search

vanvalenlab / deepcell-applications

Run DeepCell Applications
Other
8 stars 6 forks source link

Only nuclei mask for ome.tiff files/non-RGB tif files #44

Open KOBE24DUNK opened 3 months ago

KOBE24DUNK commented 3 months ago

Hi,

Thanks for the super impressive tool which helps me a lot.

But I found some issues when running deepcell mesmer: only nuclei mask.tif was obtained for ome.tiff file/non-RGB tif files. My ome.tiff file contains nine channels, the first of which is the DAPI channel and the rest are cytoplasm channels. Ideally, I would like the mask to contain the nucleus as well as the different cytoplasms.

docker run -it --gpus all -v C:\Data\deepcell:/data vanvalenlab/deepcell-applications:latest-gpu mesmer --nuclear-image /data/ROI_images/ROI.ome.tiff --membrane-image /data/ROI_images/ROI.ome.tiff --nuclear-channel 0 --membrane-channel 1 2 3 4 5 6 7 8 --output-directory /data/masks --compartment "whole-cell"

Maybe I have a misunderstanding of the mechanisms. Please forgive my stupidity in this regard!

Thanks a lot for any suggestion!

rossbar commented 3 months ago

The internal processing for combining channels is very simple - it simply adds the images together along the inferred channel axis in the image. If the channels are not properly normalized/exposure corrected to the same scale, this likely won't work very well - how this is best done is really dependent on the data, so I'd recommend experimenting/visualizing the results from different techniques of combining your non-nuclear channels.

KOBE24DUNK commented 3 months ago

The internal processing for combining channels is very simple - it simply adds the images together along the inferred channel axis in the image. If the channels are not properly normalized/exposure corrected to the same scale, this likely won't work very well - how this is best done is really dependent on the data, so I'd recommend experimenting/visualizing the results from different techniques of combining your non-nuclear channels.

Thank you for your so quick response. :) I'm still trying it and test the results under different combinations of techniques. I'll get back here if I met further issues.