Primary Object Detection: nucleus

Dear All,

What is the best way to identify nucleus from the image:

Thank you in advance,
Volodymyr

Hi,

It’s quite a tricky mixture, with nuclei size range from small to large, bright nucleoli, and clumped cells. Until others come to help, I suggest to smooth the image first and identify nuclei with Otsu, as shown in the example pipeline example.cpproj (637.4 KB)

Hope that gives you some start.
Good luck.

Would you be able to elaborate on how the smoothing helps for the detection please?

I’ve been using Ilastik a lot recently but this image has me wondering how I would deal with the clumping.

The Gaussian filter smoothing with a “disk size” of about 1.5x the size of the bright nucleoli is only meant to drive the objection identification toward detecting bigger objects (which are the desired nuclei) , instead of catching the small nucleoli.
Metaphorically, if I want the segmentation to look for a mountain, instead of looking for the human with bright white hat standing on the top of the mountain, I would first blur the picture a bit, now the human is “dissolved” into the terrain and not catchy anymore.

1 Like

Awesome, thanks for your reply!