Bioinformatics Advance Access originally published online on May 29, 2008
Bioinformatics 2008 24(14):1647-1649; doi:10.1093/bioinformatics/btn247
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
CellTrack: an open-source software for cell tracking and motility analysis
1Computer Engineering Department, Middle East Technical University, Ankara, Turkey, 2Department of Computer Science and Engineering and 3Mathematical Biosciences Institute, The Ohio State University, Columbus, OH, USA
*To whom correspondence should be addressed.
 |
ABSTRACT |
|---|
 TOP ABSTRACT 1 INTRODUCTION2 METHODS3 THE SOFTWAREACKNOWLEDGEMENTSREFERENCES |
|---|
Motivation: Cell motility is a critical part of many important biological processes. Automated and sensitive cell tracking is essential to cell motility studies where the tracking results can be used for diagnostic or curative decisions and where mathematical models can be developed to deepen our understanding of the mechanisms underlying cell motility.
Results: We have developed CellTrack: a self-contained, extensible, and cross-platform software package for cell tracking and motility analysis. Besides the general purpose image enhancement, object segmentation and tracking algorithms, we have implemented a novel edge-based method for sensitive tracking of the cell boundaries, and constructed an ensemble of methods that achieves refined tracking results even under large displacements or deformations of the cells.
Availability: CellTrack is an Open Source project and is freely available at http://db.cse.ohio-state.edu/CellTrack
Contact: sacan{at}cse.ohio-state.edu
|
1 INTRODUCTION |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 THE SOFTWAREACKNOWLEDGEMENTSREFERENCES |
|---|
Cell motility is an essential part of many biological processes that are necessary for the sustenance of an organism. Free-living unicellular organisms move to avoid toxic substances or to approach nutrients. Tissue cells of multicellular organisms mobilize during embryologic development (morphogenesis), wound healing, maintenance of tissues, generation of new blood vessels, cancer metastasis and immune response. Understanding the underlying mechanisms of cell motility is crucial for curative or preventative treatments to many diseases that are caused by abnormalities in cell locomotion.
Accurate segmentation and tracking of cells in microscopic imaging is becoming to be an important step in cell-motility studies. For instance, tracking number and velocity of rolling leukocytes is essential to understanding and successfully treating inflammatory diseases (Ray et al., 2002). Mathematical modeling of cell locomotion also requires sensitive tracking of moving cells. In Coskun et al., (2007) , live cell imaging data is used to solve the inverse modeling problem in order to determine the material properties of the cells.
Manual processing of cell locomotion data is labor-intensive and error-prone. There has been a number of recent attempts to automate the detection and tracking of cells from microscopic imagery. Zimmer et al., (2002) developed a modified active contour model to detect the pseudopods of the moving cells and to handle dividing cells, however required that an initial segmentation is to be provided for the first frame. Mukherjee et al., (2004) proposed a unified search algorithm to handle segmentation and tracking problems simultaneously, using image level sets computed via threshold decomposition. Li et al., (2006) developed a two-level system where the output of the lower level, which comprised of a cell detector, a level-set tracker and motion filter, are evaluated by a track arbitrator level to detect mitotic and apoptotic events and to handle cells that move in or out of the view.
Automated cell tracking methods developed so far mainly focus on association of cells across frames and do not provide a sensitive tracking of the cell as it deforms during its locomotion. Moreover, these methods are usually developed in isolation, addressing a specific problem, but not as part of an integrated software environment. The available software packages that do incorporate some of these methods are either proprietary or closed source, which makes them inapplicable and non-extendible to different research problems.
We have implemented CellTrack, an integrated and extensible software environment for tracking cells. CellTrack provides general purpose image processing, object segmentation and tracking methods. We have implemented a new edge-based tracking method that is based on snakes (Kass et al., 1987) and relies on a new energy functional that can accurately track changes in the shape of a moving cell. We also develop an ensemble of methods to achieve more robust tracking results. We introduce these methods in the following section and provide some preliminary qualitative results to illustrate the benefits of these new methods. Extensive quantitative evaluation of these methods and their comparison with other available methods are deferred to a separate study.
|
2 METHODS |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 THE SOFTWAREACKNOWLEDGEMENTSREFERENCES |
|---|
Snakes are elastic curves that evolve on the image plane to capture object boundaries (Kass et al., 1987). Snake evolution is based on an energy minimization procedure that improves the coordinates of the snake. The snake energy is generally defined as a weighted sum of internal (Eint, e.g. continuity and curvature) and external (Eext, e.g. image gradient) energy terms over the snake's elements (snaxels). Using snakes for object tracking involves initializing a snake to its configuration from previous frame and evolving the snake again for the new frame. Traditional tracking methods focus on tracking the objects as a whole, without being concerned about providing a sensitive association of the boundary points across frames.
Our novel approach uses the configuration of the snake from the previous frame not just for initialization, but also as a constraint energy term, such that the snake in the new frame would effectively match up, on a per-snaxel basis, with the fitness of its previous configuration to the image plane. Our proposed energy functional is as follows:
|
|
|
|
I) at each snaxel position, and the arc-length (
) of the contour around each snaxel. Note that our model can readily be extended to other color or texture feature descriptors. Figure 1 illustrates the success of our energy functional. Without the Ematch terms, the snake fails to capture the boundary of the tracked object but falsely deforms to the edges of the neighboring objects. Inclusion of Ematch terms causes the snake to maintain its internal and image-relative configuration from the previous frame to correctly capture the boundaries of the same object in the new frame. Note that our approach also has the unique benefit of providing a sensitive association of the snaxels across frames (see the inset of Figure 3 for an illustration). This is especially useful in cell-motility modeling and analysis studies, where the tracking of the bulk of the object is not sufficient and such a sensitive association is required (Coskun et al., 2007).
|
Note that tracking using snakes relies on the assumption that the movement and deformation of the tracked object is small between consecutive frames. In order to relax this assumption and obtain more robust tracking results, we developed an ensemble of different tracking methods in a coarse to refined fashion. We refer the reader to Yilmaz et al., (2006) for a general survey of the object tracking methods.
