith et al., 2002; Smith et al., 2004). 3DSkullStrip performed better according to dice similarity coefficient (Bartosic et al., 2010) and hence it was used for all the 271 subjects. The final masks were then carefully checked and manually adjusted.
A primary challenge in generating a population specific template is spatial normalization, which involves registering (aligning) each individual subject’s image to the others. Spatial normalization of groups of DTI data acquired from different subjects enables accurate mapping of characteristics of the diffusion tensor, such as the diffusivities, anisotropy and orientation within these images.
In this study, the template is constructed using DTI-TK, an advanced DTI spatial normalization and atlas construction tool (http://www.nitrc.org/projects/dtitk). It constructs a template in an unbiased way that captures both the average diffusion features (e.g. diffusivities and FA) and anatomical shape features (tract size) in the population (Zhang et al., 2007a). It performs white matter alignment using a non-parametric, highly deformable, diffeomorphic (topology preserving) registration method (Zhang et al., 2007b) that incrementally estimates its displacement field using a tensor-based registration formulation (Zhang et al., 2006). The tensor-based formulation perform alignment of white matter by taking advantage of similarity measures comparing whole tensors via explicit optimization of tensor reorientation (Alexander et al, 2001, Alexander and Gee, 2000a, Alexander and Gee, 2000b). By computing image similarity on the basis of full tensor images rather than scalar features, the algorithm incorporates local fiber orientations as features to drive the alignment of individual WM tracts. Using full-tensor information as a similarity metric for non-linear warping has been shown to be effective in spatially normalizing tract morphology and tensor orientation (Park et al., 2003, Zhang et al. 2007b).
These combined steps are used to generate a representative rhesus monkey DTI brain template, which can facilitate accurate comparisons of white matter microstructure and brain connectivity across populations. The initial average template is computed as a Log-Euclidean mean of the input DT images (Arsigny et al., 2006). The Log-Euclidean tensor averaging preserves white matter orientation with minimal blurring. The average DTI template is then iteratively optimized by registering the subject images to the current average, and computing a refined average as the mean of the normalized images for the next iteration. This procedure is repeated until the average image converges. The resulting template is unbiased towards any single subject and captures the average diffusion properties of the population at each voxel with a diffusion tensor. Subsequently, the template is “shape-corrected” to ensure that it also represents the average shape of the population (Guimond et al., 2000). This is achieved by first computing an average of the deformation fields that warp each subject into alignment with the template, then warping the template with the inverse of the average deformation field. The overall schematic of the processing pipeline is shown in Fig. 1. Since the brain size of NHP is significantly smaller than that of humans, the scaling parameters in the optimization were adjusted to match the size of the macaque brain. The length scale for humans in computing the piecewise affine deformation field is 1.0 mm while that for monkeys is 0.4 mm. Our DTI template, UWRMAC-DTI271, was generated using the set of DTI data from n=271 monkeys.
The quality of a template is dependent on the anatomical consistency of the normalized images used to generate the template which in turn depends on efficacy of the registration method used. DTI-TK has previously generated very high quality DTI templates in human studies, both for adults (Zhang et al., 2007b; Zhang et al., 2011) and infants (Wang et al., 2011). But its efficacy was not examined in the context of NHP. Hence we demonstrate the effectiveness of DTI-TK, a high-dimensional tensor-based normalization method in comparison to three other commonly used intensity-based normalization strategies for monkey data. We believe such comparisons for non-human primate data can serve a valuable purpose for DTI studies in monkeys. The three other approaches use FA, co-registered FA and T1-weighted images, and co-registered b=0 (T2-weighted) and T1-weighted images from the DTI study. We will refer to as (1) FA-ANTS, (2) FA-T1 and (3) B0-T1, respectively. All the three intensity-based registration methods use a publicly available state-of-the-art intensity-based normalization software suite called Advanced Normalization Tools (ANTS) (Avants et al., 2008, Avants et al., 2006, Avants et al., 2004), which uses diffeomorphic transformations similar to DTI-TK. For both DTI-TK and ANTS we use the default setup of parameters for the diffeomorphic transformations recommened in the online documentation. For ANTS we use the recommended symmetric diffeomorphic mapping (SyN).
