Active Graph Matching for Automatic Simultaneous Segmentation and Annotation of C. Elegans

A frequently used model organism in developmental biology is the worm C. Elegans. Since C. Elegans is highly stereotypical it is well suited for comparative developmental studies. A common and time consuming image processing problem in such studies is the segmentation and annotation (labeling) of cell nuclei with their unique biological names in 3-dimensional microscopic images.

This talk presents an automated method that helps solve this problem. The method, called Active Graph Matching, integrates the popular active shape models into a sparse graph matching problem. In effect, Active Graph Matching combines the benefits of a global, statistical deformation model with the benefits of a local deformation model formulated as a second-order Markov random field. Despite the respective optimization problem being NP-hard, a new iterative energy minimization technique yields empirically good results. As compared to state-of-the art methods, Active Graph Matching yields considerably higher accuracy in annotating nuclei in 3D light microscopic images of C. Elegans. Furthermore an additional pre-processing step in the form of the generalized Hough transform allows for simultaneous segmentation and annotation of a large set of nuclei in a fully automatic fashion for the first time.