Need help for shape analysis for cells stained with anti beta catenin antibody

2016.09.26 - 488-beta-caenin.tif.tif (768.2 KB)

Hi guys,

I have been struggling to find a way to segment the cells in this image. The cells have been stained with anti-beta catenin antibody. My goal is to segment the cells and then analyze them for morphological data, such as shape, AR, circularity, etc.

Doing this by hand takes forever, so I wanted to find a quicker, at least semi automated way to do it. One thing I have used so far is StarDist, but I am not satisifed with the quality of the segmentation. I have also tried segmenting through CellProfiler and ilastik, but I can’t figure out how to get these software to recognize the cells, instead of just highlighting the boundaries.

Also, any advice on what filters to use would be great too. I am very new to image analysis (I sort of got thrown into it due to COVID) and so I finding the best way to do anything is very confusing and difficult for me.

Any help would be much appreciated! Thank you!

1 Like

Hi @Alefiyah_Bookbinder,

this looks like some cool data :slight_smile: A quick note on the format. You seem to have saved a single channel value as an RGB color image. Also the scale bar is included. So it looks like some sort of screenshot from somewhere. I would recommend saving the original image (maybe someone from your team can help with that).

Now for your segmentation task:

You have staining only on the boundaries, so the segmentation method should be able to deal with this. I’d recommend chaining two ilastik workflows:

1.) Pixel classification: to get a good prediction image of the boundaries only. So you’d want to suppress all other things that have high grey values in your image and highlight boundaries as good as possible. At the end you want to export a probability image.

2.) Boundary-based segmentation with multicut: Here you load your original image as raw data, plus the prediction image for the boundaries. Here is a quick demo on youtube on how to work with this workflow.

Alternatively for 2., you could also load the boundary probability image into Fiji, threshold the boundary channel, compute the distance transform and do watershed on the distance transform image. However, the multicut will most likely yield better results.

I hope this gives you a rough idea on how to solve this in ilastik. Feel free to ask for any information along the way :slight_smile:


1 Like