Three-dimensional reconstruction of the actin cytoskeleton from stereo images.

In eucaryotic cells, actin filaments are abundant components in the cytoskeleton where they form a complex three dimensional (3D) structural network that provides the cell with its shape and mechanical properties. However, understanding the structural and mechanical properties of actin filaments composing the cell cytoskeleton is often hampered by the inability to faithfully reconstruct the three-dimensional geometric relationships. This paper presents a vision-based reconstruction approach that automatically reconstitutes the three-dimensional structures of cytoskeletal polymers from stereo image pairs taken at the different tilt angles. The approach finds corresponding points between two images and recovers the depth information about the structures. The computational process consists of three major procedures: feature representation, stereo matching, and disparity refinement, implemented in a multi-resolution manner based on a coarse-to-fine strategy. The reconstruction depicts the three-dimensional structure of cytoskeletal polymers and their geometric relationships. New and useful information becomes available and allows quantitative analysis of the structure. Measurement of the cytoskeleton geometrical properties and the filament concentration in a defined volume are obtained by direct calculation.