Actin, a protein common in muscle cells of animals, forms fibers and fiber bundles which can be dyed with fluorescent markers and then detected by a light microscope. Quantitative measurement of the properties of Actin fibers will lead to a better understanding of how Actin interacts with other cell structures and contributes to cell locomotion and morphology.
We present a method for detection of Actin fibers in 2D images. Our approach has three stages. First, a set of pixels that follow the contours of the fibers in the image is extracted from the original gray-scale cell image using edge detection techniques. Each fiber has two edges associated with it, so the next step eliminates one edge associated with each fiber by selecting an interval of edge direction such that one half of the edge pixels lie in this interval. The surviving edge pixels that have an edge magnitude above a threshold are selected and then thinned.
The second stage connects all of the edge pixels into a minimal spanning tree, using inter mediate algorithms from computational geometry. As a result, successive points from a fiber contour will tend to be linked. However, the tree will also connect points that belong to different fiber contours partly because the fibers intersect in the image, but also because by definition a tree must provide a path of links between any two vertices in the tree.
The third and final stage extracts the individual fiber contours from the minimal spanning tree. Long links within the tree are deleted because they connect two different fibers. Intersections of multiple fiber contour at a tree vertex are handled by heuristics based on prox imity and on rough collinearity. The overall result is a set of fiber contours that appear in the a 2D cell image.
Our approach has worked well on a small set of real and synthetic images. The minimal spanning tree approach has proved fairly accurate because it groups pixels based on a glo bal perspective rather than relying on uncertain local predictions of where fibers might extend.