Nucleii + micro nucleii identification

greetings
i am somewhat new to cell profiler and i am using the software to get an idea as to what would be the best approach to identifying nucleii and micronucleii from some images i have been given. The images are of some stem cells which have been treated such that they binucleate but do not seperate, and later i will have some images which will contain micronucleii which will be of the cells after they are irradiated. I have only used the basic pipeline approach at first and i’ve tinkered around with some stuff and loaded other pipelines others have put up on the forum and in the example pipelines given on the website but to little success. I am as you can see somewhat stuck and any help and advice would be really appreciated as this is for my final year project and the further i get the better.

the pipeline i used for the attachment was incredibly basic, just load image and identify primary object on default settings, otsu global etc etc.
basic.cp (3.13 KB)


Hi,

The detection is getting confused because there are 3 levels of intensity: dim (background), middle (cytoplasm) and high (nucleus). This will probably become apparent if you right-click on the upper-left panel (the original image), and select Image contrast > Log normalized.

I would suggest changing the Otsu thresholding from 2-class to 3-class with the middle class set to Background. This should get the detection you want, and then you can work on segmentation.

Regards,
-Mark

i used your changes and it slightly improved the accuracy from # of identified objects being 6 as opposed to the previous 10. I then altered then altered the threshold correction factor until i got it to identify only 2 nucleii. This is all pretty simple of course and is in no way a move towards being able to dynamically assess a picture and each time accurately quantify the varying number of objects within. Using the same settings i applied it to a different picture to see what would happen and of the 5 nucleii within it only registered 2. So i now need to find a method to do this dynamically. Any advice?




if no ideas could you at least point me in the right direction?

Hi,

I’ve recreated the problem of the same settings not working for the two images, even though they look similar. The control image has the expected behavior for Otsu 3-class thresholding with minimal setting adjustment, whereas the 200p_2 does not. So I have a few questions to see if I can sort this out:

  • Were both images acquired with the same microscope settings?

  • Typically, we recommend that images be acquired as TIFs which is a lossless format, as opposed to JPGs which lose information. You might using re-catpuring the image as a TIF, or if it’s already a TIF, not saving as a different format and reposting them to this forum thread.

  • I noticed that the 200p_2 seems to have some weirdness associated with it which is apparent on rescaling the intensity (see posted image), i.e., the cropped edges plus stippling across the image. Were these image modified in any way after acquisition?

Cheers,
-Mark


Were both images acquired with the same microscope settings?

i enquired and yes they were?

Typically, we recommend that images be acquired as TIFs which is a lossless format, as opposed to JPGs which lose information. You might using re-catpuring the image as a TIF, or if it’s already a TIF, not saving as a different format and reposting them to this forum thread.

i will retry with said tif files, this was probably the source of the error

I noticed that the 200p_2 seems to have some weirdness associated with it which is apparent on rescaling the intensity (see posted image), i.e., the cropped edges plus stippling across the image. Were these image modified in any way after acquisition?

they were not besides being converted to jpg.

i’m gonna start again because i had to get a new set of photo’s due to the old ones being corrupted, they shall be tif’s this time however.

any advice on where i should start for tackling this problem, i need a way of being able to identify nuclei and micro nuclei for a large sample size and be able to tally the results.
issues that will arise are nucleus size, varying cytoplasm values, varying nuclei intensity, nuclei will be stuck together as they have had demecolchine introduced into the growth medium.

i’m incredibly new to all this and could use some real help.

Would you consider the 200p 2.jpg to be representative of your image set? If so, we could start with this one rather than the other which was apparently corrupted. If this image is acceptable, let me know which nuclei you would want to identify as what.
-Mark

the 220p 2.jpg was a standout example of a case where 2 nucleii have not seperated due to the colchicine i was using it as an example only.
i will add some images to this message which have more varied and seem more representative. In each image i would want to identify between nucleii and micronucleii.

as you can see the micro nucleii seem to be both small and appear quite brightly, so i’m guessing these are the factors that need to be taken into account in order to determine the difference between them and regular nucleii

and also you can see nucleii paired together and not separated which could cause problems in terms of determining them as two different entities. I hope this has been clear, any other questions ask and i’ll try to get back as promptly as possible

thankyou :smiley:



Hi Tom,

These images are much better. Attached is a pipeline which will should get you started. Look at the module notes at the top for my thoughts.

