Help with adipose tissue analysis

Hi, I have been analyzing adipose tissue by hand and came across the Cell profiler program the other day. I was hoping someone could help me figure out which modules I need to analyze the two types of adipose tissue. I have attached a few images to show you an idea of the images I have been trying to analyze. For the WAT I would like to obtain the number of cells in the image and the width of each cell. Is it possible to get a width or area for each cell or does this program just give an ave cell area for the image? I would like to get the cell width so I could also measure the line for 100um in the image and the extrapolate how many um each cell width is.
The BAT is a little more difficult to analyze because the cell size varies. If possible I would like to count how many of six or so different cell sizes are in an image and then make a histogram from that. That is take the largest cell size in an image and one of the smallest and a few others in between. I would like to use those same cell sizes to analyze all the BAT images. If anyone has some knowledge of cell profiler and could help me develop a pipeline to measure these images that would be very helpful. I have about 500 images in all to measure and would like to make everything consistent.







Hi,

Your images were challenging, but I think I have a means to handle them. Fortunately, the same general approach seems to work on both WAT and BAT images. I’m attaching the pipelines to this post.
For the WAT, the strategy is the following:

  • After loading the images, resize for memory concerns; the full resolution doesn’t seem like it’s needed, but remember to scale the results appropriately.

  • Get a single channel from the color image, picking the one the has the highest contrast (at least, by my eye)

  • Invert the image, use top-hat filtering to enhance the cell borders and reduce the intensity variation within the cell, then invert back

  • Identify the cells, using thresholding so the walls separate the cells. Adjust the declumping settings to try to further segment the cells from each other even if the walls do not. This means that the cells will be chopped up as well, but that is remedied in the next step.

  • Unify objects that are adjacent and the intensity between their centroids is sufficiently high. This is to rejoin the improperly segmented pieces of the cells back into their wholes.

For the WAT, much the same applies:

  • After loading/resizing/color splitting, use feature suppression to smooth things out a bit.

  • Identify objects using settings appropriate for the smaller objects. The thresholding is not the key thing, but use the declumping settings to segment the smaller items appropriately is important. Larger objects will also be carved up but that is remedied in the next step.

  • Again, unify objects to rejoin the cell pieces.

  • Classify the resultant objects on the basis of area.

Keep in mind that I’m picking the color channel for contrast. If the color balance changes in the acquisition of new images, you may need to adjust this choice.

Again, for both of these, some additional adjustment of the settings is probably necessary. However, after working on this, I think I can say that whatever adjustments you make, you are looking to optimize the following:

  • Partial declumping of the cells in IdentifyPrimaryObjects, to get the proper larger scale segmentation (i.e, between cells)
  • Refinement in ReassignObjectNumbers to resolve the improper smaller scale segmentation (i.e, within cells)

Hopefully this gets you started!
2010_06_18_WAT.cp (7.6 KB)
2010_06_18_BAT.cp (7.64 KB)