Automatic Fusion of 3D staining patterns from mouse embryo tomograms

Poster presented on July 21, 2008 by Ruben Schilling at ISMB 2008 Toronto.

Abstract: The spatial distribution of molecular staining patterns, e.g. from gene expression or proteins, is important to understand developmental processes on the molecular level. The information when, where and under which condition gene expression or protein concentrations are present in an organism can be recorded from in-situ and staining experiments and result in images of the subject containing specific patterns. When scanning a whole catalogue of genes for a hypothesis one typically needs to compare a large set of transcriptional patterns, which is a time-consuming process. Furthermore this type of analyis does not lead to a quantitative underestanding of the processes. This leads us to the formulation of an image registration problem, that can be stated as finding a mapping of one image to another similar image, such that their difference is minimized. Similarity is measured on the level of intensities of the images. Our method first searches for a best affine match and derives then from the similarity measure a driving force, that deforms the images over time into a matching state. Our computational method for the fusion of 3D staining patterns from whole, intact mouse embryo images is non-parametric and does neither require user input nor a segmentation. We exemplify in our results the success of our method by validating the co- and differentially expressed regions of the genes Cdx1 and Spry1 in mouse embryos of Theiler stage 15.

Download PDF of Poster.