Cell analysis with morphological tools
By analyzing changes in cells, morphology software tools automatically measure responses to chemical compounds or genetic alterations.
By John Haystead,Contributing Editor
A powerful image-processing tool, mathematical morphology is applicable to a number of vision-system applications, including industrial inspection, automatic target detection, and remote sensing. In biomedical imaging, the technique has proven effective in the segmentation and delineation of complex cell structures. Originated by Georges Matheron and associates at the Institute for Mathematical Morphology (Fountainebleau, France), morphological techniques can be applied to a range of image-processing tasks such as edge detection, noise suppression, image enhancement, skeletonization, segmentation, and pattern recognition, among others.
FIGURE 1. In morphological image analysis, a basic 3 x 3 structuring element or mask can be used for dilation and erosion operations. With the pixel of interest at the center position, the algorithm examines the values of the neighboring pixels and assigns the value of the central pixel according to these values.
In general, morphological techniques process digital images based on shape and structure. By systematically probing an image with a small geometric template or structuring element, the shape and structure of objects within the image can be identified and analyzed.
Any type of morphology operator can be built hierarchically from two operators called dilation (enlarging) and erosion (shrinking). Both operators are parameterized by particular structuring elements that can be flat (binary images or signals) or nonflat (gray-scale images or signals).
Dilation and erosion can be seen as the intersection and union, respectively, of an appropriate number of shifted versions of the original image. Each shift amount and direction depends on the content of the structuring element, and each position of the structuring element is associated with a particular shift (see Fig. 1). More complex operations, such as opening, closing, and edge detection are implemented by combining these two basic operations.
Cytokinetics Inc. (San Francisco, CA) is using MatLab 5 software from The MathWorks (Natick, MA) and the Morphology Toolbox from SDC Information Systems (Naperville, IL) to prototype cell-analysis algorithms. By analyzing certain changes in specific cellular components within microscopic images, morphology techniques can automatically measure the response of cells to chemical compounds or genetic alteration.
"The first step in the prototyping process is the selection of biological markers that reflect the physiological state of the cell under test," says Eugeni Vaisberg, a scientist at Cytokinetics. So far, the specific markers identified by Cytokinetics are proprietary, but all respond to chemical action within the cell. They include many aspects of the cytoskeleton—a three-dimensional, fibrous structure that fills the cytoplasm and determines cell architecture.
Although the cytoskeleton is transparent in standard light and electron-microscope preparations, it is a complex and dynamic cell component that maintains the cell's shape and moves parts of the cell in processes of growth and motility. Cyto-architecture influences most cellular processes and provides a dynamic and constant readout on processes occurring in the cell.
The next step is the acquisition of several images that represent the cells in different physiological states. At Cytokinetics, these images are acquired using the company's microscope-based imaging system. The resolution of the images is typically 1024 x 1280 x 16-bit gray-scale.
After combining and correlating any morphological changes to the cellular markers within the various cell images, image-analysis algorithms are used to automatically discriminate between the different cell states. "The Morphology Toolbox is a useful prototyping tool because it capsules some of the cellular-analysis algorithms that we would otherwise have to develop, implement, and test ourselves," says Vaisberg,
TOOLBOX MORPHOLOGY
The SDC Morphology Toolbox is a collection of gray-scale morphological tools that can be applied to image segmentation, nonlinear filtering, pattern recognition, and image analysis. "Gray-scale morphology is common in research, but there are few commercially available software tools," says Roberto Lotufo of the State University of Campinas (Campinas, Brazil), a developer of the product. "Our objective was to minimize the configuration and programming that users need to perform."
The toolbox comprises a family of discrete nonlinear filters based on lattice operations called morphological operators. These operators range from classical morphological filters for restoration and shape description to complex connected filters and watershed transforms.
In the toolbox, several morphological operators are also implemented as special fast algorithms, such as distance transform, watershed, reconstruction, labeling, and area opening. Users begin by examining a number of demonstration programs and then selecting those features that suit their particular application.
FIGURE 2. In blood-cell analysis, microscopic gray-scale images can be segmented and processed to separate superposed cells. The segmentation procedure is based on gray-scale-connected filtering and threshold. The separation procedure is based on the classical watershed approach. Detected watershed lines are overlaid on the cells binary image (left). The cells are separated using the watershed lines (middle). The final image shows the superposition of the contour of the detected cells on the original image (right).
One of the advanced analysis tools included in the toolbox is the Watershed Transform. The watershed cell-segmentation principle can best be understood by regarding the gray-scale image as a volumetric (or topographic) representation with lighter pixels representing higher altitudes and darker pixels representing lower altitude regions. By taking a gradient of the image, the highest peaks or cell edges appear as faint white rings against the background noise. By gradually flooding the image with differently colored fluid markers beginning at points inside and outside of the cells, but never allowing the fluids to mix, a boundary line is formed wherever they meet. The result is a continuous border always located at the highest peak of the gradient line at each point that connotes the depressed border of each cell (see Fig. 2).
Although Cytokinetics is not currently implementing watershed processing into its image-analysis algorithms, Vaisberg says they are planning to look more closely at the approach in the future. "We've tried watershed and other approaches in developing our recognition algorithms, and the toolbox allows us to test the approach and to decide whether it will be useful."
Vaisberg says that one of the benefits of the toolbox is that it requires little additional configuration and programming by the user. He adds that the tool would be more valuable if developers could view and directly implement the toolbox's underlying C code.
Company Information
Institute for Mathematical Morphology
77305 Fontainebleau, France
Web: cmm.ensmp.fr
Cytokinetics
South San Francisco, CA 94080
Web: www.cytokinetics.com
The MathWorks
Natick, MA 01760
Web: www.mathworks.com
SDC Information Systems
Naperville, IL 60540
E-mail: [email protected]
State University of Campanas
13081-970 Campinas, SP, Brazil
Web: www.dca.fee.unicamp.br