Posted by
Gabriel Landini on
Nov 19, 2008; 8:32am
URL: http://imagej.273.s1.nabble.com/counting-cells-in-multiclour-fluorescence-images-tp3694463p3694466.html
On Tuesday 18 November 2008 20:47:27 Tony Collins wrote:
> I've been playing with some semi-automated cell identification and put the
> macro so far below. It's based on automated cell segmentation in high
> content screening where a nucleic acid stain (DARQ5 in our case) is used to
> first identify the nucleus (DNA really bright) then the cytoplasm (RNA
> being fainter). You can kind of generate a similar type of image by
> splitting your RGB, dividing the red and green by 2 to make them fainter
> and adding the red green and blue back together.
>
> The macro uses two thresholds one for nucleus and one for cytoplasm which
> are here in the text (a better tweak would be to pull the upper and lower
> bounds from the user defined threshold).
>
> It uses particle detection to find nuclei then maxima detection within the
> lower threshold bound to find the cytoplasm. Once it's identified the cells
> and put them in the ROI manager then you can go crazy with the analysis.
>
>
http://www.macbiophotonics.ca/images/cellSegmentation.jpgThe problem with this type of approach is that the cytoplasms are all
different sizes, so one is not comparing the same thing. E.g. is it a fair
comparison between cell 30 and cell 12?.
What about thresholding the blue channel (nuclei), make sure that each nucleus
is separated (one could run the Binary Watershed for this) and then do
a "Dilate-no-merge" of the nuclei (let's say 3 or 4 iterations) (a plugin for
this is available in the Morphology collection).
Then subtract the nucleus mask from this "dilated-not-merged" mask (which
should leave binary "rings" that have an upper bound more comparable between
cells).
Finally do a logical AND of the ring mask and the original and analyse the
contribution of the red and green channels.
Cheers
Gabriel