Our combined method performs the following steps: the overall displacement and rotation of the object is first determined using a template matching method (Fig. 2b). The resulting contour is used as the initial state to pyramidal Lucas–Kanade (Bouguet, 2000) optical flow-based deformation (Fig. 2c); we use statistical outlier detection and local interpolation to achieve resistance against errors in the optical flow evaluation. Our extended snake method is then applied to obtain the final snake configuration (Fig. 2d). The ensemble method achieves accurate tracking even for large displacements or deformations of the objects between frames.
|
|
3 THE SOFTWARE |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 THE SOFTWAREACKNOWLEDGEMENTSREFERENCES |
|---|
The methods outlined earlier are implemented in CellTrack, a software package that aims to automate the cell tracking process. The simple and intuitive user interface allows easy navigation and analysis of image frames (see Fig. 3 for a snapshot). In addition to the tracking methods, CellTrack provides general image processing and enhancement functions such as smoothing, background subtraction, and object segmentation. For each of these methods, the user can change the default parameters to meet the discrepancies and demands of different tracking applications. The ability of immediate previewing of results makes it easy to investigate the effect of each parameter.
|
The speed, area, deformation, trajectory and detailed tracking of the cells are computed and displayed for analysis. Besides automated cell detection and tracking capability, the interface also allows manual editing to initialize or modify the tracking data. CellTrack can be used to work with movie or image files of a variety of file formats. The tracking results can be exported either as raw text data for further numerical analysis, or as movie or image files for visualization, sharing and publishing.
CellTrack is developed in C+ so as to avoid dependence on commercial development and deployment products and to avoid the computational overhead of higher level languages. The image processing functions are based on OpenCV library (Bradski, 2000) and the graphical user interface is implemented with wxWidgets (Smart et al., 2005). Both of these libraries are open source and available for a broad range of platforms including Linux, Windows and Mac OS, making CellTrack also a cross-platform software. CellTrack has been developed in a modular design pattern where the processing logic is separated from the user interface details and implemented as separate plugins. This design pattern makes CellTrack easily extendable to incorporate new methods. Finally, we wish to note that CellTrack is by no means a final solution to all cell tracking problems, but a platform for continuing development. Please see the user manual of the software distribution for the current capabilities and limitations of CellTrack.
|
ACKNOWLEDGEMENTS |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 THE SOFTWAREACKNOWLEDGEMENTSREFERENCES |
|---|
Funding: This research is partially supported by US National Science Foundation (NSF) Grant IIS-0546713 and Turkish Scientific and Research Council (TUBITAK) Grant 107E173.
Conflict of Interest: none declared.
|
FOOTNOTES |
|---|
Associate Editor: Olga Troyanskaya
Received on March 8, 2008; revised on May 7, 2008; accepted on May 23, 2008
|
REFERENCES |
|---|
|
TOPABSTRACT1 INTRODUCTION2 METHODS3 THE SOFTWAREACKNOWLEDGEMENTSREFERENCES |
|---|
Bouguet J. Pyramidal implementation of the Lucas Kanade feature tracker: description of the algorithm. In: Technical Report. Intel Corporation, Microprocessor Research Labs, OpenCV Documents (2000).
Bradski G. The OpenCV Library. Dr. Dobb's Software Tools for the Professional Programmer (2000) 25:120–125.
Coskun H, et al. Ameboid cell motility: a model and inverse problem, with an application to live cell imaging data. J. Theor. Biol (2007) 244:169–179.[CrossRef][Web of Science][Medline]
Kass M, et al. Snakes: active contour models. Int. J. Comput. Vision (1987) 1:321–331.[CrossRef]
Li K, et al. Online tracking of migrating and proliferating cells imaged with phase-contrast microscopy. In: IEEE Proceedings of the 2006 Conference on Computer Vision and Pattern Recognition Workshop (CVPRW'06) (2006) New York. 65–72.
Mukherjee D, et al. Level set analysis for leukocyte detection and tracking. IEEE Trans Image Process (2004) 13:562–572.[CrossRef][Web of Science][Medline]
Ray N, et al. Tracking leukocytes in vivo with shape and size constrained active contours. (2002) 21:1222–1235.
Smart J, et al. Cross-Platform GUI Programming with wxWidgets (2005) Indianapolis: Prentice Hall PTR.
Yilmaz A, et al. Object tracking: a survey. ACM Comput. Surv (2006) 38:13.[CrossRef]
Zimmer C, et al. Segmentation and tracking of migrating cells in videomicroscopy with parametric active contours: a tool for cell-based drug testing. IEEE Trans. Med. Imaging (2002) 21:1212–1221.[CrossRef][Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  S. Jaeger, Q. Song, and S.-S. Chen DYNAMIK: a software environment for cell DYNAmics, Motility, and Information tracKing, with an application to Ras pathways Bioinformatics, September 15, 2009; 25(18): 2383 - 2388. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||