FA-ANTS refers to FA-based normalization using ANTS. In this method the individual FA maps are first aligned to the FA of the template (obtained from DTI-TK). ANTS has been demonstrated to be among the most accurate intensity-based normalization method among fourteen different methods (Klein et al., 2009). The estimated transformations are then applied to the individual DTI data. Finally, tensors are reoriented to obtain the tensors in the template space.
FA-T1 and B0-T1 are both T1-based normalization approaches but use different intra-subject registration techniques to map the diffusion data onto T1 space (i.e. FA to T1, B0 to T1). First, the individual FA or B0 maps are aligned to their corresponding T1-weighted images using affine registration. The T1-weighted images are then aligned to a population specific T1 template (Oler et al., 2010) again using ANTS. These transformations are then applied to the individual DTI data by reorienting the tensors as in FA-ANTS.
Although the tensors were reoriented in all the above approaches using an identical state-of-the-art preservation of principle directions (PPD) scheme (Alexander et al., 2001), the key difference between those approaches and DTI-TK is that the tensor reorientation also plays an integral part in estimating the transformations in DTI-TK (Zhang et al., 2006), using a computationally efficient “finite-strain” strategy (Alexander et al., 2001).
The efficacy of a DTI registration method may be measured by the accuracy of the WM alignment. The four registration methods were compared using several metrics (Zhang et al., 2007b, Zhang et al., 2011) that are described below. The brain alignment performance was first evaluated using a subset of n=30 monkeys (called DTI30) for higher computational efficiency relative to the full sample. The following metrics are used to investigate the spatial variance and consistency of the scalar, directional and entire matrix information in the diffusion tensor.
The diffusion anisotropy and the dominant direction of diffusion are two features that account for much of the variations in WM structures (Pierpaoli et al., 1996). Misalignment of WM structures yields large voxelwise variations in either one or both of these features. Therefore, the normalized standard deviation of the FA and the dyadic coherence (Jones et al., 2002) were selected as voxelwise statistics to gauge the normalization quality. The normalized standard deviation of FA, FA, is defined as the ratio of the standard deviation over the mean of the FA values. The Dyadic coherence, κ, which ranges from 0 to 1 (0 for randomly oriented directions and 1 for identically oriented directions), captures the variability in the dominant direction of diffusion. We calculated FA and κ at each voxel within the WM regions of the DTI30 templates for all the four methods. We also computed normalized standard deviation of the tensor trace (TR), similar to FA.
Cross correlation between two scalar maps X1 and X2 is computed as
where υ indexes over all the voxels. This value ranges between 0 and 1. The higher the cross correlation, the higher the similarity between two maps. We computed the cross correlations of the WM voxels between the DTI30 subjects and the corresponding template for each of the four normalization methods using both fractional anisotropy (FA) and trace (TR) of the tensors.
Eigenvalues and eigenvectors contain complementary information, which together specify the size, shape, and orientation of the diffusion tensor ellipsoid that characterizes diffusion (Basser and Pajevic, 2000). The OVL between two tensors is computed as:
where λ, ε and λ′, ε′ are the eigenvalue-eigenvector pairs of the two tensors respectively. A higher average OVL indicates that the method generates more consistently normalized subjects. The OVL of the WM voxels between DTI30 subjects and the corresponding templates for each of the four normalization methods was computed.
The two distances between diffusion tensors (Zhang et al., 2006) are (a) the Euclidean distance (ED) defined as:
and (b) the deviatoric distance (DD) which is the Euclidean distance between the corresponding deviatoric tensors that takes into account the anisotropic component of the tensors defined as:
Lower tensor distances suggest better accuracy in WM alignment. We computed the above two distances between WM voxels of the DTI30 subjects and the corresponding templates of the four normalization methods.
Statistical comparisons of the four methods were performed with the same white matter mask for all the registration methods. DTI-TK and FA-ANTS already share a same white matter mask. The template FAs of FA-T1, B0-T1 are mapped on to the DTI-TK space, using FSL’s linear image registration tool (FLIRT) (Smith et al., 2004). This removes the size of the white matter mask as a confounding factor in the performance comparison. The set of white matter voxels was defined as those having FA >= 0.2. Histograms (distributions) for all four normalization methods were then generated for each metric. Many of these histograms are skewed and thus violate normality. So paired t-tests would be inappropriate. Hence the distributions were evaluated using the respective empirical cumulative distribution functions (CDF). A CDF is the probability that a variable has a value equal to or higher than a specific value. For each metric we perform Kolmogorov-Smirnov (KS) two-sample tests (Massey, 1951) to estimate statistical significance of the differences between the CDFs. The KS statistic measures the largest possible distance between two CDFs.
