The atlas-guided image expert (AGIE) will semi-automatically delineate structures in brain imagery using intensity and textural variation in the imagery itself, expected position and shape information coming from a brain atlas, and procedural knowledge (i.e., the best order in which to find structures and how to work around various problems in the imagery). Finding internal structures is an important stage in many of the methodologies used in neuroscience including immunocytochemistry, enzyme histochemistry, metabolic mapping, in situ hybridization, and receptor autoradiography. Currently, outlining is done by hand and it takes days to outline the structures of interest in a single brain. Outlining brain structures is a difficult task, since many do not have clear boundaries in histological images. Some of the information that a neuroanatomist uses to make decisions comes from the context rather than the imagery. They know the shapes and relative sizes of the structures that they are looking for, and their positions relative to easier to find structures. The key innovations we are proposing are developing: 1) a single delineation algorithm that can use intensity, texture, and expected position and shape in selecting the boundary points of a structure; 2) an agent for directing the delineation algorithm that will use many of the routines that human experts use in delineating brain structures. This effort will leverage our prior work in each of the technical areas that is key to AGIE development. We have already created high-resolution digital brain atlases of rat and mouse; developed a delineation algorithm that uses intensity and expected position and shape information; and created knowledge-based agents that direct extended visually guided behaviors. Our successes on these prior efforts give us confidence in our ability to complete the challenging work we have proposed. Once completed, the AGIE system will improve the speed with which brain structure delineation can be done, and will decrease the variability in the boundaries of the structures found compared to hand outlined boundaries.
