Generalize manual correction scripts

[valosekj]

Description

This PR proposes an initial version of scripts for manual correction workflow. For details, see README.
The scripts are based on ukbiobank-spinalcord-csa scripts. To allow easy tracking of changes proposed in this PR, I pushed the ukbiobank-spinalcord-csa scripts to the main branch.

Compared ukbiobank-spinalcord-csa scripts, I added:

• new input flags (-path-derivatives, -path-derivatives, -suffix-files-seg, -suffix-files-gmseg, -suffix-files-label, -label-list, -viewer) in f89a9ac
• correction of gmseg in 291be5f
• selection of viewer (FSLeyes, ITK-SNAP. 3D Slicer is not implemented yet, though) in 44a54f6
• fetch_subject_and_session function (to unify get_subject and get_session under a single function) in 0e312aa and daf74dc
• fetch_yaml_config function to simplify loading of yaml config file among scripts in 272c4c9
• further improvements (such as manipulation with paths, comments to clarify code, etc.)

I would extremely appreciate feedback from SCT users, such as @sandrinebedard, @kiristern, @naga-karthik, @rohanbanerjee, and @mchen1110. You can test the PR using the instructions below or on some dataset you are currently working on.

This PR partly addresses #1. But still, a lot of features describes in spinalcordtoolbox/spinalcordtoolbox#3353 (comment) have to be implemented.

How to test this PR

1. Clone repo, create venv, and install dependencies:

git clone https://github.com/spinalcordtoolbox/manual-correction.git
cd manual-correction
git checkout jv/generalize_manual_correction_scripts
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. Download the testing dataset. The dataset contains a sample output as provided by sct_run_batch. The dataset also includes the files_to_correct.yml file with a list of files for manual correction (obtained from the SCT QC html report.

3. Run package_for_correction.py:

This simulates the real-world workflow in which a dataset is processed on a remote server, and you do not want to copy the whole dataset (GB of data). Instead, only the images and labels that need to be corrected (listed in files_to_correct.yml) are zipped by the package_for_correction.py:

python package_for_correction.py -path-in ~/manual_correction_example_dataset/data_processed -config ~/manual_correction_example_dataset/files_to_correct.yml -o ~/manual_correction_example_dataset/data_to_correct

(Now, you would have copied the zipped file from a server to your local machine. Since we are just simulating real-world workflow, the copy from a server to a local machine is skipped for now.)

4. Run manual_correction.py:

The package_for_correction.py outputs data_to_correct.zip. Unzip this archive and run manual_correction.py:

python manual_correction.py -path-in ~/manual_correction_example_dataset/data_to_correct -config ~/manual_correction_example_dataset/files_to_correct.yml -path-out ~/Desktop/manual_correction_example_dataset -viewer fsleyes

Note: you can specify a viewer (FSLeyes, ITK-SNAP) for SC and GM segmentation corrections. (3D Slicer is not implemented yet)

First, you will have to enter your name (which will be used for .json sidecars). Then, the package_for_correction.py will iterate across files listed in files_to_correct.yml. It will gradually open FSLeyes (or ITK-snap) for SC and GM segmentation correction and sct_label_utils for disc label correction. The script saves manually corrected files under the derivatives/labels folder (you can change this folder by the -path-derivatives flag). The script also outputs a QC report (--> you can check how you corrected the labels).

Once finished, you can rerun the command. In this case, the script detects manually corrected files under derivatives/labels and asks you if you want to overwrite them.

Fixes: #1

[valosekj]

@naga-karthik Thank you for the implementation of this super useful feature! 👍🏻

[valosekj]

The PR is ready for merge. If there are no additional comments or suggestions, I will do merge.