Reusable Modules for High-Content 3D and 4D Image Analysis

Abstract: Fully automated high-throughput and high-content experiments need reusable general modules, that can be combined in a flexible way to build the solution. Even though the biological objects or structures may not share any common features, the transformations that act on the structures (like rotations, translations or deformations) are nearly the same in every experiment. In the talk I will show general concepts for the detection, recognition or comparison of complex high-level structures, that we use in our daily work. Invariant features are computed by the integration over the transformation group. Rotation-invariant detection is done using an image representation that encodes the local surrounding of every voxel in spherical harmonics basis functions. A general module for tracking of complex structures can be based on recent high accuracy optical flow methods. The successful applications of these concepts are, e.g., the classification of pollen grains, the rotation-invariant detection of landmarks in zebrafish embryos and the analysis of growth patterns in frog kidney.