Kuss,A., Hege,H.-C., Krofczik,S., Borner,J.
Abstract:
Digital atlases of the brain serve as a spatial reference frame which can be used to relate data from different image modalities and experiments. In this paper we describe a standardized pipeline for the creation of extendable surface-based anatomical atlases from 3D image data of a population of individuals. The techniques are utilized for creation of atlases of insect brains. The pipeline consists of the major steps imaging and preprocessing, segmentation, averaging, surface reconstruction and surface simplification. At first, 3D image data sets (in our case from confocal microscopy) are resized, stitched and initially displayed using standard image processing and visualization tools. Then brain structures such as neuropils and neurons are labeled by means of manual segmentation and line extraction algorithms. The averaging step comprises affine and elastic registration and a mean shape selection strategy. Finally non-manifold surfaces of the labeled and aligned structures are reconstructed using a generalized surface reconstruction algorithm. These surfaces are simplified and adapted to further needs by decimation and retriangulation. The chosen methods of each step are adequate for a variety of data. We propose a recursive application of the pipeline in order to build the atlas in an hierarchical fashion integrating successively more levels of detail. The approach is applied in several different neurobiological research fields.