Looking to maximize dynamic range of calculated intensities

Hi there!

I’ve been working with .nd2 formatted 4-channel immunofluorescent images to quantify and compare expression of FITC and TRITC-labelled proteins across various cell lines and tumour models from cell microarrays and tissue sections, respectively (images linked). In brief, the pipeline I’ve been using looks something like the following and are linked below:

Cell microarray pipeline

  1. Identify primary objects (i.e. nuclei)
  2. Identify secondary objects (i.e. FITC and TRITC-labelled cells)
  3. Identify tertiary objects to exclude nuclei objects from cell objects
  4. Measure object intensity (i.e. tertiary objects)

Tissue pipeline

  1. Image correction across all 4 channels (e.g. illumination correction, align)
  2. Threshold by epithelial marker to identify ROI
  3. Mask FITC and TRITC-labelled areas to isolate ROI
  4. Threshold by intensity to identify auto-fluorescence
  5. Mask image from Step 3 by threshold from Step 4 to eliminate auto-fluorescence
  6. Measure image intensity

It seems like a simple enough analysis, but we just aren’t seeing the dynamic range we would typically expect for these samples (i.e. 0.5 to 8-fold vs. 100 to 1000-fold differences), particularly with IF. After digging through the forums, I understand that intensities of all images are automatically scaled from 0 to 1, in my case, based upon the camera’s maximum intensity value specified in the metadata of my nd2 files. But to the extent of my knowledge, this shouldn’t impact dynamic range regardless of which intensity metric I use. Previous suggestions have been to tinker with the camera settings to ensure that it’s set correctly but I was wondering if you had any other ideas or leads we could pursue in addition to that.


Thanks, that’s a super helpful amount of detail! Your pipelines off hand sound reasonable, though I haven’t personally checked them to make sure.

A few random questions, in no particular order

  1. What sort of illumination correction and alignment are you doing? Have you spot-verified that the results of both look reasonable?
  2. Have you spot-verified your segmentation of all kinds of objects? This is particularly important because you’re graphing total intensity, rather than mean or median, so a large dim thing and a small bright thing will look roughly the same by integrated intensity but quite different by mean.
  3. If you hover over a couple raw images in FIJI and/or CellProfiler to look at the raw pixel values, do you in fact see something closer to 3-fold or 1000 fold signal to noise?
  4. Same question, but for your illumination corrected images in CellProfiler?

You’re right that rescaling to 0-1 shouldn’t affect signal to noise, as long as you’re not rescaling elsewhere min-max. A number of scope/camera settings might, such as gain/gamma, or if significant areas of the image are saturated, or if your microscope is automatically set to auto-rescale all images min-max to make them “pretty” (I’ve actually encountered that last one in the wild before, after days of painful troubleshooting), but you’ll have a better sense on if that’s needed based on the answers to the above.