Qualitative comparisons of the FA of the templates using DTI30 for all of the spatial normalization methods are presented in Fig. 2. Visually, it is evident that the full-tensor based (DTI-TK) and FA based (FA-ANTS) templates are the sharpest, while the FA-T1 and B0-T1 templates are much more blurry. Compared to FA-ANTS, some of the finer WM structural details, such as the separation between the internal and external capsule, are better delineated using DTI-TK (see arrows in Fig. 2). In all the quantitative evaluation figures DTI-TK is represented using blue, FA-ANTS using red, FA-T1 using green and B0-T1 using cyan colors.
Good spatial normalization will ideally maximize the anatomical consistency in the imaging measures, which decreases measurement variance, while preserving individual differences. The empirical CDFs and histograms (insets) of the above variables for all four methods are shown in Fig. 3. DTI-TK has the smallest standard deviation in FA and the highest eigenvector coherence. According to this criteria, B0-T1 and FA-T1 perform poorly, as indicated by the right shift in FA and left shift in κ, compared to DTI-TK and FA-ANTS. Although FA-ANTS is performs similarly in terms of FA, DTI-TK yields more consistent tensor orientations as demonstrated by higher κ of the tensors.
The spatial maps of FA, TR and κ are shown in Figs. 4 and and5,5, which visually demonstrate reduced FA and TR and increased κ for the DTI-TK and FA-ANTS. TR is significantly lower than that of FA. This is expected since the contrast between white and gray matter is quite less in TR maps. As can be seen both qualitatively and quantitatively, DTI-TK yielded the lowest inter-subject variance in the scalar measures and higher eigenvector coherence, which we believe is desirable in a DTI spatial normalization method.
The empirical CDFs and histograms (insets) of the cross-correlations for FA, TR and eigenvalue-eigenvector overlap (OVL) for the four methods are shown in Fig. 6. The plots DTI-TK are shifted furthest to the right, though FA-ANTS has similar performance. These results suggest that DTI-TK yields more consistent tensor information at the voxel level.
The histograms and CDFs of the tensor distance metrics (ED and DD) are shown in Fig. 7. The plots of DTI-TK are significantly shifted to the left. These results support that DTI-TK yields higher between-subject consistency of the full diffusion tensors.
In this section example maps of different measures extracted from UWRMAC-DTI271, the template generated from all 271 animals using DTI-TK, are presented. Maps of the average fractional anisotropy, trace, axial diffusivity and radial diffusivity of are shown in Fig. 8. Axial slices (every 5th slice) of our DTI template are shown in Fig. 9 as RGB-encoded fiber orientation maps (Pajevic and Pierpaoli, 1999). The intensities of the colors are modulated by the corresponding FA map. The bottom row of Fig. 9 shows a subset of slices (corresponding to second row) from a registered T1 template (McLaren et al., 2009). The UWRMAC-DTI271 was aligned to the Saleem Logothetis (SL) (McLaren et al., 2009) and Paxinos (Paxinos et al., 2009) space by affine registration (as used in the FA-T1) of our template FA to the corresponding T1 weighted templates. This registration approach (FA-T1), as can be seen, from the evaluation experiments was the next best performing method, in the absence of diffusion tensor data. The tensors in the corresponding T1 spaces were resampled and reoriented using the Log-Euclidean and PPD framework respectively using the implementations in DTI-TK.
Tractography on the DTI template can be used to estimate the trajectories of white matter pathways (Jones et al., 2002, Park et al., 2003) in the macaque brain. To demonstrate the potential of tractography applications, the UWRMAC-DTI271 was used to reconstruct five major white matter pathways – the corpus callosum, the inferior fronto-occipital fasciculus, the cingulum bundles, the fornix and the uncinate fasciculus. Tracts were generated and selected based on strategies described by Catani et al. (2008) and Mori et al. (2002). These major WM pathways share similar topology with those in humans. Several additional 3D renderings of these tracts are available online as supplementary material.
