Question about LoG object identification


#1

Hi,
I’ve been experimenting with using CellProfiler to automate our FFA (fluorescent focus assay) cell counting. In this assay, GFP-labelled antibodies attach to the surface of cells if certain virus genes are expressed in the cells. From the examples, it seems like the best way to count the cells is to use a DAPI stain first, but we’re kind of locked in to a test method without it. Thus, I’m trying to use the IdentifyPrimAutomatic to detect cells by their membranes, not their nucleuses.
Among the options for IdentifyPrimAutomatic, I noticed a Laplacian of Gaussian (LoG) counting option.
The inputs listed are size of neighborhood (height, width), Sigma, Minimum Area, Size for Weiner Filter (height, width), Threshold.
There is no explanation for this function or its variables within the help function. I assume Sigma is the analog of the Sigma for the LoG edge-finding function, and that threshold is the same as elsewhere, but I am unsure of how to handle the other variables (neighborhood size, minimum area, weiner filter size). For a starting point for those variables, should I use some function of object size, or average distance between objects, or what?

Thanks,
Russ


#2

To further my question, I have been experimenting with putting random variables in to the LoG identification function within the IdentifyPrimAutomatic module, trying to see if I could get it to work, with the following inputs:
200,200,6,100,100,100,.105
ie Size of neighborhood(height, width): 200,200
Sigma: 6
Minimum Area: 100
Size for Weiner Filter(height,width): 100,100
Threshold: 0.105

The IdentifyPrimAutomatic function would not finish, and I got the following error:

Error Dialog:
There was a problem running the analysis module IdentifyPrimAutomatic
which is number 02. Error using ==> IdentifyPrimAutomatic at 896
Image processing was canceled in the IdentifyPrimAutomatic module because
no DAPI regions were generated.

I never called anything DAPI within the pipeline. Does this only work with the DAPI stain? If something doesn’t look exactly like a nucleus, does that cause the error I got? Or is it something else, like my variables being out of whack?

Thanks for any insight you can give,
Russ


#3

Hi Russ,
It is not typical to use LoG in CellProfiler.

Somehow, we have forgotten to site the original author in the current release of CellProfiler. This issue has been resolved and will appear in the help for all future releases of CellProfiler.
The citation will appear as follows:
% Laplacian of Gaussian method:
% This is a specialized method to find objects and will override the above
% settings in this module. The code was kindly donated by Zach Perlman and
% was used in this published work:
% Multidimensional drug profiling by automated microscopy.
% Science. 2004 Nov 12;306(5699):1194-8. PMID: 15539606
% Regrettably, we have no further description of its variables.

However, you will receive that particular error message if your threshold is too high.
To try to further explain these variables:
-Neighborhood size refers to the size of the neighborhood used to create the LoG filter.
-Sigma: standard deviation of the LoG kernel.
-Size of Weiner filter: the block size to use to reduce the effect of noise.

Hope this helps.
~martha