search

vanvalenlab / deepcell-tf

Deep Learning Library for Single Cell Analysis
https://deepcell.readthedocs.io
Other
425 stars 99 forks source link

Question regarding pixel expansion #700

Open tbudiman7 opened 1 year ago

tbudiman7 commented 1 year ago

Hi VanValen Lab,

I conducted a pixel expansion after segmentation on my image and noticed that there are small additional cells present in the pixel expansion generated outline that did not exist in the nuclear segmentation. After going through your notebook, I noticed that this happened in your example images as well - see screenshot attached below. I was wondering why it does this as my understanding is that the pixel expansion argument adds a pixel outline on top of the nuclear segmentation outlines.

Screenshot 2023-11-21 at 12 32 28 PM
ngreenwald commented 1 year ago

Hi @tbudiman7, good question. It is indeed using the same seeds for the expansion. It's possible that there are very small objects which do not meet the minimum size threshold using the raw predictions, but after expansion they are over the size threshold? You could take a look at the post-processing logic and see when the expansion is applied vs when the cutoff is applied.