Trouble with getting segmentation with epithelial cells

This is the pipeline and the two images I am using. One image is mcherry tagged for nucleic protein, and the other is YFP tagged for cytosolic protein. I am attempting to segment, and the pipeline is not sensitive enough to detect differences when the cells are close together. Additionally, no background was taken with just media. How do I go about getting more accurate segmentation?




subtract1_skipsaves.cp (8.29 KB)

Hi,

My suggestions/thoughts are the following:

  • In IdentifyPrimaryObjects, your nuclei segmentation seems pretty good as-is.I’d probably change the thresholding method to Otsu Global, 2-class thresholding, and the thresholding correction set to a bit less than 1 (maybe 0.8?).
  • . Increase the upper diameter limit a bit, perhaps 20, and decrease the lower diameter limit to perhaps 5. Of course, this depends if you know for a fact that you don’t want want nuclei above 15 pixels or below 8 pixels, as you have it set now.
  • Uncheck the “Automatically calculate size of smoothing filter” and manually set it to 5.
  • In IdentifySecondaryObjects, I would change the input image to the original green image from LoadImages. Set the thresholding method to Otus Global with 3-class thresholding and the middle class set to the foreground. Decrease the threshold correction factor until you capture as much of the foreground area as you can. However, the cell boundaries start looking more angular, and don’t reflect reality quite as well.
  • Remove the ImageMath module since it doesn’t add value to the pipeline.

Lastly, is the YFP image the one that you are using to quantify things, and if so, do you expect the staining to change according to experimental treatment? The reason I ask is that it is typically a good idea to have a separate stain for determining the cell boundary (one that doesn’t change with experimental perturbations) versus the stain that you are quantifying (which is expected to change). In that way, the segmentation remains consistent from image to image, across experiments.

Regards,
-Mark