Allen Cell Discussion Forum

Filling empty pixels in 3D segmentation

Hi,
I am currently trying perform 3D segmentation on cells using beta actin channel. One issue I am facing is that the walls appear to be ‘porous’ - i.e. there are holes in the wall that connect the cells. This makes it difficult to segment the cells properly since most of the cells appear to be connected to each other. Is there a good way to overcome this problem? I would appreciate any suggestions that can help remove the gaps.

Thanks!
Pushkar

Hi @pushkar, thanks for your interest in using Segmenter. I totally understand your challenge. There are a few things you could try.

Option 1: If your cells are relative distant from each other, you may simply use a morphologica closing (with proper structure element) to close the holes.

https://scikit-image.org/docs/dev/api/skimage.morphology.html#skimage.morphology.binary_closing

Option 2: If your cells are very close to each other, you may still use a morphological closig. But, this will not help you get indiviual cells which are close to each other. For this purpose, you may try distance transfer + watershed to seperate them.

Option 3 and more: If neithor of above two works, there are more to explore, but may require longer development time.

Thanks,
Jianxu