Summary: We are designing real-time algorithms for massive yeast cell segmentation and tracking. The key idea is to use low-cost trackers for individual cells combined with a joint estimation of the global movement in order to relax time resolution constraints. We are also investigating the possible extraction of statistical features to improve the robustness of the algorithms.
Introduction
The focus in systems biology is now shifting towards a dynamical understanding of how cellular processes play out in complex regulatory situations. Global precision measurements of cellular networks will enable the scientific community to establish principles governing biological function and provide the necessary details to eventually reach the point where predictive modeling becomes an integral part of biology. To understand cell cycle and growth control on a systems level, it is necessary to measure and analyze the temporal characteristics of protein abundance and localization during the mitotic cell cycle, and in response to changing growth signals.
Sequence of yeast cells growing in a MITOMI chamber.
Main Contributions
Among the various state-of-the-art segmentation strategies available (e.g., snakes, levels sets, graph cuts, etc.), we can identify snakes (or active contours) as being the most relevant for our purpose because the shape of yeast cells is quite constrained (essentially circular). Snakes are polygonal or parametric curves that evolve on the image by minimizing an appropriate energy function (which typically favor contours) and that eventually snap onto an object. Snakuscules are probably the classes of active contours most adapted to our needs because they admit a minimal parameterization (few spline knots because the outline of a yeast cell is rounded) and also because it is easy to constrain their shape.
Snakes are usually deployed to detect single objects and there have been only few attempts in applying them to the simultaneous detection of multiple targets, and certainly not at the scale that is required by this project.
We are currently working in providing the snakuscule with a more general shape while keeping its essence; that is, its low computational cost. We are also investigating the use of probabilistic models to increase the robustness of the tracking algorithms by constructing motion estimators and predictors.
Collaborations: Prof. Michael Unser, Prof. Sebastian Maerkl (LBNC-EPFL)
Period: 2008-ongoing
Funding: Swiss SystemsX.ch initiative under Grant 2008/005
Major Publications
[1]
P. Thévenaz, R. Delgado-Gonzalo, M. Unser, "The Ovuscule," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 33, no. 2, pp. 382-393, February 2011.
[2]
R. Delgado-Gonzalo, N. Dénervaud, S. Maerkl, M. Unser, "Multi-Target Tracking of Packed Yeast Cells," Proceedings of the Seventh IEEE International Symposium on Biomedical Imaging: From Nano to Macro (ISBI'10), Rotterdam, The Netherlands, April 14-17, 2010, pp. 544-547.