Finally Fig. 11 shows corresponding axial slices of FA maps of six different subjects and the corresponding slices from the template. It can be seen that the major white matter structures are consistently aligned. Additional movies showing the accurate alignment for an axial and a coronal slice are available as supplementary material. Such consistent alignment is highly desirable for voxel-based analyses.
In this paper we described the construction of a population-specific DTI template, the UWRMAC-DTI271, for young rhesus macaques. The DTI template was resampled and aligned using affine registration of our template FA to the corresponding T1 weighted templates in both the Paxinos atlas (Paxinos et al., 2009), (241×320×190, 0.25×0.25×0.25 mm3) and the Saleem and Logothetis atlas (McLaren et al., 2009) (256×256×240, 0.5×0.5×0.5 mm3) and is currently available as a free, image-based resource in the standard image NIFTI (.nii) format for researchers worldwide. We believe that it will be of significant interest to investigators in the fields of brain imaging, systems neuroscience, developmental neurobiology, comparative neurobiology, primatology and psychiatry. This template, along with the WM pathways, may be downloaded at http://www.nitrc.org/projects/rmdtitemplate/. It may be loaded and viewed with many publicly available image analyses software packages such as FSL (http://www.fmrib.ox.ac.uk/fsl/), ITK-SNAP (http://www.itksnap.org/pmwiki/pmwiki.php) and AFNI (http://afni.nimh.nih.gov/afni). The WM tracts may be loaded and viewed in TrackVis (http://www.trackvis.org/). The contributing factors to the high image quality of our template are (1) the extremely large sample from which the template was created and (2) the high performance of the spatial normalization across animals.
Consistent with recent DTI template creation studies in humans (Zhang et al., 2007b; Zhang et al., 2011; Wang et al., 2011), this study showed that DTI-TK generated a DTI template of the rhesus macaques with the least inter-subject variability and high anatomical consistency. Although evaluating all possible registration methods is beyond the scope of the current paper, we can conclude that the tensor-based registration with DTI-TK yielded more consistent mapping of the scalar DTI measures, and the tensor orientation (e.g., eigenvectors, full tensor) in particular, compared to intensity based methods. The latter is not too surprising as previous works such as Park et. al. 2003 demonstrated that under common transformation model, using full tensor information performs better than using rotation invariant indices of the tensor. Hence although the improvement of DTI-TK may be due to different implementation of the diffeomorphic transformations the primary reason may be attributed to the use of tensor-based similarity. We also note that FA-ANTS had similar performance for registration of FA (and TR) maps, and that this approach would likely generate similar results when evaluating just scalar DTI measures. The apparently worse performance of the T1 normalization approaches (FA-T1 and B0-T1) is likely caused by poor registration of the DTI to the T1-weighted images, but might also be caused by overall worse performance of normalization using T1-weighted images. We believe that the presented comparisons can help primatologists make informed decisions when choosing a registration method and provide useful guidance in spatial normalization of monkey DTI data.
Although UWMRAC-DTI271 is obtained from a very large number of subjects using a well-tested normalization method there are several limitations that need to be considered. First, the spatial resolution of the original DTI acquisition was not isotropic and was relatively coarse in the slice (anterior/posterior) dimension (2.5 mm). However, both the template and the spatially normalized DTI data of each individual animal (Fig. 8) appear much sharper than the original data, including the slice dimension. Clearly the DTI template benefits from averaging across such a large number of images, resulting in minimal apparent blurring. Second, the field-map based distortion correction prior to spatial normalization (see Methods section for details) cannot completely compensate for highly distorted anatomical areas in the ventral prefrontal areas and temporal lobe regions around the inner ear. This is a limitation with nearly all DTI studies in both monkeys and humans. In the future, an improved DTI template may be constructed with true isotropic resolution and parallel imaging, using the spatial normalization strategies described here. Third, the DTI template was generated using DTI data from monkeys over an age range of 0.7370 to 4.2027 years, which may be suboptimal for use in studies of animals at ages outside the range. Future work is needed to generate either DTI templates at other ages or a DTI template with a temporal dimension to accommodate for age-related changes (Hart et al., 2010; Fonov et al. 2011). Finally, it is well known that the diffusion tensor model has some limitations with regard to describing both the diffusion and geometric properties of the tissue in regions of fiber crossing (Wedeen et al., 2008). Despite this limitation, there is a huge potential for DTI, and it is widely used for most clinical and scientific imaging research studies of white matter.
