Colonies: partial overlap treatment?

Hello,
I am at this point interested in colony counting.

A) Is there a way (already coded in some module), to correctly numerate colonies, in the case there is some overlap ?

B) I encounter a case where colonies give more or less nice circular spots, but with many occurences of partial overlap (the eye is able to grasp then that two or three circular discs are partially overlapping). Yes, I admid that the student who did the plate did a bad job (too many clones spread). Is nevertheless CellProfiler able to cope gracefully with this ?

C) I read in the doc, that the jpg pictures should be avoided; this is nevertheless the only format I can get from a small digital camera. Do you have recommendations with respect to the resolution of the image ?

D) What are the units used in the image treatment output: are they pixels ?

E) Would there be some module that would take a sample calibration image ( for example taking with the same camera that take the pictures of the colonies, a calibration image consisting of a laser-printer printout of an appropriate calibration image); then this info would be used by CellProfiler to give the correct scales on the subsequent images ?

Thanks !

For this situation, if the colonies overlap, you would probably need to adjust the appropriate settings in IdentifyPrimAutomatic (if you’re using CellProfiler 1.0; it’s called IdentifyPrimaryObjects in CellProfiler 2.0). The settings labeled “Method to distinguish clumped objects” and “Method to draw diving lines…” both deal with segmentation, which is process of separating objects that are joined/overlapping.

The same answer as above applies here too. But to be more specific. since the colonies are circular, you may have good luck with the declumping method set to “Shape” and the dividing line method set to “Distance” (which I think it already is the pipeline). However, if the area of overlap between adjacent colonies is marked by a dim line dividing the two, using “Intensity” for both may also work.

There’s no precise answer to this one. Basically, you would want to go with the resolution that gives you sufficient results for segmentation (object separation) but not so big that CellProfiler chokes due to memory :smile: .

That is correct, for length-based measurements.

No module does this specifically. However, if you identify an appropriate feature in this calibration image (for example, a white scale bar on a black background) and measure it with MeasureObjectSizeShape, you can obtain the length which can be used as a scaling factor for your other measurements offline (that is, outside of CellProfiler).

Hope this helps!
-Mark

Hello,

[quote] wouldbeuser wrote:Is there a way (already coded in some module), to correctly numerate colonies, in the case there is some overlap ?

For this situation, if the colonies overlap, you would probably need to adjust the appropriate settings in IdentifyPrimAutomatic (if you’re using CellProfiler 1.0; it’s called IdentifyPrimaryObjects in CellProfiler 2.0). The settings labeled “Method to distinguish clumped objects” and “Method to draw diving lines…” both deal with segmentation, which is process of separating objects that are joined/overlapping.[/quote]

O.K., I will try.

[quote] wouldbeuser wrote: I encounter a case where colonies give more or less nice circular spots, but with many occurences of partial overlap (the eye is able to grasp then that two or three circular discs are partially overlapping). Yes, I admid that the student who did the plate did a bad job (too many clones spread). Is nevertheless CellProfiler able to cope gracefully with this ?
The same answer as above applies here too. But to be more specific. since the colonies are circular, you may have good luck with the declumping method set to “Shape” and the dividing line method set to “Distance” (which I think it already is the pipeline). However, if the area of overlap between adjacent colonies is marked by a dim line dividing the two, using “Intensity” for both may also work.[/quote]

O.K.

[quote] wouldbeuser wrote: I read in the doc, that the jpg pictures should be avoided; this is nevertheless the only format I can get from a small digital camera. Do you have recommendations with respect to the resolution of the image ?

There’s no precise answer to this one. Basically, you would want to go with the resolution that gives you sufficient results for segmentation (object separation) but not so big that CellProfiler chokes due to memory :smile: .[/quote]

O.K.,

but it would be nice to have a dedicated module that would do the resampling of the image !
Typically it would be nice to be able tu use the digital pictures “as is” from a camera, without having to mess with the resolution; I would find also more proper to archive high-resolution images (for eventual later re-treatment), and have the pipeline able to disgest those high-resolution images by a first resampling module (resampling to a user-chosen lower resolution) !
Could you plan it / do it ?

[quote] wouldbeuser wrote:What are the units used in the image treatment output: are they pixels ?
That is correct, for length-based measurements.[/quote]

O.K.

[quote] wouldbeuser wrote:Would there be some module that would take a sample calibration image ( for example taking with the same camera that take the pictures of the colonies, a calibration image consisting of a laser-printer printout of an appropriate calibration image); then this info would be used by CellProfiler to give the correct scales on the subsequent images ?
No module does this specifically. However, if you identify an appropriate feature in this calibration image (for example, a white scale bar on a black background) and measure it with MeasureObjectSizeShape, you can obtain the length which can be used as a scaling factor for your other measurements offline (that is, outside of CellProfiler).[/quote]

O.K.

The module you long for exists, and is called Resize.

Cheers,
David