Posted by
Kenneth Sloan on
May 21, 2014; 2:58pm
URL: http://imagej.273.s1.nabble.com/Segmentation-of-confluent-cells-tp5007785p5007831.html
Voronoi analysis is very powerful (it’s what I have used almost exclusively in
analysis of en face images of human retina; we’ve been doing this for .. awhile...).
The one caveat is that you must understand that the Voronoi polygons are not IDENTICAL
to the cell boundaries, and if your subsequent analysis depends critically on
what is happening near the boundaries, you may be introducing an extra
source of variability.
And…especially with the “interesting” cases, the cells may not be convex…
And…the nuclei may not be particularly close to the “center” of the cell…
But, again - that’s what we do with (for example) RPE cells in human retina.
We stain the cytoskeleton, semi-automatically identify the “center” of each cell,
allow a trained observer to correct, extend, or completely re-do the “findCenters” step,
and then construct a Voronoi Diagram. The Voronoi polygons are then used as surrogates for
the cells.
[beware of edge effects - we ask a trained observer to draw a large polygon (not necessarily convex)
which encloses the “good” portion of a field of view. The Voronoi Diagram involves ALL cell centers found,
but subsequent analysis uses only the Voronoi polygons which lie entirely inside the “good” region. In simpler
cases, all you need to do is to shrink the field of view and analyze only the central region.]
In normal tissue, this works great. In diseased tissue, there are issues that are typically resolved by
having someone closely examine the results to be sure that the Voronoi Diagram is truly representative of
the actual cells. In some cases (we haven’t gotten there, yet) it *might* be necessary to actually
trace the cell boundaries (and use a somewhat more complicated data structure to hold the result).
If you do this, take care to identify the CENTERS (defined in a fairly loose way) of the cells - this is not
necessarily the same as the positions of the nuclei. Be especially careful of measures such as “number of neighbors”,
a measurement which is very sensitive to the precise locations of the Voronoi “sites”. Do periodic, systematic reviews
of the Voronoi Diagrams and the original images to be sure that the VD is capturing what you think it is.
Cell density is no problem. Mean values of cell area or distance to neighbors are no problem, but there may
be some noise in variance. # of neighbors requires extra care (but is so powerful that it’s worth the effort).
If all you need is cell density, you can be much more tolerant of low-precision localization of the cell centers, and
simply detecting nuclei will be fine. Only…beware of making this decision and THEN LATER having someone
decide to use the VD for more sensitive measurements. Sometimes, the second level of analysis conveniently forgets, or ignores, or is simply unaware of, the limitations imposed by the initial data collection. Don’t over-interpret!
--
Kenneth Sloan
[hidden email]
"La lutte elle-même vers les sommets suffit à remplir un coeur d'homme; il faut imaginer Sisyphe heureux."
On May 21, 2014, at 08:54 , Unruh, Jay <
[hidden email]> wrote:
> Note that for many confluent cell layers, the cells extend beneath one another to a fairly significant extent. The "boundary" will underestimate the area by a fairly significant amount. Given this uncertainty and the difficulty in segmenting the boundaries anyway, you are probably better off simply using the distance to the neighboring nuclei as an estimate of cell size. Voronoi analysis will give you a slightly more sophisticated version of this.
>
> Jay
>
> -----Original Message-----
> From: ImageJ Interest Group [mailto:
[hidden email]] On Behalf Of Cammer, Michael
> Sent: Wednesday, May 21, 2014 8:49 AM
> To:
[hidden email]
> Subject: Re: Segmentation of confluent cells
>
> When I mentioned cell-cell junction proteins, I was thinking of something like e-cadherin.
> The red in
https://www.flickr.com/photos/mcammer/2234624761/>
> ===========================================================================
> Michael Cammer, Microscopy Core & Dustin Lab , Skirball Institute, NYU Langone Medical Center
> Cell: 914-309-3270 Lab: 212-263-3208
>
http://ocs.med.nyu.edu/microscopy &
http://www.med.nyu.edu/skirball-lab/dustinlab/>
> From: ImageJ Interest Group [mailto:
[hidden email]] On Behalf Of Matthew Pearson
> Sent: Wednesday, May 21, 2014 7:40 AM
> To:
[hidden email]
> Subject: Re: Segmentation of confluent cells
>
> Hi guys,
>
> Thanks for your responses. I'll look into region growing algorithms used in ImageJ. I now a least know that this isn't particularly straight forward. I'll make sure i'm there for the image capture so the images can be as good as possible and we'll take it from there.
>
> cheers,
>
> Matt
>
> ------------------------------------------------------------
> This email message, including any attachments, is for the sole use of the intended recipient(s) and may contain information that is proprietary, confidential, and exempt from disclosure under applicable law. Any unauthorized review, use, disclosure, or distribution is prohibited. If you have received this email in error please notify the sender by return email and delete the original message. Please note, the recipient should check this email and any attachments for the presence of viruses. The organization accepts no liability for any damage caused by any virus transmitted by this email.
> =================================
>
>
> --
> ImageJ mailing list:
http://imagej.nih.gov/ij/list.html>
> --
> ImageJ mailing list:
http://imagej.nih.gov/ij/list.html--
ImageJ mailing list:
http://imagej.nih.gov/ij/list.html