Cellprofiler puncta identification massive false positive

I have two images for puncta analysis, one at the 0 min and one at the 60 min using this pipeline, puncta.cpproj (442.7 KB) , the 0 min mark produces an incredible amount of false positives, while the 60 min produces none. I can’t see the differences between the image backgrounds or why there would be such a false positive rate in the 0 min mark.
Additionally, after using the enhance feature in CP it works for the 60 min and beyond images, but for the 0 min image it still leaves behind artifacts, as seen in the 0 min gamma reduced image. Is there any way to ignore those?
0 min original

0 min image marked

0 min gamma reduced

60 min original

60 min image marked

I don’t know much about CP, but usually in a time series behavior like that is due to bleaching changing the necessary threshold over the time course. Without some sort of adaptation (time series based adjustment) it is difficult to set a single threshold that will work for the first and last images. While it might not be visible looking at the picture, I would recommend checking the exact pixel values in those areas, in the channel where you are searching for puncta. Hard to say more without seeing the original first frame without the spots in the way (maybe share the raw data for those two frames only?).

It does appear to picking up objects that should have been thresholded away. I’ll update the OP with the images

It does look like whatever processing your are putting the image through prior to spot detection is strongly enhancing edges (* and variation), as most of the scattered points in the 0min image are around the borders of the clusters of cells. If that area/background bleaches first, that could lead to the reduced detection of such points in later frames.

I think that further help will require someone who can use CellProfiler, good luck!

Hi @mdl54

Your pipeline is very disjointed. I created a much simpler version that will do what you want. I also added puncta per cell (as puncta identified not in a cell is not real).
Use the enhance speckles function to clean up your images (see below)

Identify cells

Identify Puncta

1 Like


puncta.cpproj (647.3 KB)


1 Like

Thanks very much, this is helpful.