Viewing positive intensity correlations on image

Is there a way to visualize a particular object within an image that has a positive intensity correlation. For example, can I “highlight” a particular nucleus that has a positive intensity correlation value? If so, can I do this in CellProfiler, or do I need to use CellProfiler Analyst? And HOW?

Thank you,

Hi Anna,

To answer this question, you would need to describe what the correlation is relative to on a per-cell basis. CellProfiler has a MeasureCorrelation module that might do what you want, but it requires two images to compare. Does that sound like it would suit your needs?


Hi Mark,

I’m attaching my current pipeline and a couple of images. As you can see from the pipeline, I am using that very module (MeaureCorrelation). One of the handy outputs that I obtain from this pipeline is a spreadsheet with correlation values for each object in the image. In Cell Profiler Analyst, I can open an image from ImageViewer, then choose “show cell numbers” under “View”. Ideally, I’d like to be able to “highlight” those cells in that image which have positive correlation values (showing nuclear translocation).

You may wonder why I’m only using one measurement in my pipeline, even though there are 3 suggested in the “Image-based screening using subcellular localization of FOXO1A in osteosarcoma cells: A computer exercise using CellProfiler & CellProfiler Analyst software” paper of which you are an author. I was not able to put together a good enough parameter setup for the identifying the cell body as a “secondary object”. The closest I got was Propagation, Manual, Regularization factor of 0.02. Since I’m primarily interested in nuclear localization, I figure that the one measurement is OK, though I could be mistaken (wouldn’t be the first time).

Thank you for the help.


03-27-13_Nuclei_onlyoutputtestMOD.cp (8.69 KB)

One way to do this in CellProfiler without CellProfiler Analyst is to use ClassifyObjects to classify the cells according to positive and negative correlation values. It then gives you a color-labeled image which you can later save. Alternately, you can also use DisplayDataOnImage to display the per-cell correlation values overlaid on top of each cell; this image can also be saved later as well.

Yes, I see from your GFP image that the staining seems to capture more of the cell wall than the cell body. Two suggestions I have are the following:

  • One approach is to use ImageMath with the GFP as input and “Invert” as the operation to invert to the pixel intensities so that light pixels are dark and vice versa. You may have better luck with the inverted image as input into IdentifySecondaryObjects than the original image alone, and maybe changing the thresholding method to Otsu global with 3-class thresholding with the middle class set to foreground. Of course, this is only recommended if the stain remains close to the walls in all cases, for all treatments. If it shifts location substantial, then this approach may fail for those cases. I’m just going on the basis of the one GFP image you posted.
  • Something mentioned in the written exercise (but not dwelled upon) is that you can use the Distance-N method in IdentifySecondaryObjects to impose a cell body in cases where the cell stain is absent, fluctuating or indistinct. If you provide a value of N that captures most of the cell body area, this can act as a surrogate and you may still get good results re: assessing translocation.

Hope this helps!

Hi Mark,

This is what I got when I tried “ClassifyObjects” (see attached). Do you see anything that could be optimized? This was a great suggestion. Thank you very much!


What you have shown is basically what I would do. The final output is a per-image percentage of what falls into what category, plus a per-object measurement stating which bin it ended up in. Does this fit your needs?

This does fit my needs.

Thanks so much!