Cell detection and segmentation

Note: this is effectively a cross-post from http://www.mathworks.co.uk/matlabcentral/newsreader/view_thread/328425 - I apologise if I have picked the wrong place to post this.

Recently, I've been using and writing code in Matlab to perform segmentation and tracking of cells in images. Guides such as the the one here ( http://blogs.mathworks.com/steve/2006/06/02/cell-segmentation/ ) have been very useful for some image and cell types, but there is one image sequence in particular that I am struggling with. Here is an example of an image from the sequence: http://i.imgur.com/EB8vVwr.jpg

As you can see, there are a few reasons why the segmentation here is tricky; the intensity values of the insides of the cells are very similar to the outside of the cells, and the cell boundaries are often poorly defined. Any advice would be very much appreciated!

So far, I've tried using combinations of intensity thresholding followed by various morphological operations, edge detection, thresholding of gradient strength maps, and the recently added activecontour function, to little avail. As I'm also interested in tracking and have multiple frames available, I tried using the outline suggested in http://www.mathworks.co.uk/help/vision/examples/motion-based-multiple-object-tracking.html, but I have had no luck in getting a reliable output.

Does anyone know of a way to successfully segment an image like this?

Answer 1

[Image Analyst]

What are you trying to measure? The count? Have you tried ginput() to let the user manually click on cells? The problem is there is no definite count with this image. It looks like you have some cells in the focal plane but you have others partially in the focal plane. Those are difficult to find and count. You might be able to find something searching here: http://iris.usc.edu/Vision-Notes/bibliography/contents.html Search it for your type of microscopy, like DIC or whatever.

It's possible the level set method may work in your situation: http://en.wikipedia.org/wiki/Level_set_method There is some code in the File Exchange: http://www.mathworks.com/matlabcentral/fileexchange/index?utf8=%E2%9C%93&term=%22level+set%22

[Ess]

Thanks for the reply. In answer to your questions and comments:

• I am interested in the count and the shape of the cells. I accept that for both, I won't get a perfect answer, but the more accurate, the better.
• I have not used ginput so far as I am hoping to have an automated method, or at least a method for which parameters would be entered on the command line rather than a more GUI-base method, but I might experiment with it.
• Implementations of the level-set method that I have tried have not yielded a successful result, but I might try a couple of others in the hope that some adjustments may make it more appropriate for my problem.