Identify 3 nuclei types

Hi, I have just begun using CP so I have downloaded the latest version of CP (1.0.7522) running on a Mac Pro at home (where I’m developing the pipeline) and will be running the pipeline eventually on a PC running Windows XP. The images are from an ImageXpress Micro (epifluorescent microscope). The object is to count cells in a population (8 images to cover entire well) that fall in to the following 3 different phenotypes:

  1. unchanged, normal nuclei
  2. mitotic nuclei (higher intensity nuclei but not bar-shaped because treatment causes breakage of DNA into high intensity spots within the nuclei)
  3. multinuclei (intensity typically low like normal nuclei but borders of several nuclei hopefully can be traced within the outer nuclear boundary)

Here is an image containing a mix of these cells.

This image has all three cell types, hopefully I’ve described them well enough for you to pick them out. I couldn’t create good images for each of the 3 types using the software I have at home.

Here is the pipeline that I’ve been working on, everything up to the PauseCP step.

I basically feel pretty good about defining the outer nuclear boundary for all of the cells in the first IdentifyPrimAutomatic. Then I thought that I could Crop so I can identify each cell and then associate (children, parent) further objects within each nucleus as either being mitotic or multinuclear. If there are no further objects found in the nuclei (or 1) then it would be counted as a normal (planning to use CellAnalyzer for final counting).

So, I’m able to get a good outline (and measurement) of mitotic using a threshold within the second IdentifyPrimAutomatic, but can’t figure out best way to get the multinuclear population outlined. I had an idea that I could expand the mitotic until it reached the background intensity and then exclude these nuclei from the next IdentifyPrimAutomatic but it is not working and I feel this is not the correct approach anyway. The tough thing is that the intensities of the 2 phenotypes are so different that I assume I need to deal with each one separately. With your expertise, I’m hoping that you can steer me in the right direction but giving me some potential module/pipeline steps. I have a short timeline on developing this so I’m hoping that I can get through this and devote much more time in understanding the nuances to this flexible analysis tool. Thank you for your help! And please feel free to send private messages to me if you need more information.

One thing I’ve discovered is that the background is currently very high and the outer boundary of the nuclei is only 1.5 or so above background. We are going to work on trying to modify final media conditions to bring that down (to improve the outer nuclei detection). The general pipeline development is still valid using this image though!

Mary West
ExampleShannonPIPEspecklNoIllcorr_expand_excl.mat (2.01 KB)

Hi Mary,

Unfortunately, the TIF of the image didn’t come through. Could you try to either (a) upload as a different image type (like PNG) or (b) post it on Google Picasa or something similar?


Hi Mary,

Attached to this message is a pipeline that may serve as an initial attempt to do what you want. I had to do some guesswork based on your desdrptioon, since I’m not familiar with the phenotype. Basically, it works in the following way:

]Identify AllNuclei in the manner you had before/]
]Use top-hat filtering via SmoothOrEnhance to highlight the high-intensity speckles. Crop the filtered image into the AllNuclei objects, and identify the speckles using per-object thresholding. RobustBackground is good for detecting objects several deviations above a mean (e.g, very high intensity peaks) in a consistent way./]
]Relate the Speckles to AllNuclei. Use FilterByObjectMeasurement to obtain nuclei with 1 speckle or more. These are either normal or multi-nucleated/]
]Crop the original image into the NormalOrMulti nuclei. Identify Lobes within the cell with per-object thresholding./]
]Relate the Lobes to the NormalOrMulti nuclei. Use FilterByObjectMeasurement to obtain nuclei with 1 lobe or less. These are normal nuclei./]

Naturally, the part that needs the most tweaking is the 4th step, identifying multiple objects within the cell to get a sense of the number of lobes. The size of the filter in the 2nd step may also need adjustment.

I should mention that I removed the RescaleIntensity module from your pipeline for the image that you uploaded. You may need to put it back in for your raw data.

Also, this is the type of project that would work well with our machine learning tool, CellProfiler Analyst. With it, you can classify example nuclei into phenotypes of your choosing, and allow the software to determine what measurements are required to decide between them. The software is due for release soon, but perhaps not early enough for your needs. Stay tuned, though…

2009_10_13_PIPE.mat (1.61 KB)