Localization of sections within the brain via 2D to 3D image registration

The mouse brain library (MBL) is a database of brain images, each consisting of sparse coronal or horizontal sections. To facilitate morphometric research, it is necessary to index each of these sections by its location within a canonical 3D atlas (NeuroTerrain). This is done with a 2D to 3D matching technique which was developed in this study. The registration method is imaged-based and uses a genetic algorithm to find the local-affine transformation that maximizes a mutual information metric. The average distance between the registration results achieved with the proposed method as compared with manual matching by an expert was 250 microns. This compares well with repeated manual trials where inter-trial matching distance was on average 200 microns.