New data added !! - 16/01/2025

Hello everyone,

Thank you for participating in Fuse My Cells!

We added new data in the database. Be careful about the Z degradation (ascending / descending) and some new images has no Z degradation but X(Y) degradation.

PS: Do not hesitate to ask questions, to share information and to inform us of any problem.

I might be wrong but, although you say

we have updated all the images with a few z-plane crops to reduce the empty/black parts of the images. This changes the 3D sizes, which are no longer standard (and reduces the final file size, lightening the final size of the database)

I seem to see empty/black part of images in most of the images I downloaded. For example, image_4_nucleus_angle.tif and image_4_nucleus_fused.tif are both empty on Z-slices 0-25 and 203-230.

Apologies if that's a misunderstanding from my end.

Thank you for your message,

I checked this image and I understand what's going on. When cropping the images, I set a threshold of 150 pixels (instead of 0 after checking various images). So, I only deleted shots that didn't contain pixels larger than 150 px.

As it happens, there are certain areas of the shots that have “little artifacts” of light, so these shots were not considered “ full black ”. This is what happens in image_4_nucleus_fused.tif (and maybe some few others).

I see, it makes total sense, thanks for clarifying, I was just afraid I was displaying images wrong / not understanding the data well, I'm reassured now :)

Is it normal to have almost completely dark images for slices of the ground truth in the middle? E.g. image 181 nucleus (fused, so supposedly ground truth) has clear images with relevant information for low Z and high Z but for Z = 444 for example, there is nothing.

Minimal Reproducible Example:

import matplotlib.pyplot as plt
import tifffile

file_path_fused = "data/image_181_nucleus_fused.tif"
image_3d_fused = tifffile.imread(file_path_fused)
plt.imshow(image_3d_fused[444, :, :], cmap='gray', vmin=0, vmax=image_3d_fused.max())
plt.axis('off')
plt.show()

Hello,

That's an excellent question, thank you!

I tested it on my side to see if we'd made a mistake ; and rest assured, everything's fine. The images are not “entirely black” even in your example of image 181 slice 444.

It's not an easy matter, because managing image intensity on microscopy images is quite special.

Firstly, I've checked with print(np.unique(image_3d_fused[444, :, :])) and there are 3912 different values inside.

Then, I've tried to plot only plt.imshow(image_3d_fused[444, :, :], cmap='gray'). And we can see that there is something on the image. So I looked in detail at the code you share and I think it's related to the management of intensity specific to microscopy images, in particular for 3D images.

When writing the vmax with image_3d_fused.max() you calculate the max value over the full 3D image and by printing it it's 64057.

However the max value of a slice is not always the same as the one of the full image and for example the maximum value of the slice 444 (image_3d_fused[444, :, :].max()) is 7375.

Then you may write the "slice vmax" then you can see the image:

plt.imshow(image_3d_fused[444, :, :], cmap='gray', vmin=0, vmax=image_3d_fused[444, :, :].max())

I see, thank you, although I'm not completely convinced about the quality of some fused images (again, it might come from a misunderstanding on my side, apologies if it was).

Let me take a better example:

Here (that might be very minor now that I think about it), there is a sort of a "glow effect" around the fused nucleus (you can take out vmin and vmax)

import matplotlib.pyplot as plt
import tifffile

file_path_fused = "data/image_51_nucleus_angle.tif"
image_3d_angle = tifffile.imread(file_path_fused)
plt.imshow(image_3d_angle[80, :, :], cmap='gray', vmin=0, vmax=image_3d_angle.max())
plt.axis('off')
plt.show()

file_path_fused = "data/image_51_nucleus_fused.tif"
image_3d_fused = tifffile.imread(file_path_fused)
plt.imshow(image_3d_fused[80, :, :], cmap='gray', vmin=0, vmax=image_3d_fused.max())
plt.axis('off')
plt.show()

If you change the slice to slice 145-150, both are incomplete (or maybe it's a biology thing and that's actually all there was?). That's what I was most concerned about.

Thank you so much for this post!!!

Everything is normal for some reasons: - these are not the nucleus but membranes imaged with a less powerful laser hence the 'glow effect'. - for the incomplete side of planes 145-150 it's due to the shape of the embryo and its depth in the imaging frame: hence the incomplete side (present on the 2 images so more “normal” than only one).

Thanks again for allowing us to see this error (the acquisitions should have been in channel 0 membrane and 1 nucleus, here channel 1 is a membrane with a less powerful laser).

I've updated zip 41-60 and I'll make a post.

Does this answer your question?

Thanks!

Can you explain this « present on the 2 images so more “normal” than only one »? I’m not sure I understand what it means.

Yes, I was a bit ambiguous, sorry.

I meant that the shape and the “incomplete” part are the same for the angle image as for the fusion image, i.e. “the way the biological sample is and looks”. As we've imaged biological samples, not all embryos are ball-shaped, and in this case, going deeper, we can see a "crescent moon" shape.

Is this clearer ?

Yes thanks