Wheat germ agglutinin staining

cellprofiler

#1

Hi,

I’m new to Cell Profiler, but am hoping to set up a pipeline to measure cell area of cardiac cells in heart tissue using wheat germ agglutinin stain (WGA). I’ve been able to manipulate some of the filters to remove background within the cells, but the objects that IdentifyPrimAutomatic is defining as objects are not what I’m seeing as outlined cells on the slide. Is there a way to manipulate the automated process since I’m hoping to use it to measure a large number of slides and would like to avoid manually outlining the cells myself?

Also, if I can solve that problem, I’d like to set up a second filter to remove cells/obejcts that are not of a specific dimension (like in longitudinal section when I’d like to measure cross section) before making the measurements. Is this possible, and can you recommend a module to work with for this?

Thanks so much!

Mike



Memory problem
#2

Hi Mike,

Have a look at the attached pipeline, and see if it works for your needs.

Also, in order to remove cells according to a criteria, you can use MeasureObjectAreaShape to the cells, and then use FilterByObjectMeasurement to remove cells above or below a particular feature range. The pipeline I’m attaching filters cells according to a specified major axis length, just as an example.

Hope this helps!
-Mark
2009_02_19_PIPE.mat (1.39 KB)


#3

Thanks for the reply! I fooled around with the pipeline some more and came up with the attached one, which works reasonably well as long as there’s not too much background. I filtered by eccentricity, which allows me to exclude cells that are less likely to be in cross-section (less round).

I’m in the process of working on a few other pipelines for image analysis of heart tissue, in you have any suggestions. One is to quantify tissue fibrosis (Masson’s Trichrome stain), in which areas of collagen deposition stain blue and cells red. The difficulty I’m running into is figuring out how to count fibrotic sections as objects in order to calculate the percentage of fibrosis per slide. The fibrosis ranges from a few sections of 3 x 20 pixels, to large areas of tissue (up to 30% of the total area), and thus it’s difficult to find an object identification method that works. If you have any suggestions, or know of another way to measure fibrosis without counting these areas as objects, please let me know.

Thanks again!

Mike
WGAPIPE.mat (1.64 KB)


#4

Re: additional pipelines - If you like, you’re more than free to post images of these other sections. We can take a look and see what can be done.
-Mark


#5

Thanks again for the help!

I’ve attached a few pictures of cardiac tissue stained with Masson’s trichrome, in which fibrotic sections are blue and the cells are red. My current method for quantification using ImageJ is to use RGB split to convert the image to greyscale; take the blue image, invert the intensity so that the fibrotic area (blue) is dark, and adjust the threshold to make it binary, then count the dark pixels; then take the red image and adjust the threshold likewise so that the tissue section is black, then measure pixel count. I calculate the fibrosis fraction by dividing the blue pixel measurement by the blue plus red. As a control, I then combine the two binary images (in different colors) to form the measured image.

The main limitation for automation (or even the manual method) is that there’s often a fair amount of blue pixels scattered throughout the cells, and the threshold often has to be fine-tuned for each blue image so that I’m not counting background as fibrosis. Ideally, I’d either like to include a filter that puts a lower limit to the number of pixels that can be counted as ‘fibrosis’ (say, at least 10 - 15 pixels in contact with each other) and then measure pixel count for the whole image, or a method of counting continuous dark pixels as an object, and the calculating the area of this object. But if you have any other ideas, I’m all ears…

Thanks again!

Mike





#6

Hi Mike,

Attached is a pipeline which attempts to contrast enhance the separation between the red and the blue, tentatively identifies the blue areas and then expands those areas to try to capture the full extent of the blue, and then measures it.

A few notes:

  • The threshold in IDSecondary may need some adjusting. Setting the object fraction to 0.4 seemed to be optimal for the images you gave, but you may need to play with it some more.

  • I used RobustBackgroundGlobal for IDPrimAuto since it usually acts the best when you have no foreground to identify (i.e, no blue), but if that doesn’t work, you can always adjust the lower threshold bound upwards.

  • I used IDPrimAuto a 2nd time just to generate a set of outlines for saving, nothing more.

Hope this helps!
-Mark
2009_02_20_FibrosisPIPE.mat (1.53 KB)


#7

That’s great! I’ll try it out on a few different levels of fibrosis to see if I need to adjust the threshold (for example in other sections with < 5% fibrosis), but at least for those slides, it seems to work well.

Thanks so much!

Mike


#8

Hi,

Thanks so much again for the help with this fibrosis pipeline. One issue that I wanted to ask that I seem to be having trouble with in this pipeline is that it’s also measuring white background as blue. Presumably, this is because following the RGB split white is 255 for all three, but I was wondering if there is a way to filter this color out to avoid measuring it as fibrosis?

If you can help, I would greatly appreciate it.

Thanks again!

Mike