Signal vs background



I’ve been trying to measure different signals (calcium, mitochondrion) vs. the background. However, although the signal is quite clear to the eye, there is barely any difference if I look at the mean intensities on the spreadsheet. Could it be that too much signal is accounted for the background?
On another note, in the attached image you can see how I identify the secondary objects. This is in concordance with the general approach to just draw a ring around the nuclei and measure the signal. However, as the calcium is only presented on one site of the cell and not surrounding it, I’m also measuring a lot of background which weakens the value of the signal. I haven’t found a more suitable algorithm to identify the 2nd objects… My questions is, is there a way to only select a part of the ring for measurement (e.g. highest 70% of the pixel) or can I somehow make the indentification algorithm better?

Here is the pipeline: autoPipeU2Otreated.cppipe (27.3 KB)
and an example image:
Any help would be highly appreciated




Can you check if the UpperQuartileIntensity would reflect better the observation? You’re right that if you include a lot of background in the ring, the mean value would not be that helpful.

[UpperQuartileIntensity: The intensity value of the pixel for which 75% of the pixels in the object have lower values.]



Hi Minh,

thank you very much for your suggestion. The UpperQuartileIntensity is a bit higher, but the max background is still higher than the max signal… So I’m just wondering if I can approve the pipeline to better distinguish between the background and signal

Cheers, Anja


If you’re worried about picking up background, rather than using the Distance-N method (which wille xpand out no matter what) you could use one of the other methods in IdentifySecondary (Propagate, Watershed, Distance-B)- they’re less likely to pick up areas of background.


Hi Beth,

I’ve tried that but with those algorithms I’m getting even more background. If I have a lot of nuclei (with little treatment) the 2ndary objects are being identified like this:


In that case, if you wanted to pursue alternate segmentation methods I’d definitely do Distance B (to keep it from going out too far) and also set a minimum threshold; that’d allow you to let it propagate out if there is signal underneath it, but not move if it didn’t.


The problem is that my background is varying so much, that I can’t set a global threshold… If I set it the lower bound too low, the background is counted as signal, too high, I’m loosing signal to the background. The background seems to be changing from top to bottom on the plates


In that case, you probably want to add some background subtraction to your pipeline (at least upstream of the segmentation, you can then measure intensity on the original images); the help for the CorrectIlluminationCalculate and CorrectIlluminationApply modules are probably the best places to start in terms of which exact settings will work best for your purposes…