Nuclei counting: stuck on a tricky stripy noisy image


I am trying to count the nuclei (DAPI stained) in 12 images using CellProfiler 3.1.8. The images have slightly different dimensions and illumination artefacts. The images also have stripes/banding which messes with the object detection (see screenshot).

I have removed most of the illumination artefacts (see screenshot), but haven’t had much luck with the bottom stripy half of the image.

How can I clean up this image and count the nuclei? I’ve experimented with EnhanceOrSuppressFeatures and TopHatTransform (which seems to find the stripes), but haven’t had any luck so far and am mostly stumbling around in the dark. Any ideas or suggestions?

I’ve attached a cropped region of the image (since the original tif is ~50 MB)

dapi_cropped.tif (8.1 MB)

It seems the proper approach would be to fix the problems with the microscope. Looks like a problem with the confocal scanner or the light source (flickering).
If you really can’t redo the experiments:
One method to remove such periodic structures is in the frequency domain. Unfortuately Cellprofiler does not come with tools for this. A quick demonstration in Fiji.

  • crop such that you only deal with the affected region
  • FFT
  • Blot out the bright parts of the vertical line
  • Inverse FFT

If you have lots of images this is probably too tedious. The best approach is to go back to the microscope and capture these images properly.