On Monday 04 February 2008 18:26:12 David Knecht wrote:
> I am trying to help a user calculate the area of cells in an image.
> The cells are all interconnected in a continuous sheet ( tissue
> section) with the membranes visible in the brightfield image. The
> outlines can be fairly easily identified by thresholding, but I am not
> sure how to deal with the interconnected nature of the cells so that
> each closed unit is counted as a separate object.
Hi,
When looking into solid tumours and epithelial tissues, we also faced the same
problem. Most often one cannot find the exact boundaries of cells.
One alternative we found useful was to use the nuclei to create "virtual"
cells: divide the tissue compartment into exclusive areas based on the
position of the nuclei (similar to a voronoi diagram, but based on the full
nuclear profiles to avoid boundaries passing through the nuclei).
We found that those v-cells, although artificial, still capture enough
architectural information to allow for comparisons and classifications.
There are a few different ways to extract these v-cells. We use greyscale
reconstruction to extact "domes" (i.e. the dark nuclei after colour
deconvolution) and then applying a watershed algorithm to partition the
remaining space into the v-cells.
Here are some examples with details:
http://www.head-face-med.com/content/pdf/1746-160X-2-4.pdfhttp://www3.interscience.wiley.com/cgi-bin/fulltext/109584758/PDFSTARThttp://www.blackwell-synergy.com/doi/pdf/10.1046/j.1365-2818.2003.01113.xMaybe this could help with your problem too.
We plan to upload some examples page in the "not too distant" future :-).
Cheers,
Gabriel
Locked
