The MRI scans were processed in a pipeline that required several steps including a) skull stripping using FSL (Smith et al., 2004 ), diffeomorphic registration to a symmetric population-specific template chimpanzee brain (Avants et al., 2011b (link)) and subsequent probabilistic segmentation of the T1-weighted images into grey matter, white matter and CSF following procedures and software that have been described in detail elsewhere (Avants et al., 2008 (link); Avants et al., 2010a (link)).
After brain extraction, we used the openly available Advanced Normalization Tools (ANTS) (http://www.picsl.upenn.edu/ANTS/) to perform multivariate normalization and structure-specific processing of our data (Avants et al., 2008 (link); Klein et al., 2009 ). ANTs encodes current best practices in image registration, optimal template construction and segmentation and is scalable to large-scale, distributed computing environments. ANTs cross-sectional studies deform each individual dataset into a standard local template space and/or a canonical stereotactic coordinate system. The core processing maps T1 structural MRI to an optimal template space, which is defined as the population-specific, unbiased average shape and appearance image derived from a representative population (Avants and Gee, 2004 ; Avants et al., 2010b (link)). The average template was constructed to optimally represent both the original and a flipped version of the dataset such that the final template is symmetric about the midsagittal plane. The coordinate deformations themselves are smooth and invertible, that is, diffeomorphic – neuroanatomical neighbors remain neighbors under the mapping. At the same time, the algorithms used to create these deformations are biased towards the reference space chosen to compute the mappings. Moreover, these topology-preserving maps capture the large deformation necessary to aggregate populations of images in a common space. Recent evaluation studies suggest that ANTs-based normalization is currently the most stable and reliable method available. After defining the template image to target image coordinate transformation, we employ template-based priors and N4 inhomogeneity filed correction to accurately segment cortical gray matter segmentation and perform cortical parcellation (Das et al., 2009 (link)). After these initial processing steps, the diffeomorphic registration-based cortical thickness (DiReCT) method is used to compute cortical thickness (Das et al., 2009 (link)).
DiReCT uses the segmentation probability images to compute a continuous voxel-wise estimate of cortical thickness (Das et al., 2009 (link)). DiReCT is unique in that it exploits tissue segmentation probability maps to identify a maximum likelihood correspondence between the white matter surface and the outer gray matter surface where the correspondence mapping is constrained to be spatially regular, differentiable and invertible, i.e. diffeomorphic. As a consequence, DiReCT thickness estimates incorporate both shape constraints and subtle probabilistic information about the likely position of sulci that may not be visible in a hard segmentation. Thus, DiReCT is a robust image-based technique for identifying voxel-wise and regional thickness information. An independent implementation and validation of this method showed that it is competitive with Freesurfer, a commonly used program for estimating cortical thickness (Clarkson et al., 2011 (link)). We note, however, that the implementation used by Clarkson et al. (2011) (link) was never itself directly compared to that available in the ANTs toolkit and we suspect that differences between the Clarkson et al implementation and our own gold standard implementation may exist.
As has been done in some studies of cortical thickness in humans and monkeys (Styner et al., 2007 ; Van Essen et al., in press ), we used a seed or region-of-interest approach to quantify GM thickness in 12 select areas of interest (see Table 1) including the a) dorsal, mesial, and orbital prefrontal cortex b) superior, middle and inferior temporal gyri c) pre- and post-central gyri d) inferior frontal gyrus e) posterior superior temporal gyrus f) supramarginal gyrus and g) superior parietal lobe. The landmarks used to define each region of interest are provided below and they were selected for theoretical and pragmatic reasons. Specifically, there has been considerable comparative interest in the evolution of the prefrontal, temporal and parietal cortex (Deacon, 1997 ; Schoenemann et al., 2005 (link); Rilling, 2006 ; Sherwood et al., 2012 ) and therefore we focused on these regions. Similarly, there has been significant interest in the evolution of cortical organization and lateralization in the homologs to Broca’s and Wernicke’s area in chimpanzees (Cantalupo and Hopkins, 2001 (link); Keller et al., 2009 (link); Hopkins and Nir, 2010 (link); Schenker et al., 2010 (link); Spocter et al., 2010 ) and this was our reasoning for quantifying the inferior frontal gyrus and posterior superior temporal gyri. Finally, we also included the pre- and post-central gyri as regions of interest because they are primary motor and sensory cortex and we hypothesized that cortical thickness would be lower in these regions compared to the others. The region of interest masks were drawn on the chimpanzee template brain using a mouse-controlled pointer and then transformed back to the individual GM thickness maps for each subject, using the inverse matrix of the original registration. The masks were then applied to the individual subjects’ GM thickness map to derive average thickness measures for each of the 12 regions within each hemisphere. We also computed an average cortical thickness measure for each hemisphere. This was done by drawing a mask that covered the entire left or right hemisphere, excluding the brain stem and cerebellum. We note that, because the template is symmetric, the masks only had to be drawn on one hemisphere of the template brain and could then be transformed to label both left and right sides of the individual subject’s brain.