These experiments were approved by the Animal Care Committee and Lab Animal Services at the Hospital for Sick Children. This study follows the Canadian Council on Animal Care (CCAC) standards.

Eight male Yorkshire piglets of 22 ± 6 days old (body weight: 6.1 ± 1.2 kg) were used in this study. All piglets were housed in pairs in a temperature- and humidity-controlled environment with a 12-h light/dark cycle and fed with a commercial piglet milk replacer prior to MRI scanning. The piglets were euthanized by intravenous injection of pentobarbital sodium (120 mg/kg) upon the completion of all imaging procedures and while under general anesthesia.

The animals were pre-anesthetized with a Ketamine (10 mg/kg) solution intramuscularly, intubated and maintained under anesthesia with 2.5% isoflurane and 2 L of oxygen delivered via MRI-compatible ventilator during the MRI scanning. Heart rate and oxygen saturation were monitored and a circulating water blanket was used to maintain core body temperature of 37°C. The animals were placed prone and head first on the diagnostic table. In the Philips Achieva 3T MR scanner, a 32-channel receive-only head coil was placed around the head to provide MR imaging. A three-dimensional T1-weighted magnetization prepared gradient echo (MPRAGE) sequence was used. The sequence parameters were: repetition time (TR) = 8.15 ms, echo time (TE) = 3.72 ms, flip angle = 8°, matrix = 224 × 224, field of view (FOV) = 224 mm × 224 mm, slice thickness = 1.00 mm.

The diffusion-weighted images were collected with a spin-echo single-shot echo-planar imaging (EPI) sequence using sensitivity encoding (SENSE; TR = 5844.97 ms; TE = 105.90 ms; flip angle = 90°; matrix = 128 × 128, FOV = 205 mm × 205 mm; slice thickness/spacing = 1.60/1.60 mm; SENSE reduction factor = 2) with a b-value of 800 s/mm² along 128 directions. A baseline image with b = 0 s/mm² (b0 image) was acquired for both forward and reverse phase encoding directions.

Specific ROI-based fiber tracts were extracted from the population-based whole-brain fiber tracts. The ROIs include genu and splenium of the CC, optic chiasm, fornix and middle cerebellar peduncle. 3D Slicer was used to extract the fibers passing through specific fibers.

Mean (±standard deviation) body weight, age and brain size of the piglets used to generate the templates were calculated in Matlab (version 2014b, MathWorks, Natick, MA, USA).

The average age of the eight piglets was 22 days (range: 17∼35 days old). The average brain size of the piglets was 65500 ± 3806 mm³, and the template brain size was 67294 mm³. For the six pigs used to generate the DWI related template, the average body weight, age, and brain size were 6.5 ± 1.2 kg, 23 ± 6 days old and 66434 ± 3927 mm³. The body weight, age and brain size of the eight piglets used to generate T1 template are shown in Table 1.

The brain pattern is well preserved in the final T1 template brain compared to that of a single subject T1 image (Figure 1, row 1). The final T1 template brain is shown in Figure 1 (row 2) in the axial, coronal and sagittal views. Qualitatively, the outline of brain structures and GM/WM contrast are clearer in the template than in the single subject. Based on the intensity differences, the T1 brain image could be segmented into GM, WM and CSF. Figure 1, rows 3–5 shows the probability maps of GM, WM and CSF in the template space.

The maximum distance was 1.73 mm for one subject at the posterior CC, while the mean distances across subjects were smaller than 1 mm for all three landmarks. The differences between coordinates in selected landmarks are shown in Table 2.

DWI metric templates, including FA, AD, RD, and MD maps, were generated (Figure 2). Figure 3 shows the population-based whole-brain fiber tracts transformed from individual space, with each color representing fibers from each individual. The top row shows the fibers from a single subject in the template space. In general, the whole brain fibers from different subjects were overlapped in the template space. Figure 4 shows the merged tracts overlaid with the DWI metrics on each fiber. Consistent with the DWI metric maps, FA values on the fiber tracts were higher in the regions with highly aligned structures, such as the CC, fornix, and trigeminal nerve. On the contrary, AD, RD and MD values were higher in the regions with more fluid, such as olfactory bulbs, ventricles and cisterns.

The ROI-specific tracts were extracted from the population-based whole-brain tracts, including tracts from the fornix, optic chiasm, middle cerebellar peduncle, genu and splenium of the CC (Figure 5). Column 1 and 3 represent fibers colored by the direction of the tensor and FA values, separately. Column 2 represents fibers colored by individual identities, showing that the tracts from restricted ROIs were consistently present across subjects. Specifically, the fornix could be visualized as the C-shaped fiber bundle from its body, approximately at the level of the anterior commissure, and reaching to the level of the crura. The crossing fibers of the optic chiasm were clearly visualized. The middle cerebellar peduncle fibers were visualized with projections deep into the cerebellum. Similarly, the genu of the CC projections to the bilateral dorsolateral prefrontal region were also seen, with fibers visualized distally and approaching the cortex. For the splenium, the majority of the tracts traveled along the external boundary of lateral ventricles, bilaterally to the temporal lobe (tapetum of the CC), while few tracts traveled to the visual cortex (forceps major).

This template images and tractography are available online⁴, and can be visualized in 3D Slicer.

Compared to the existing two piglet T1 templates, this T1 template includes higher GM/WM contrast retaining more detailed information than the one from Conrad et al. (2014) and is more comprehensive, including both the optic chiasm and olfactory bulb, relative to the one generated by Gan et al. (2014).

