Help with pipeline for counting nuclear foci (speckles)

Hello world :smile: .

I’d like to ask for some help on how to count nuclear speckles (gamma-H2A.X foci)
I have worked with the Speckle counting example pipeline and made it work, but I’m not satisfied with the segmentation of the foci.
Images were taken with a Leica TCS SP5 confocal microscope and a 63x/1.4 oil immersion objective

The problem is that a number of cells have a rather diffuse staining where the foci are not so well defined. The pipeline as it is now (attached below) in some cells detects speckles which I would say are not existing and on the other hand fails to detect some that are clearly present (pics of example cells are attached, white arrows highlight false positives, yellow arrows point to false negatives).

I’ve played around with the IdentifyPrimaryObjects module for the speckles:

  • Increasing object diameter didn’t help

  • I’ve increased the threshold correction factor, which helped a bit, but with further increase I get more false negatives

  • For detection of clumped objects I tried settings Intensity and Laplacian of Gaussian and the latter seems to work better

  • I didn’t manage to optimize the segmentation by adjusting the smoothing filter size and maxima suppression distance

Any help regarding which parameters could be changed in which way to improve foci detection would be greatly appreciated.
Some original full-size images to work with I’ve also attached as “Input images.zip” (Hoechst DNA stain in the blue channel has ch01 in the file name, the green channel with the gamma-H2A.X signal has ch00 in the filename).

Thank you very much in advance
Dominik
Input Images.zip (962 KB)



ds_gH2AX-foci_use me.cp (14.1 KB)

Hi Dominik,

The pipeline you posted doesn’t seem to give the same results as in the screenshot. Are you sure it’s the same one?

In any case, a couple of suggestions in the IdentifyPrimaryObjects module that identifies the speckles:

  • The false negatives might be falling below the size criterion. You may want to adjust the lower size limit downward.
  • Increasing the lower threshold limit from the default value of 0 might take care of the false positives

Regards,
-Mark

Thank you very much for your answer, Mark.

[quote=“mbray”]Hi Dominik,

The pipeline you posted doesn’t seem to give the same results as in the screenshot. Are you sure it’s the same one?[/quote]

It’s indeed not the same one, sorry for that. I tried so many things and saved the different pipeline versions that I posted the wrong one. The one that was used for the example images had a different size criterion in the IdentifyPrimaryObjects module for speckles: typical diameter of objects was 4-35 pixel units instead of 5-35 as in the posted pipeline.

[quote=“mbray”]In any case, a couple of suggestions in the IdentifyPrimaryObjects module that identifies the speckles:

  • The false negatives might be falling below the size criterion. You may want to adjust the lower size limit downward.

  • Increasing the lower threshold limit from the default value of 0 might take care of the false positives

Regards,
-Mark[/quote]

This already helped. I decreased the lower object size to 3 pixel units and the lower threshold limit to 0.15 and the identification appears to be better. I need to test the pipeline on some more images, though.