Page - 100 - in Proceedings of the OAGM&ARW Joint Workshop - Vision, Automation and Robotics

Fig. 1. Life cell imaging of SW13 cells expressing fluorescent HK8-CFP and HK18-YFP proteins (frames were recorded every 30 sec). (A) Snake evolution on a single frame; (B) Tracking result for an individual filament on a time-sequence of 40 frames; (C) Initial position of the snake on the first frame and (D-F) after 15 frames: (D) without stretching term and length constraint; (E) with stretching term only; (F) with stretching term and distance-based potential; network and allow snake endpoints to be captured by the force field induced by the potential (Fig. 1F). C. Overall tracking procedure In our setting, the tracking of individual filaments consists of two main routines: refinement of the position of the filament on the current frame and transition of the filament from the current to the next frame in the sequence. For the second step, we apply pyramidal Lucas-Kanade optical flow computation [1]. It allows to obtain a reasonable fit in case of large deformations of the filament. Incorrect mappings obtained by the optical flow algorithm require the repetition of the refinement step using active contours. Thus, we propose the following tracking procedure (see Fig. 2): Fig. 2. Block-diagram of the overall tracking algorithm (A) Initialization: The filament is initialized on the first analyzed frame. This can be done manually by user or additional (semi-)automatic segmentation procedures. (B) Image preprocessing: Gaussian smoothing; Hessian ridge detector; gamma contrast correction. (C) Calculate the GVF on the preprocessed image. (D) Optimize the position of the snake on the current image based on the GVF obtained in (C) and take into account a stretching term for open ends [3] and potential for the endpoints. (E) If the current image isn’t the last one in the analyzed sequence, go to the next step. Otherwise, exit the procedure here. (F) Calculate the pyramidal optical flow of the current image with respect to the next image in the time- sequence as described in [1]. (G) Transfer the snake to the next image in the sequence based on the calculated optical flow field. (H) Select the next image and repeat starting from step (B). A result obtained by this procedure is depicted in Fig 1. Fig.1Ashows theconvergenceof thesnakeonasingle frame with an “external energy” as defined above. Fig. 1B shows a filament being tracked in an image sequence of 40 frames. ACKNOWLEDGMENT This work has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 642866, and the DFG (LE 566/22-1). REFERENCES [1] J.-Y. Bouguet, “Pyramidal Implementation of the Lucas Kanade Feature Tracker: Description of the Algorithm,” Intel Corporation Microproces- sor Research Labs, Tech. Rep., 2000. [2] M. Kass, A. Witkin, and D. Terzopoulos, “Snakes: Active contour models,” International Journal of Computer Vision, vol. 1, no. 4, pp. 321–331, 1988. [3] H. Li, T. Shen, D. Vavylonis, and X. Huang, “Actin filament tracking based on particle filters and stretching open active contour models,” Medical ImageComputingandComputer-Assisted Intervention, vol. 12, no. 2, pp. 673–681, 2009. [4] H. Li, T. Shen, M. B. Smith, I. Fujiwara, D. Vavylonis, and X. Huang, “Automated actin filament segmentation, tracking and tip elongation measurements based on open active contour models,” in IEEE Interna- tional Symposium on Biomedical Imaging: From Nano to Macro, 2009. [5] D. M. Toivola, P. Boor, C. Alam, and P. Strnad, “Keratins in health and disease,” Current Opinion in Cell Biology, vol. 32, pp. 73–81, 2015. [6] C. Xu and J. L. Prince, “Gradient vector flow: A new external force for snakes,” in IEEE Conference on Computer Vision and Pattern Recognition, 1997. [7] T. Xu, H. Li, T. Shen, N. Ojkic, D. Vavylonis, and X. Huang, “Extraction and analysis of actin networks based on open active contour models,” in IEEE International Symposium on Biomedical Imaging: From Nano to Macro, 2011, pp. 1334–1340. 100