Registration is a process that attempts to manipulate two or more 3D datasets into a common 3D space, so that similarities and differences between individual samples at equivalent locations can be recognized. This has been used to great success in neuroanatomy studies (Spring et al., 2007; Dorr et al., 2008; Lerch et al., 2008). It has recently been applied to mouse embryo studies of general anatomy, though this work has tended to focus on older, larger embryos that have been imaged using magnetic resonance imaging (MRI) (Zamyadi et al., 2010; Cleary et al., 2011) or micro-computed tomography (micro-CT) (Wong et al., 2012b). The embryo registration method employed here is based upon the current mouse brain registration algorithm employed in our laboratory (Kovacevic et al., 2005), but has been modified with additional steps to make it sensitive to the very fine features of the mouse embryonic vasculature. We first looked at C57Bl/6 embryos. We collected autofluorescence and vascular OPT scans of somite stage–matched embryos (20–23 somites; n=7). During the registration algorithm, the autofluorescence data (Fig. 2A–G, represented as volume textured samples) are first used to register the embryos together, where first a set of 3D affine transformations (translations, rotations, scaling factors, and shearing factors), and secondly non-linear, voxel-by-voxel transformations are used to generate a representative consensus average image of all embryo inputs (Fig. 2H). These transformations are encoded in a deformation field for each 3D input image, which, when applied to the original autofluorescence image of each embryo, result in registered images that look very similar (Fig. 2A'–G', A''–G'', and A'''–G''', shown in three orientations). The corresponding deformation fields are then applied to the vascular scans of the individual embryos (Fig. 3A–G, represented as volume-textured images in their original orientations), which approximately registers the individual vascular trees together. However, since the vascular structures are so fine, there are evident registration mismatches (data not shown). Therefore, the vascular images are then subjected to an additional registration algorithm that consists of another set of non-linear, voxel-by-voxel transformations. Thus, a representative consensus average of all vascular maps is calculated (Fig. 3H), and the same process as previously described results in registered vasculature images that look very similar (Fig. 3A'–G', A''–G'', A'''–G''', shown in three orientations).
Figure 2. Autofluorescence registration of seven E9.5 embryos. Autofluorescence scans of the embryos (A–G) are used for the first stage of the registration process (see Experimental Procedures section). Following both linear/affine and non-linear registration, a population average of the seven embryos is generated (H), which represents the global anatomy of the embryos. All of the registration steps necessary to reach the population average are then applied to each individual embryo to see how each embryo has changed (A'–G', A''–G'', A'''–G'''; shown in three different orientations). All data can be fully manipulated in 3D. Scale bars = 500 μm.
Download figure to PowerPoint
Figure 3. Vascular tree registration of seven E9.5 embryos. Following registration of the global anatomy, the original vascular scans of each respective embryo (A–G) are then subjected to the deformation field produced by the registration of the autofluorescence data, followed by further non-linear registration steps in order to account for small residual differences in the vascular scan data (see Experimental Procedures section), which again generates a population average vasculature image of the embryos (H). The registration steps generated in this process to reach the population average can once again be applied to the original vascular scan data in order to visualize the vascular patterns of each individual embryo in equivalent orientations (A'–G', A''–G'', A'''–G'''; shown in three different orientations). All data can be fully manipulated in 3D. Scale bars = 500 μm.
Download figure to PowerPoint
We next performed the same procedure on a set of somite stage–matched mouse embryos on the CD-1 background (20–23 somites; n=7). Autofluorescence scans and vascular scans were obtained in the same fashion as for the C57Bl/6 embryos. Registration of the autofluorescence data produced a representative consensus average image of all embryo inputs as before, and we were similarly able to register the vascular scans of the embryos to produce a representative consensus average image, and thus register the vascular maps in 3D to a high degree (data not shown). The genetic background of mice, therefore, does not appear to have any bearing on whether or not registration of their anatomy or vascular trees is possible.