The average baboon brain: MRI templates and tissue probability maps from 89 individuals

Highlights

• Unbiased population-average baboon brain MRI templates are provided.

• Tissue probability maps in template space are also provided.

• Images are useful for normalization and segmentation of baboon neuroimaging data.

• Provides a Talairach-like standardized anatomical coordinate space for the baboon

Abstract

The baboon (Papio) brain is a remarkable model for investigating the brain. The current work aimed at creating a population-average baboon (Papio anubis) brain template and its left/right hemisphere symmetric version from a large sample of T1-weighted magnetic resonance images collected from 89 individuals. Averaging the prior probability maps output during the segmentation of each individual also produced the first baboon brain tissue probability maps for gray matter, white matter and cerebrospinal fluid. The templates and the tissue probability maps were created using state-of-the-art, freely available software tools and are being made freely and publicly available: http://www.nitrc.org/projects/haiko89/ or http://lpc.univ-amu.fr/spip.php?article589. It is hoped that these images will aid neuroimaging research of the baboon by, for example, providing a modern, high quality normalization target and accompanying standardized coordinate system as well as probabilistic priors that can be used during tissue segmentation.

Introduction

Given the phylogenetic proximity between humans, apes and monkeys, research on non-human primate models is an essential component to understanding both healthy brain function and disease (Belmonte et al., 2015, Roelfsema and Treue, 2014). After the anthropoid apes, Old World monkeys are the next closest relatives to humans (Stewart and Disotell, 1998). As well as being an excellent natural model for epilepsy (Killam, 1979, Szabo et al., 2011a, Szabo et al., 2011b) baboons (Papio), an Old World monkey, possess several features that make them a particularly fruitful model for understanding human brain structure and function (Black et al., 2009). A baboon brain, for example, is on average two times larger than the brain of a rhesus macaque (Macaca mulatta, Leigh, 2004) — one of the most common Old World monkeys found in laboratories. The baboon brain also has a larger degree of gyrification (folding) than other Old World monkeys and contains all the primary cortical structures found in humans (Rogers et al., 2010). Accordingly, the baboon model has been used in numerous structural and functional neuroimaging experiments (e.g., Kochunov et al., 2010a, Kochunov et al., 2010b, Kroenke et al., 2005, Kroenke et al., 2007, Liu et al., 2008, Miller et al., 2013, Phillips and Kochunov, 2011, Phillips et al., 2012, Rogers et al., 2007, Salinas et al., 2011, Szabo et al., 2007, Szabo et al., 2011a, Szabo et al., 2011b, Wey et al., 2013).

The application of multi-subject statistics to such neuroimaging experiments requires, in general, that images acquired in a participant's native-space be normalized to the standardized coordinate space of a template image. This process ensures, as accurately as the normalization algorithm allows, that the same coordinates within the space represent corresponding anatomical regions from each individual. This now standard practice increases statistical power, enables conclusions to be generalized to the population as a whole and facilitates the generalizability and comparability of results across studies.

Pioneering human neuroimaging work by Fox and colleagues (Fox et al., 1985) used an X-ray to objectively map PET activation sites to the Talairach space (Talairach et al., 1967). This technique was soon updated to use a T1-weighted (T1w) magnetic resonance imaging (MRI) scan instead of an X-ray (e.g., Evans et al., 1992, Seitz et al., 1990). Due to the limitations caused by Talairach space being based on a single 60-year old female, this development also led to the creation of a new population-average template and an associated standard space (MNI space, Evans et al., 1993). While the MNI space continues to be extensively used and further developed (Evans et al., 2012) for some applications it is arguably not the most appropriate target space, for instance, an age-appropriate template is generally to be preferred in pediatric studies (Fonov et al., 2011, Sanchez et al., 2012, Wilke et al., 2002, Yoon et al., 2009).

The importance of non-human primate neuroimaging research also necessitated the creation of appropriate brain templates (Black et al., 2001a, Black et al., 2001b). The first non-human primate population-average template was constructed from the T1w MRI images of nine olive baboons (Papio anubis, Black et al., 1997, Black et al., 2001b) and was mapped to the Davis and Huffman (1968) baboon brain atlas. Since, template images for several non-human primate species have ben made available: the pig-tailed macaque (Macaca nemestrina, Black et al., 2001a), the rhesus macaque (Macaca mulatta, Frey et al., 2011, McLaren et al., 2009, Rohlfing et al., 2012), the Japanese macaque (Macaca fuscata, Quallo et al., 2010), the cynomolgus monkey (Macaca fascicularis, purl.org/net/kbmd/cyno), the vervet monkey (Chlorocebus aethiops, Fedorov et al., 2011, Maldjian et al., 2014, Woods et al., 2011), the common marmoset (Callithrix jacchus, Hikishima et al., 2011, Newman et al., 2009) and the olive baboon (Papio anubis, Black et al., 2001b, Greer et al., 2002). Recently, tissue probability maps have also been provided with non-human primate brain templates (e.g., Fedorov et al., 2011, Hikishima et al., 2011, McLaren et al., 2009, Rohlfing et al., 2012).

The existing baboon (Papio Anubis) templates were made available almost 15 years ago (Black et al., 2001b, purl.org/net/kbmd/b2k; Greer et al., 2002, sites.google.com/site/baboonmriatlas). Several factors warrant the creation of a new, updated and improved brain template in this species. First, a baboon template would ideally be created from a large number of individuals — in the existing templates the maximum number is 9. The larger the number and consequently the more heterogeneous the sample of animals used to create a template the better it represents the variability in brain morphometry within a species. Second, there have been significant improvements in the techniques used to coregister T1w MRI images, which is a critical component of template creation (Klein et al., 2009). Third, to our knowledge there is no currently available symmetric template image for the baboon. For both human and non-human primate brains, it is well known that there are differences in morphology across the two cerebral hemispheres (e.g., Barrick et al., 2005, Pilcher et al., 2001). Symmetric templates are critical when investigating these hemispheric differences using, for example, voxel-based morphometry (VBM, Ashburner and Friston, 2000); normalization to an asymmetric template introduces a confound as it is impossible to conclude whether hemispheric differences are real or simply caused by normalization to an asymmetric template (Fonov et al., 2011, Pepe et al., 2014). Fourth, to our knowledge there are no tissue probability maps available to aid in the normalization or the segmentation of baboon brains. Finally, despite being an important and visibly protuberant brain region anteroinferior to the frontal lobe, the olfactory bulb is not present in the Greer et al. (2002) baboon template. Although present in the template of Black et al. (2001b), it is not encompassed by the brain mask provided by the authors.

For these reasons the current work aimed to produce both asymmetric and symmetric unbiased population-averaged T1w baboon (Papio anubis) brain templates and their associated gray matter, white matter and cerebrospinal fluid tissue probability maps primarily using the open source toolkit Advanced Normalization Tools (ANTs, http://stnava.github.io/ANTs/; Avants et al., 2011a). ANTs has been independently evaluated as containing one of the top performing registration algorithms (Klein et al., 2009, Ou et al., 2014) and employs current best practice in template creation (Avants et al., 2010, Hopkins and Avants, 2013). In the current article, we detail the data and methods used in the creation of these templates and associated tissue probability maps. The main template, Haiko89, inherited its name from the 100th baboon scanned in this project. Haiko was the oldest baboon of the social group and passed away naturally before the end of the MRI project. Created entirely using freely available software, the templates are being made available to the scientific community to facilitate, for example, normalization to a standard coordinate space, skull stripping and tissue segmentation of individual baboon brains.