Create converter for btrack files

[K-Meech]

Create a converter for btrack output files to the cell tracking challenge format.

Output files

Btrack outputs to an HDF5 file containing various outputs:

• summary of HDF5 file structure - scroll down to 'Notes' section
• It seems to already contain a text file in the correct format under tracks/obj_type_1/LBEPR. It has two additional un-needed columns: R for the label of the root track and G for the generational depth. These columns could be removed or ignored.

Existing converters

There are no existing CTC converters, but btrack already seems to be using pretty much this format. We can extract the text file with python:

import h5py
with h5py.File("tracks.h5", "r") as f:
    dset = f['tracks']['obj_type_1']["LBEPR"]
    arr = dset[:]

or directly use their export_LBEP function We may want to consider adding options to the btrack napari plugin, for those who don't use any python code.

[K-Meech]

Extracting other features (not directly in CTC file)

Inter-frame time (i.e. time in seconds that each frame represents)

• On a first look, btrack seems to only use frames for its time features.
• I've asked in the discussion on their repo if the inter-frame time is stored anywhere in the .h5 file

Update: one of the main maintainers confirmed in their repo's discussions that: 'Unless the timestamps were explicitly added as metadata, then there's no guarantee that they will be stored. The file format uses an integer index by default for the frames.'

Right-censoring

Btrack assigns each track a fate which could be useful here. The different fates are available on their repo, but don't have much documentation. In the btrack paper under 2.3.4 Global Track Optimization they are listed as:

• false positive track,
• initializing at the beginning of the movie or near the edge of the FoV,
• termination at the end of the movie or near the edge of the FoV,
• a merge between two tracklets,
• a division event
• an apoptotic event.
• In addition, we added “lazy” initialization and termination hypotheses, allowing cells to initialize or terminate anywhere in the movie.

Browsing around in their example jupyter notebook, as well as visualising in napari would mark the following as relevant:

• TERMINATE_BACK = reaches end of the movie
• TERMINATE_BORDER = goes out of frame left, right, up or down
• TERMINATE_LAZY = track terminates at any point in the movie. From browsing through the examples, these are normally tracking mistakes where it fails to link two tracks together, or it fails to recognise a division event and so ends unexpectedly.
• FALSE_POSITIVE = these are normally very short tracks (may only last a frame or two) that tend to have no children
• APOTOSIS / DEAD = mark dead cells

These are assigned with a certain likelihood based on their hypothesis model, so I think using them to create the right-censoring column should be optional. We could have two flags:

• Set all tracks with TERMINATE_* fates as right-censored
• Remove tracks marked as FALSE_POSITIVE and their descendants (if any)

Missing frames

I couldn't see anything directly related to missing frames.