Cheers,
-Mark
2011_01_13.cp (10.4 KB)

Hey Mark,
I cannot thank you enough for this, it works very very well. Is it possible to identify an object based on 2 properties, in this case intensity and size. e.g. if said object is above this brightness and lower than this width then it is a micro nuclei. and vice versa for regular nuclei. I’m presuming the best idea would be to split the selection in stages and filter out the rest e.g

                                                                                   original picture
                                                                                 *filtering process*
                                                               dark ones                                       bright ones
                                                        *filtering process*                              *filtering process*
                                            large dark ones   small dark ones       bright large ones              bright small ones
                                                                              classify these 4 objects
                                                  nucleus             nucleus             nuclei stuck together            micronuclei

which i’m assuming would just require adding and removing a few modules.

thanks again and let me know your thoughts

regards Tomb

Hey,
I managed to get the cells to be isolated by 2 metric variables (intensity and area), however i thought of a better way in which i could isolate the binucleated cells.

Is it possible to classify how many cells by the number of nucleii that lie within the cytoplasm? e.g. if 2 nucleii within a cytoplasm = binucleated etc

what would be the best approach to do this?

If you observe this image under log normalised you can see that there are clearly 3 cells with bi-nucleation occuring within. So how would i go about classifying them?

I’ve attached a pipeline which attempts to identify and classify the binucleated cells.
-Mark
2011_02_02.cp (10.4 KB)

thanks for the help mark,

i attempted to use the pipeline for some other images taken from the same sample and it did not work.Can you think why this would be?, i’m assuming it had something to do with the way in which it recognises segmented nuclei based on shape and that its just a case of finding a way to get the pipeline to auto-adjust accordingly for each image given to it.

any suggestions, i attached some pictures for an example.

regards Tom





If you log-scale the 27.tif image, you can see that there is a stripped abberation down the image, which I assume is an acquisition issue.

If you omit the IdentifySecondary step, you can still do the RelateObjects, ClassifyObjects and FilterObjects step with the UnifiedNuclei rather than the Cells. However, the problem you are running into is how to distinguish the binucleated nuclei from two distinct nuclei that are merely touching. In this case, you will have to think about what makes these two cases different in a qualitative sense, and then consider which modules can carry out this decision process. For example, if you have a bright nucleus touching a binucleated cell, a solution may be to:

  • Segment all nuclei, binucleated or not
  • Filter out the bright nuclei
  • Mask the image to remove the bright nuclei and re-identify the nuclei, this time without segmenting
  • Relate the segmented nuclei to the unsegmented objects, and classify the unsegmented objects according to the number of children (this is the same strategy as my prior post)

The key point here is what criteria would you use to distinguish the phenotypes of interest if you were doing this without software (i.e., manually)? If such a set of rules exist, CP may be able to replicate this process. If not, CP probably won’t be able to do it either.

Alternately, if this problem seems to be beyond a simple set of filtering steps, you can use the machine-learning tool in CellProfiler Analyst (CPA) to make the decision between these phenotypes for you. In this case, you should probably segment all the nuclei, include the ExportToDatabase module to write out the measurements plus a properties file for CPA to use. There is a CPA forum where you can post questions for help on this.

Regards,
-Mark

Hello mark me again, i was wondering if you could point me towards a method of illumination correction that would work dynamically for each image. The settings i have attempted are in the pipeline attached.

if you look at the images logarithmically you will see some lovely bright streaks across the image. It is different for each image.

I believe it is adding bias to my detection of bi-nucleated cells as it is altering the intensity readings.

please let me know what you think









and here are the rest
2011_03_14.cp (13.5 KB)



The streaks don’t seem to be particularly severe, at least not enough to alter detection. How exactly do you think they are affecting the “intensity readings?”
-Mark

in closely clumped cells it is sometimes obscure as to whether or not the cells are bi-nucleated or are simply cells close together. I feel the intensity smears may create a variance in intensity between 2 nuclei that may cause the pipeline to consider a binucleated cell, actually 2 cells.