embl-cba / spindle3d

0 stars 0 forks source link

Background #22

Open TobiasKletter opened 3 years ago

TobiasKletter commented 3 years ago

Some montages of the 6 top and the 6 bottom slices: https://www.dropbox.com/s/z3akt7ug29d8spa/Background_Montages.zip?dl=0

Looking at these, I could imagine it would be smart to sample pixels in these corners of the top slice, because they seem to be consistently the darkest spots and we it would still consider whatever contribution the imaging medium might have... Screenshot 2021-03-16 at 12 51 32

However, this approach only works when cells grow in a flat environment and not on top of each other...

tischi commented 3 years ago

I am surprised by the relatively strong intensity drop off at the edges. ping @ssgpers

  1. Which microscope was this?
  2. How bright are the cells in comparison?
  3. How do the bottom slices look like (ideally with same B&C settings as the top ones)?
TobiasKletter commented 3 years ago

Which microscope was this?

This is from a Nikon spinning disk (The dataset that we acquired at ALMF was done on a Zeiss LSM and basically has zero offset)

How bright are the cells in comparison?

image

How do the bottom slices look like (ideally with same B&C settings as the top ones)?

I packed some of these in the .zip at the very top of the thread.

tischi commented 3 years ago

Ok, looking at the image an offset of around 600 would be appropriate, which is close to the value in the corners. The intensity pattern on the top may be due to out-of-focus light from the cell (spinning disk is not so good in optical sectioning as confocal)...

I packed some of these (bottom slices) in the .zip at the very top of the thread.

Could you just paste a montage here? 😄

TobiasKletter commented 3 years ago

Might also work:

image image

Actually, looking at the orthogonal views, I prefer the bottom slices: image

tischi commented 3 years ago

Tricky.... as always.

This might be something very special about your spinning disk, that the bottom looks better than the top. So I don't think we can conclude based on this.

@ssgpers What would you do in such data to estimate the fluorescence background outside of cells?

ssgpers commented 3 years ago

Hi guys,

As I was not part of part of the discussion from the beginning, can you please clarify for me whether 25x25 um is entire image or crop from the bigger image.

In the second case original image maybe can be used to estimate the background using tricks like filter with very big radius.

If we need to work with a crop image I can think of the following strategy:

TobiasKletter commented 3 years ago

@ssgpers Hi Alex! In this particular case I would also have the bigger image, but the idea is to have a strategy that works on the already cropped input images for Tischi's morphometry analysis. We constructed the workflow such that it expects cropped cells...

If you want to have look, here would be an example: https://www.dropbox.com/s/rts95zg6je4dauw/Raw_image.tif?dl=0

and the typical cropped cell ready for analysis: https://www.dropbox.com/s/0r40re908dyrbtv/Cropped_mitotic.tif?dl=0

edit: here is the same cell with a resampled isometric voxel size of 0.25 µm and the binary cell mask (channel 2) https://www.dropbox.com/s/q21mqqy70r53ho2/Cropped_mitotic_isometric_CellMask.tif?dl=0

tischi commented 3 years ago

@TobiasKletter

If we limit ourselves to tissue culture I think using the median intensity in the lowest plane may be the best option for now (in a sense the only simple option).

We could add an option for the user to manually enter a background intensity, should we rather do this? This would be the safest. And I think it would make sense because often, for a whole data set one can (and should) use the same background.

tischi commented 3 years ago

Does the ground truth set cover different cell types and microscopes?

TobiasKletter commented 3 years ago

@tischi
Only saw this now, sorry!

If we limit ourselves to tissue culture I think using the median intensity in the lowest plane may be the best option for now (in a sense the only simple option).

We could add an option for the user to manually enter a background intensity, should we rather do this? This would be the safest. And I think it would make sense because often, for a whole data set one can (and should) use the same background.

Yes I'd be happy with both options! Maybe the default would be to take the median intensity in the lowest plane, if there is no user-specified number that was entered?

Does the ground truth set cover different cell types and microscopes?

This belongs to the other issue, right?

TobiasKletter commented 3 years ago

Memo: Let's wait for feedback whether or not this needs to be implemented