Using Trainable Weka Segmentation to segment cfos-ir cells in an image


I am trying to do segmentation on several images of brain tissue that has gone through immunohistochemistry (IHC) for cfos in order to segment out the cfos cells to do a cell count.

I have cropped the image to contain only the brain region that I am interested in.

Two such images are shown.

As you can see from the two images, one is brighter than the other and in each image, some of the cfos cells are really dark whereas others are much lighter. Additionally, the background is not completely uniform. All things that might make segmentation more difficult.

Also shown is a small section from the first image with the cfos cells outlined.

I was working with the Weka in the hopes to segment the images, but I am running into problems being able to count both dark and light cells, and especially having a weka model that works on both darker and lighter images.

I tried preprocessing the images by duplicating the image, performing a gaussian blur with sigma = 5, and then subtracting the two images before doing weka. This seemed to work better, but then I had the obvious problem with the Weka thinking that the edge of the brain slice are cfos cells.

So, I am hoping for some suggestions on how to use the weka for this segmentation, what training features to use, preprocessing suggestions, or even a different application/approach altogether.

Thank you so much.

With Image J , I find this.

that looks very nice!
What did you do in ImageJ to result in that image?
And can you apply the same technique to the other image and get consistent results?

Hi Lydia
The second image gives this!


That looks pretty good! a few faint cells arent counted but nothings perfect.

What did you do to achieve this?

Here is the main part of the macro.
run(“Duplicate…”, “title=1”);
run(“Duplicate…”, “title=2”);
run(“Gaussian Blur…”, “sigma=3”);
run(“Convert to Mask”);
run(“Analyze Particles…”, “display summarize add”);
roiManager(“Show All without labels”);

Sorry but I do not know how to put the macro “active”!
If anyone could inform me by a tutorial?

Perform some tests and please report.

the macro works fairly well on the two images I had already provided, but when I moved onto other images, it didnt perform as well (see below)… This is one of the reasons I was hoping the weka would be more helpful than a standard threshold.