Track cells with predefined segmentation masks

[tuanpham96]

I have data that have already been segmented for each day but not aligned across days for either mask ids nor field of views. I was wondering whether/how I can reshape my data somehow to load into Cell_ACDC for aligning the masks.

Plus, if the tracking is automated, is it possible to have manual correction afterwards?

Thanks!

[ElpadoCan]

Hi @tuanpham96, you can absolutely do the tracking on previously segmented data. And you can choose to do it either automatically or manually, or first automatically and then manually correct.

You are going to need to structure your data in a specific way. If you have the microscopy files I recommend using the module
0. Create data structure from microscopy file(s)... of Cell-ACDC (first button on the little launcher that opens when you launch Cell-ACDC). See the section called Create required data structure from microscopy file(s) of the user manual.

Alternatively, you can create the data structure manually. Have a look at the section called Manually create data structure from microscopy files(s) in our user manual here.

Briefly, you need one .tif file per position per channel (if you have multiple fluorescent channels for example). Then you put the .tif file(s) into a subfolder with relative path ...\Position_1\Images.

Once you have the structure, inside the Images folder, you also put the segmentation file that needs to be a NumPy array in .npz format. If you have other formats we have a utility for converting it to .npz. You can find conversion utilities from the menu on the top menu bar of the Cell-ACDC launcher called Utilities --> Convert ....

The segmentation file needs to have the same start name (which I call the basename) of the .tif file and it needs to include the word segm after the common basename. So for example let's suppose you have a .tif file called my_experiment_phase_contrast.tif, then valid segmentation file names are for example my_experiment_segm.npz, or my_experiment_segm_phase_contrast.npz, or whatever text you want to have after segm_.

Finally, if you want to work with just the segmentation file (without loading any signal like phase contrast, bright field, or fluorescent), I just thought now that it could be a cool feature to add. Essentially, I could add a feature that allows you to simply load a segmentation file from any folder you like, with any format you like. Cell-ACDC would then internally convert it to npz and it would work on that file.

I hope it's clear enough and what you think about loading only the segmentation file.

Cheers,
Francesco

[ElpadoCan]

Hi @tuanpham96, I got your data and I will do some tests in the next few days. I'm at a conference for two days so I will not have much time, but when I'm back I will definitely check it out.

I'm very excited to expand Cell-ACDC scope to calcium imaging so rest assured that we will make it work and then we see if it's useful for you. Here are my answers:

Q1: By the way, does it matter whether the starting index of Position_X is 1 or 0?

It doesn't matter, but I always started at 1.

Q2: Do you recommend phase-corrected before inputs into acdc?

I would definitely try it. If it doesn't make things worse, it will probably be beneficial.

Q3: Just out of curiosity, do the data that acdc is originally meant for usually have z-planes of the same object?

Cell-ACDC can work with 2D, 3D (either single z-stack snapshot one per position, or 2D data over time) or 4D data (z-stack over time). A z-stack is a sectioning of the same sample that usually takes a few minutes to be taken with a microscope.

I hope I answered everything and I will keep you updated with the tests I will do with your data. Have a good week!

Cheers,
Francesco

[ElpadoCan]

Hi @tuanpham96,

I had a look at your data and I didn't know you already tried to get the right structure for Cell-ACDC, nice job!

So the plane_inside data can be loaded without problem. The plane_as_position instead has the wrong shape. The shape of the .tif file that you created is (5,1,512,512), while the correct shape is (5,512,512) without the 1 for the plane. This can easily be achieved by calling the NumPy function numpy.squeeze which automatically removes axes with dimension 1.

Basically, you can add the squeeze function at line 151 of your utils.py file as follows:

imwrite(file_name, np.squeeze(mat))

The _segm.npz file instead is correct in both cases. Just to be sure I would also add the squeeze function at line 144 of your utils.py file as follow:

np.savez(file_name, np.squeeze(mat))

I hope it's clear and let me know if it works!

By the way, since I am guessing you probably are going to need some new features, I think the easiest would be for you to install from source instead of from PyPi. This way you can get new features before they are released and we can release them once you tested them.

To install from source you need to clone the repo with git and then install the locally cloned repo. You could do this in a separate environment (that you can call for example acdc-source or whatever you like). This way you have both the release version and the developer version.

You can find instructions on how to install from source here. Let me know if you need help with that.

Cheers,
Francesco

[ElpadoCan]

I forgot one thing. At the moment you are working with only one .tif file (one channel) called for example ME04_plane0.tif where ME04_plane0 is the basename and there is no channel name. This is of course fine for one channel, but remember that you can add a channel name after the basename, for example, ME04_plane0_GFP.tif. This is handy and required with two or more channels where you could have another .tif file called ME04_plane0_mNeon.tif or whatever the channel name is. The channel name can be any text you want, I usually use the fluorophore name, but you could for example call it nuclei.

For the segmentation file, it is the same idea. You can have ME04_plane0_segm.npz or you could also have ME04_plane0_segm_GFP.npz if you plan to have more than one segmentation file (for example a segmentation mask of another channel, or one 2D segmentation file and one 3D).

I hope it's clear enough and let me know if you have questions.

Cheers,
Francesco

[tuanpham96]

The shape of the .tif file that you created is (5,1,512,512), while the correct shape is (5,512,512) without the 1 for the plane.

I don't remember why, but I think it didn't pick up that there's 1 plane and 5 frames when I was doing (5,512,512) in the first place, hence I didn't squeeze it. But it was probably when I was tinkering around. Anyway, I can add squeeze back.

I think the easiest would be for you to install from source instead of from PyPi.

Good point. I'm currently using it from venv anyway so it's easy enough to switch.

This is handy and required with two or more channels

Thanks for the tip! This would be very helpful as there are others in the lab doing 2 channels.

if you plan to have more than one segmentation file (for example a segmentation mask of another channel, or one 2D segmentation file and one 3D).

So generally we'd have 1 segmentation file in cases of 2 channels. How does the program handle 2 tifs and 1 segmentation file? Does it load everything in to track?

[ElpadoCan]

Hi @tuanpham96,

The squeeze should work because to load your data I also added squeeze to Cell-ACDC and I was able to load it. Note that even if the shape is (5,512,512) you are still required to provide the number of z-slices, and by inserting 1 z-slice Cell-ACDC will ignore that and expects data with shape (T, Y, X).

So generally we'd have 1 segmentation file in cases of 2 channels. How does the program handle 2 tifs and 1 segmentation file? Does it load everything in to track?

You can choose how many channels to load for the computation of numerical features based on the loaded segmentation file. If for example you have two channels GFP and mNeon and one segmentation file ending with segm_cells.npz you can load this segmentation file, then you can load the two channels and compute the metrics. Cell-ACDC will save a file ending with acdc_output_cells.csv containing features of both channels (for example GFP_mean and mNeon_mean) based on the segm_cells.npz masks. All of this can be done from the menu on the top menubar of the main GUI (module 3. from the launcher) called Measurements --> Set measurements.... This will open a dialogue where you can choose to load channels and choose which features to save. Each feature has an info button with an explanation about that feature.

Let me know how it goes, thanks!

Cheers,
Francesco