A standard coordinate system is important for brain research, where reported brain locations could be easily compared across studies and imaging modalities. In order to maintain consistency with other templates (Felix et al., 1999; Watanabe et al., 2001; Saikali et al., 2010; Conrad et al., 2014; Gan et al., 2014), we set the same origin at the anterior limit of the posterior commissure in the midsagittal plane. This eases the comparison of results from studies with different templates. The validation results presented show only a small deviation (≤1 voxel) between the template and each individual brains. Probability maps of segmentation are also presented here to serve as a priori input to aid future single image segmentation.

Diffusion templates introduced for humans (Mori et al., 2008; Verhoeven et al., 2010; Wang et al., 2011), primates (Adluru et al., 2012; Zakszewski et al., 2014) and rats (Rumple et al., 2013), have played an important role in subsequent ROI-delineation and brain registration (Zhang et al., 2010; Turken and Dronkers, 2011; Vetreno et al., 2015). To date there has been no DWI metric template introduced for the piglet brain, which limits any significant level of information about WM anatomy. In this study, the DWI metric template was generated based on six piglets, providing voxel-wise diffusion information. The different scalar maps could be used as a template for brain normalization. Volumetric-based morphometry analysis of a DWI metric map can also be accomplished with this template.

In this study, with the conjunction uses of HARDI and the method proposed by Tournier et al. (2007) to determine the FOD, orientations that are separated by small angles could be resolved. The fiber tract model thus could represent crossing fibers reliably, e.g., the optic chiasm (Figure 5). The population-based whole-brain fiber tracts were generated through transformation of individual streamlines into the template space. These tracts thus represent homologous white matter connections between individual animals. The tracts connecting the same regions are highly overlapped across individuals, reflecting consistency of tracts across subjects (Figures 3 and 5), and suggesting that the inter-subject template registrations are within an acceptable range. On the contrary, tracts that show less overlaps may indicate higher structural variability across subjects or lower consistency from tracts generation. Merged tracts are different from tracts generated based on an average diffusion template (Verhoeven et al., 2010; Adluru et al., 2012; Zakszewski et al., 2014), where individual subject diffusion orientation information would have been lost. Furthermore, as the population-based tracts preserve individual tracts information, researchers are able to merge tracts from their own data to the population-based tracts. With the augmented fiber tracts from more subjects, the degree of consistency of the population-wise fiber distribution may be visualized and inspected more confidently.

Specific ROI analysis shows that fiber morphology is analogous in piglets to humans in the fornix (Chen et al., 2015; Kehoe et al., 2015), optic chiasm (Hofer et al., 2010), the genu of the CC (Verde et al., 2014) and middle cerebellar peduncles (Chanraud et al., 2009). However, the shape of the fibers passing through the splenium of the CC is different in piglets and humans. The majority of the splenium fibers project to the visual cortex (called forceps major) while other fibers were part of the tapetum that extended laterally into the temporal lobe in adult human brains (Hofer et al., 2010; Miller et al., 2011; Turken and Dronkers, 2011) as well as neonates (Gilmore et al., 2007; De Bruïne et al., 2011). However, in the piglets, the majority of these fibers form the tapetum, with only a small portion of the tracts projecting to the visual cortex (Figure 5). The discrepancy is expected as many brain structural differences exist between the two species (Sauleau et al., 2009). For example, the telencephalon of the pig brain is less curved than that of the human brain (Sauleau et al., 2009).

The newborn human brain is approximately 36% the size of the adult human brain. With a fast growth speed, the TBV reaches ∼72% of adult brain size at 1 year, and ∼83% at 2 years (Knickmeyer et al., 2008). A similar developmental curve until the maximum is present in the pigs. It is suggested that the piglets reached its maximum brain growth rate at around week 4, when the TBV reached 50% of its maximum, and reached 95% of its maximum by 21–23 weeks of age (Conrad et al., 2012). The age of our piglets is between 17 and 35 days, with an average of 22 days old. The brain size of our piglets is similar to that of the piglets at around 2–5 weeks reported by Conrad et al. (2012). Thus, our template represents the brain with a total volume at around 50% of its maximum brain volume.

The major limitation is that we have a small number of subjects to generate the population-averaged template. However, it has been previously reported that a template stabilizes at around 10 subjects for most human populations based on observations (Avants et al., 2011a). As one of the ungulates, piglets show less complicated convolutional pattern of the brain than humans (Turner, 1890; Welker, 1990; Roth and Dicke, 2005), this eight-subject-based template should therefore be a good representation of brain morphology and its variability of the normal population at this weight range. The mean squared difference comparison between 8-subject-based-template and 2-, 4-, 6-, 8-subject-based templates showed that the templates perceive small differences between the 6-subject-based and 8-subject-based templates (Supplementary Material, Table S1 and Figure S1). Thus, although the sample size is small, the 6-subject-based DWI metric template can serve as a template for the normal piglets at this weight range. Furthermore, the fiber tracts were merged instead of averaged across subjects, extra number of subjects would not influence existing fiber locations or shape. Researchers can augment the tracts by merging tracts from their own data to the population-based tracts. To our knowledge, our atlas template study is the first study investigating the white matter connections of the piglet brain. The population-based fiber tracts presented here will facilitate ROI delineation study of the piglet brain.