Geometrical constraints to improve elongated structures segmentation applied to actin filaments

Abstract

Elongated structures are widely found in biological systems such as cell cytoskeleton and neurons dendritic tree. These structures are observed through optical microscopy and quantified by image processing techniques. However, in fine structures such as filaments quantification methods commonly fail to identify them due to their small size, irregular intensity, and intersections. In this work we propose an image processing method to identify filaments by introducing novel geometrical constraints based on maximal filaments distance and curvature in order to improve segmentation. An example application is shown by applying the proposed method to stress fibers of culture cells. Our work also illustrates how fiber identification can be formulated as an optimization problem.