remove small lables

[JustineJiao]

Hi there, It's me again^_^

I tried to precess labeled images generated by Cellpose with LabelsToROIs, but I constantly found some small labels/particles that equels to 1 pixel, which affect cell size calculation.

When I "Analyze Regions" in MorpholibJ, there is no such small labels shown, so I suspect that this is caused by LabelsToROIs.

Do you have any idea to remove these small lables?

Thanks a lot!

Justine

[ariel-waisman]

Hi Justine, Could you send the original label image for this example? By hovering with the mouse on that pixel you can check the label value (or Id), which appears at the bottom left of imageJ main window. This will tell if the problem is with CellPose segmentation or further down the line. Cheers, Ariel El jue., 19 may. 2022 01:08, JustineJiao ***@***.***> escribió:

…

Hi there, It's me again^_^ I tried to precess labeled images generated by Cellpose with LabelsToROIs, but I constantly found some small labels/particles that equels to 1 pixel, which affect cell size calculation. When I "Analyze Regions" in MorpholibJ, there is no such small labels shown, so I suspect that this is caused by LabelsToROIs. Do you have any idea to remove these small lables? Thanks a lot! Justine [image: CT4-1-1 1_cp_masks_Erosion_0px_Outlines] <https://user-images.githubusercontent.com/58281975/169202279-5617deb3-188e-45f7-8961-36eb9bf5c4c7.png> — Reply to this email directly, view it on GitHub <#9>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFAA54CKNPJOGCC2R6ZQ6ADVKW5DZANCNFSM5WK2YYFA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[ariel-waisman]

Hi Justine,

Just saw that you had already sent the original image :D
I just saw that the problem lies in the original label image. That pixel
has the same label ID (8) that the brown big cell to the right

Is this the result of Cellpose's segmentation just as it comes from the
software? If this s the case, it must be a glitch of Cellpose, maybe you
can generate an issue showing this in Cellpose's github page.

For some "cells" as the one shown, the label also includes a disconnected
pixel. Since the overall label (extra pixel + cell) has a large area, you
cannot size filter it using MorpholibJ/Label Images/Label Size Filtering.

But there is a small trick that you could use to circumvent this:

1. Based on Cellpose's label image, you can generate a "re-mapped" one with
   new labels using the "Connected components labeling" plugin from Fiji, with
   options 16 bit and connectivity of 4. (bare in mind, there are two similar
   connected components labeling plugins, use the one that says componentS in
   plural.
   This will generate a similar label image, but with a unique label ID for
   all the "connected" label pixels. This means that the "extra pixel" will
   now have a different ID than the remaining on the cell. Now, you can size
   filter it to exclude small labels or single pixels using MorpholibJ/Label
   Images/Label Size Filtering

See how now that pixel is not included in the final label. Thus,
LabelsToROIs will not identify it as a ROi.
I also show you how you can change the color of the gray scale
labels, which are not that confortable to see.

Hope this helps!
Cheers,
Ariel

[JustineJiao]

Great thanks! It's very helpful. You are right, this maybe a glitch of Cellpose.
Again, thank you so much!
Justine Jiao