Preparing data in preprocessing for SC segmentation task

[uzaymacar]

This PR introduces modifications to preprocess/preprocess_data.sh to such that it prepares a cleaned data processed folder for carrying out spinal cord (SC) segmentation. More explicitly, a user input argument <TASK> is added for specifying whether the preprocessing prepares data for lesion segmentation (lesionseg) or SC segmentation (scseg). For scseg, we copy raw subject images as inputs and SC masks as target labels. More documentation on how to run can be found in the modified README.md in this PR.

It is important to note that we have subjects with and without manually-corrected SC segmentations. This means that for some subjects, the ground-truth will be taken as the output of sct_deepseg_sc. This might be reasonable because these segmentations were marked "good to go" by @jcohenadad during SCT quality-control.

[jcohenadad]

Nice-- just this to address #18 (review)

[uzaymacar]

The last five commits implement copying the relevant JSON for the SC segmentations s.t. we use the auto-generated JSON for manually corrected and create a new JSON for sct_deepseg_sc generated SC segmentations. I re-tested out training with the um/ivadomed_scseg_training branch and it works. The newly created JSONs (XXX_seg-manual.json) have the following form:

{
    "Author": "Generated with sct_deepseg_sc",
    "Date": "2021-12-17 16:40:34"
}

This form is equivalent to the one outputted by the manual correction script.

Hey @jcohenadad, is this okay from a convention perspective?

[jcohenadad]

Hey @jcohenadad, is this okay from a convention perspective?

this is great!