Splitting of signals from the same fluorescence channel

Hello image processing comunity,
we struggle a bit with the following problem:
We imaged a C. elegans embryo with 3 different fluorescent tags on a Zeiss Airyscan confocal microscope (3D image series over time, see example image). The tags are: GFP-gamma tubulin (spindle poles), mCherry-histones (DNA) and mKate2-PH (cell membrane). For technical reasons we cannot split the DNA and the membrane signal during image acquisition. So we end up with 2 distinct signals in the red channel.
Is there an image processing approach trying to separate the DNA from the membrane signals? I thought about segmenting the DNA and substrating it from the original image as the membrane staining is quite strong and has a distinct pattern compared to the DNA signal.
I would appreciate very much any advise.

1 Like

Hi @Gunar,

A trick is that your membranes are all connected. So they are a big object, so when you look for connected components, you can filter them.
There is one drawback, DNA that touch membrane can’t be separated.
Another problem is the background. Will you be able to discriminate DNA from background signal ?


1 Like

Two things to keep in mind, or that you may have already considered!

First, AiryScan processing can do… interesting things to your pixel values, as I have seen in some of the colocalization plots of pixel values! Beautiful mathematical loops that almost certainly do not exist in nature. Hopefully your comparison of Airyscan-Airyscan would be relevant, but watch out for that. Second, even with the thin optical slice from Airyscan, you are likely to get some light contamination from cytoplasmic staining above and below your nucleus in cells that size, so any measurement you take will be partially dependent on how close to the “top” or “bottom” of the cell your nucleus is (maybe not if it is dead center?).

Good luck!

Thanks for your replies.
I think I will try Nicos idea and go for the big membrane segments.
I have not experienced weired AiryScan artifacts so far, but I never looked for them. Is it something one should worry about?


Not terribly, but it makes me nervous about trusting the intensity values at small scale.

You can see some nice loops here when looking at high resolution data over only a small array of pixels. They look more like something out of one of my old toys with a pen on a pendulum than normal pixel distributions.
Lets see if this lets me put up a gif.

Neat. Just playing around with ActivePresenter. Anyway, that’s a recording of my dragging that small circle around within the nucleus of a 63xOil Airyscan image. Maybe it is just me, but those don’t look like the same pixel distributions I expect from biological data. I also have not tested this extensively, so be liberal with all of your grains of salt.

1 Like