Morph:Close operation and eliminating dumbbells

I have two questions:

  1. I’m doing a co-localization analysis for two proteins that show up as tiny granules in a fly neuron’s axons (and in the cell body but we’re ignoring that for the moment) and I’m having trouble identifying the objects. However, when I use the morph:close operation it greatly enhances object recognition. Now, since the operation can inflate the actual size of the object I’m wondering if I shouldn’t use it. Then again, since it’s co-localization and I don’t care about the size or shape of the objects does it really matter since its a proportion of object x that is overlapping with object y? In other words, the expansion in pixels occurs in both sets of neurons so it shouldn’t affect the relative comparison. Is my reasoning correct?

  2. When analyzing nuclei, I’m still getting these dumbbell shapes consisting of two nuclei touching one another. The de-clumping options (de-clumping by intensity, shape or lapcalian of gaussian) seem to work rarely. If I increase the threshold to a point where the objects do de-clump, I’ve usually lost identification of the majority of the other nuclei. Is there a better way to de-clump? Or a way to filter these dumbbell shapes out?


Morphological closing tends to close up holes and fill in gaps. If this accurately reflects the physical nature of the objects (more or less), then it’s definitely fine. Because you’re dealing with co-localization then yes, because both objects are being closed and then checked for overlap, the proportion ought be similar. Once concern is that if objects expand, then objects might be detected as co-localized when they are farther apart then they actually are. Do you think this is a problem in your case.

I could take a look at an image and pipeline to see if I could improve it, but it sounds like you’ve optimized as much as you can. You could measure the object morphology and then use the FilterObjects module, filtering on the basis of a measurement, perhaps FormFactor or Solidity. I often use DisplayDataOnImage to preview the measure to confirm that it’s usable for filtering purposes.


  1. I guess the best way to test if co-localization is unaffected is just to see if the ‘close’ operation changes the ratio of co-localization.

  2. I’ll try that with solidity. Do you ever just manually eliminate bogus objects after exporting to a spreadsheet?

Thanks again.

No (for high-throughput, it’s not practical), but you can do that manually before exporting by using EditObjectsManually.