search

tischi / fiji-plugin-morphometry

0 stars 1 forks source link

Spindle: Get DNA threshold in same way as tubulin threshold (?) #84

Open TobiasKletter opened 5 years ago

TobiasKletter commented 4 years ago

Some testing data with different degrees of signal-to-noise in the DNA channel:

https://www.dropbox.com/s/03plpdjott78voy/Dim_DNA_signal.tif?dl=0 https://www.dropbox.com/s/133xnp5ses6ugpz/Mid_DNA_signal.tif?dl=0 https://www.dropbox.com/s/tr3oabu8jm2mv6h/Bright_DNA_signal.tif?dl=0

tischi commented 4 years ago

@TobiasKletter Could you play a bit with the different auto-thresholding methods in ImageJ?

  1. [ Image > Adjust > Auto Threshold ]
  2. [ Image > Adjust > Auto Local Threshold ] (unfortunately I don't think they are really implemented in 3D though

Is there one that works well for all examples?

I could imagine as an algorithm:

  1. roughly segment the DNA using our current method
  2. from the rough segmentation define a region that includes the DNA and also pixels around it
  3. in that region run a histogram based automated thresholding algorithm (like the ones in ImageJ).

Step (2) is important because most algorithms change their output depending on the fraction of pixels that should be above threshold. So, if we can keep this kind of constant it would be a good thing.

TobiasKletter commented 4 years ago

Representative central slices:

Dim_DNA_signal_centralplane Mid_DNA_signal_centralplane Bright_DNA_signal_centralplane

Judging by eye, I think Otsu does a good job on all of them. These are montages of all slices in the stack:

Autothreshold_Stack_DimDNA-1 Autothreshold_Stack_MidDNA-1 Autothreshold_Stack_BrightDNA-1

The "Default" also gives a nice output, might actually be even identical to Otsu.

tischi commented 4 years ago

In that case maybe we can go for the Otsu and then we can cite Julius' paper (see discussion on the forum).

TobiasKletter commented 4 years ago

Agreed, we should give it a shot like this.