Automatic segmentation and tracking of keratin filaments in

• Dmytro Kotsur
• Saransh Vora
• Rudolf Leube
• Reinhard Windoffer
• Julian Mattes

Keratin filaments form a highly branched dynamic network, which impacts the mechanical properties of epithelial cells and which protects them against mechanical stress and damage. This network has highly variable properties differing from one tissue (cell type) to another. The investigation of the network properties requires a deep and accurate analysis of a huge amount of image data, which would be extremely time-consuming and difficult to accomplish manually.

Therefore, we present an image analysis tool, which allows to segment and track the trajectory of individual keratin filaments within their network. The segmentation procedure is based on the Voronoi medial axis, which allows to obtain initial position of the filaments with sub-pixel accuracy. The tracking procedure employs an active contour model (snake), which we extend by an additional term for an accurate snake’s endpoints tracking. The proposed algorithm was evaluated on 300 ground truth trajectories using different error measures and compared with two state-of-the-art methods. We also investigated the problem of automatic hyper-parameter tuning for the snake algorithm, especially heuristic parameter choice rules with respect to the gradient vector flow problem.