Using membrane channel for cell boundary segmentation

[mdayao]

Hi @eric-czech,

I am looking to use a membrane stain channel for segmentation the cell boundaries in CODEX data. I see that there is a "processor: cytometry: membrane_channel_name" in the config file for me to specify the membrane channel to use; however, it is unclear to me what the different parameters under "processor: cytometry: segmentation_params" should be set to in this case.

Additionally, I know that Cytokit uses CellProfiler's U-Net for nuclei segmentation, but what is the method of segmentation for the cell membranes?

Thank you, Monica

[eric-czech]

Hi @mdayao, the membrane segmentation is pretty straightforward and is based on CellProfiler methods that made sense in the context of tissue images. It works like this:

• Apply some gaussian blur; controlled by memb_sigma
• Calculate a threshold; default method is li same as CP but can be changed with memb_tresh_method
• Fill holes in the binary mask from the threshold; controlled by memb_hole_size
• Optionally, clip all membrane pixels to have some minimum or maximum distance from the nearest nucleus pixel; controlled by memb_min_dist and memb_max_dist
• Segment the cell boundaries using Voronoi-Based Segmentation of Cells on Image Manifolds; controlled by memb_propagation_regularization where lower values mean more weight is given to the intensity of the original membrane image and high values mean more emphasis is put on splitting cells at something closer to equidistant points (see here for more details)

I never documented that anywhere but the code for it is here. This config also shows some settings I used for membrane segmentation in the publication.

[mdayao]

Great, thank you! This is very helpful.