Voronoi-based skeletonization algorithm for segmenting the network of intermediate filaments

• Dmytro Kotsur
• Saransh Vora
• R. Windoffer
• R. E. Leube
• V. Tereshchenko
• J. Mattes

Intermediate filaments (IFs) form a highly dynamic branched network inside living cells, which contributes to physiological and biomechanical properties of the cell. In order to facilitate investigations of the properties and functions of the IF network in living cells, we developed an algorithm, which allows to extract centerlines and the topology of the IF network based on 2D fluorescence microscopy data. In contrast to pixel-level approaches for center line extraction (e.g., distance map-based or morphological thinning-based methods), our method is based on the Voronoi-skeleton constructed for the contours of the network silhouette. This allows us to immediately obtain a graph representation of the filament network with subpixel precision and to apply a wide diversity of graph mining algorithm to analyze such data. We evaluated our algorithm on 70 synthetic ground truth images and 10 manually labeled fluorescence microscopy images and compared its performance with the SOAX software for biopolymer network segmentation.