We would also like to note that, although population averaged computation templates represent anatomical variation in the species with high SNR and low bias (McLaren et al., 2009, Joshi et al., 2004, Zhang et al., 2007a), single-subject post-mortem templates (Martin and Bowden, 1996; Paxinos et al., 2009) are very valuable in cases where DTI/MRI cannot match the histological specificity and precision. Single-subject computational atlases with high SNR have also been developed by averaging repeated acquisitions of a single subject for accurate segmentation and atlas based analyses (Mazziotta et al. 2001; Oishi et al. 2011). Some voxel based analyses (like TBSS), although recommend using population-specific template, provide an option to choose “most representative” single-subject as a template that is then transformed to MNI152 (Smith et al. 2006). Finally, there are more recent Bayesian and multi-atlas approaches that may ultimately result in better representation of population properties (Ma et al. 2008; Natasha et al. 2009; Wu et al. 2011; Koikkalainen et al. 2011).
There are many potential avenues to apply the UWRMAC-DTI271. First, it is an average representation of brain neuroanatomy and diffusion measurements in the monkey, which will be valuable for education, atlasing, anatomic localization, and comparative neuroanatomy (Stephan et al., 2001; Ramnani et al., 2006). Future work will include detailed anatomic labeling of the template similar to the human DTI atlas created by Mori et al. (2008). The UWRMAC-DTI271 template may be useful as a reference template for spatial normalization of other DTI data sets, with matching age-range, for voxel-based analyses. Depending on the age-range and anatomical status of the sample, some monkey DTI studies may still need to use a study-specific template. In such cases, researchers can build upon the thorough evaluations and registration framework presented in this paper and can subsequently map the results onto this template for standardized reporting. By mapping regions-of-interest (ROIs) and/or tract reconstructions onto this template, automated region- and structure- specific analyses may be efficiently performed. Further, since the template is in standardized spaces (Paxinos et al., 2009; McLaren et al., 2009), it is now possible to integrate other brain templates and regional atlases to create multimodal maps of the monkey brain.
Another unique attribute of the UWRMAC-DTI271 template is that the major directional components of the diffusion tensor orientations are preserved. This facilitates the application of tractography in the average template as can be seen by the example reconstructions of five major white matter pathways shown in Fig. 10. We will use this template to reconstruct more of the fiber pathways and create a more complete tractography atlas of the rhesus monkey brain. Importantly, identifying WM pathways may facilitate efficient tract specific analyses (TSA) in nonhuman primates (Yushkevich et al., 2008; Zhang et al., 2010). One clear advantage of DTI studies in the monkey is that tract reconstructions may be compared with tracer studies that are not possible in humans. Spatial standardization will also enable the mapping and comparison of prior tract tracing studies in the CoCoMac database (Croxson et al., 2005), which describes cortical connections in the rhesus macaque. Our template, in conjunction with a reference label atlas (Wisco et al., 2008), can be used for developing in vivo brain connectome models for the rhesus monkey using tractography (Hagmann et al., 2010). This will be a critical step in understanding the influence of connectivity of the primate brain.
In addition to its application in the study of development per se, we believe that the developing rhesus monkey DTI brain template presented here is an important step in furthering the study of psychopathology. In particular, studies demonstrate that several forms of psychopathology often begin during childhood or adolescence (Paus et al., 2008). Moreover, evidence suggests that many adolescent and adult anxiety and affective disorders are chronic developmental illnesses that have their genesis in early childhood (Pine et al., 1998). Thus, early childhood and adolescence are a period of increased risk, and since anxiety disorders are among the most common forms of psychopathology (Kessler et al., 2005), a focus on developmental non-human primate models of brain white matter will be increasingly important to understand the neural bases of anxiety-related psychopathology (Kalin and Shelton, 2003; Nelson and Winslow, 2009).
This work was partially supported by NIH grants MH62015, MH084051, MH080826, MH46729, MH81884, MH018931, R03-EB009321 and University of Wisconsin-CIBM-MIR fellowship.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.