Clarification of manual-correction.py use cases

[valosekj]

There was some confusion about how and when to use manual_correction.py script. This issue intends to clarify possible use cases.

Use Case I - manual correction of automatically generated labels

In this use case, images and labels for each subject are located in the same directory, for example:

data_processed
└── sub-amu01
    └── anat
        ├── sub-amu01_T2w.nii.gz         # image
        ├── sub-amu01_T2w_seg.nii.gz     # label
...

data_processed is usually created by sct_run_batch.py. Labels are usually generated automatically, for example, using sct_deepseg_sc. This use case is used for the manual correction of these automatically generated labels. The manually corrected labels are saved under derivatives folder.

Use Case II - re-correction of already existing labels

In this use case, images are located under the root directory, and labels are located under derivatives, for example:

├── derivatives
│   ├── labels
│   │   ├── sub-amu01
│   │   │   ├── anat
│   │   │   │   ├── sub-amu01_T2w_seg-manual.json     # label
│   │   │   │   ├── sub-amu01_T2w_seg-manual.nii.gz
...
├── sub-amu01
│   ├── anat
│   │   ├── sub-amu01_T2w.nii.gz         # image
...

The dataset is usually already git-annexed on our data server (for details see here) and labels under derivatives have already been manually QCed and corrected. This use case is thus just about the re-correction of these already existing labels. The corrected labels overwrite the labels under derivatives folder.

Use Case III - creating labels from scratch

In this use case, only images are available. Either under data_processed or under the root directory of an already git-annexed dataset. For example:

data_processed
└── sub-amu01
    └── anat
        ├── sub-amu01_T2w.nii.gz         # image
...

├── sub-amu01
│   ├── anat
│   │   ├── sub-amu01_T2w.nii.gz         # image
...

The purpose of this use case is to manually create new labels from scratch, for example in the case of MS and SCI lesions (we do not have a robust automatic algorithm) or in the case of vertebral labeling (vertebral labeling fails for some contrast and FOV). The manually created labels are saved under under derivatives folder.

[NathanMolinier]

To avoid future confusion regarding these different cases ("which case am I currently facing ?"), I created a pull request to directly specify:

• the folder where images are stored using -path-img
• the folder where labels are stored using -path-label
• the folder where corrected labels will be stored using -path-out