Cell Profiler after Deconvolution

deconvolution
dapi
identifyprimary

#1

Hello,

I have a DAPI image which I did deconvolution on, this helped clear it up and the nuclei are a lot cleaner.
I am using CP to IDPrimary objects, however the identification is actually better before the deconvolution.
Does this make sense to you guys? I’m pretty puzzled about it, I’ve played around with the identification parameters but could not get the deconvolved image to be identified well.
Thanks for your reply :slight_smile:


#2

Hi,

In default IdentifyPrimaryObjects, there’s certain level of smoothing. So the segmentation algorithm would see a nucleus in the non-deconvoluted images as a whole (a bit blurry) object. With the same algorithms, the grainy crystal-clear complex structure of a nucleus in deconvoluted images will result in over-segmentation.

You can try to do either:

  • Use the original (non-deconvoluted) images to do segmentation, then overlay on deconvoluted images. OR
  • Use the deconvoluted images, but uncheck “Automatically calculate size of smoothing filter” and “Automatically calculate minimum allowed distance between local maxima”, and manually tune these parameters (recommend to increase) .

Hope that helps.


#3

Hi Minh, Thanks for the quick response.
Unfortunately playing with this parameter does not improve the detection enough, the cells are still over segmented.
Perhaps you have another idea?
I’m also using IDPrimary for speckle detection, and here also, the deconvoluted images are not recognized well by the software (tried playing with the smoothing filter and min. distance).


#4

Can you then give us some sample image (w/ and w/o deconvolution). I’ll see if I can help.

Thanks.


#6

Hey, I’ll be happy to but I can’t seem to be able to upload them here.
Is there an email address to which I can send it?


#7

You can zip them and post to a google drive or dropbox. Then send us the link with open authorization.

Let us know if that’ll work for you.


#8

Good idea, here it is:

https://drive.google.com/open?id=0B1SU45fC1TsWRS1Ja2E0MlRMSXc


#9

Hi,

Please try this segmentation.cpproj (636.9 KB)
on other deconvoluted images to see if you get somewhat acceptable results.

I guess you also have to do IlluminationCorrection, since the right size of the image is a bit dimmer.

Hope it helps