-
-
Notifications
You must be signed in to change notification settings - Fork 149
repositories
Jan Valosek edited this page Jan 31, 2023
·
14 revisions
Each model is saved in its own repository under the ivadomed organization. The convention for naming repositories is the following:
model_task_animal_pathology_region_contrast_architecture
Should be small letters only.
Fields:
- task = {seg, label, find}, default=seg
- animal = {human, dog, cat, rat, mouse, ...}, default=human
- pathology = {ms, sci}
- region = {sc, gm, csf, brainstem, axon, myelin, ...}, default=sc
- contrast = {t1, t2, t2star, dwi, sem, tem, oi, ...}, default=None
- architecture = {unet2d, unet3d, filmCharley, hemisAndreanne}, default=unet2d
Examples:
model_seg_monkey_sc_t1_unet3d
# multi-channel, multi-class
model_seg_sc-gm_t1-t2_unet3d
Models to be used by 3rd party software (e.g. SCT) should be uploaded as 'assets' to a release of the repository. The steps are:
- Create a release of the repository. The tag and title of the release should be
rYYYYMMDD
, example:r20211223
. - Put the model and JSON file inside a folder that has the name of the model.
- Zip the folder and upload it as an asset in the release
- Publish the release.
Convention:
data_purpose_study_animal_anatomy_contrast
Fields:
- purpose = {testing, example, tumor, ms, ...}
- study = {} Study associated with the dataset
- animal = {human, dog, cat, rat, mouse, ...}, default=human
- anatomy = {sc, brainstem, brain, foot, ...}, default=sc
- contrast = {t1, t2, t2star, dwi, sem, tem, oi, ...}, default=None
Examples:
- data_testing
- data_example_spinegeneric
- data_axondeepseg_sem
This is a folder at the root of the dataset, which includes derivatives from the data, including manual segmentations/labeling to use as ground truth for training. According to BIDS, these data should go under derivatives/
folder, and follow the same folder logic as the sub-* data. Example:
data-multi-subject
│
├── dataset_description.json
├── participants.json
├── participants.tsv
├── sub-ubc01
├── sub-ubc02
├── sub-ubc03
├── sub-ubc04
├── sub-ubc05
├── sub-ubc06
│ │
│ ├── anat
│ │ ├── sub-ubc06_T1w.json
│ │ ├── sub-ubc06_T1w.nii.gz
│ │ ├── sub-ubc06_T2star.json
│ │ ├── sub-ubc06_T2star.nii.gz
│ │ ├── sub-ubc06_T2w.json
│ │ ├── sub-ubc06_T2w.nii.gz
│ │ ├── sub-ubc06_acq-MToff_MTS.json
│ │ ├── sub-ubc06_acq-MToff_MTS.nii.gz
│ │ ├── sub-ubc06_acq-MTon_MTS.json
│ │ ├── sub-ubc06_acq-MTon_MTS.nii.gz
│ │ ├── sub-ubc06_acq-T1w_MTS.json
│ │ └── sub-ubc06_acq-T1w_MTS.nii.gz
│ │
│ └── dwi
│ ├── sub-ubc06_dwi.bval
│ ├── sub-ubc06_dwi.bvec
│ ├── sub-ubc06_dwi.json
│ ├── sub-ubc06_dwi.nii.gz
│ ├── (sub-ubc06_acq-b0_dwi.json)
│ └── (sub-ubc06_acq-b0_dwi.nii.gz)
│
└── derivatives
│
└── labels
└── sub-ubc06
│
├── anat
│ ├── sub-ubc06_T1w_seg-manual.nii.gz <---------- manually-corrected spinal cord segmentation.
│ ├── sub-ubc06_T1w_seg-manual.json <------------ information about origin of segmentation (see below). Note: this file is optional for ivadomed's loader.
│ ├── sub-ubc06_T1w_labels-disc-manual.nii.gz <------- intervertebral disc labels. Voxel located at the posterior tip of each disc. The value corresponds to the disc, e.g.: C2/C3: value=3, C3/C4: value=4, etc.
│ ├── sub-ubc06_T1w_labels-disc-manual.json
│ ├── sub-ubc06_T2star_gmseg-manual.nii.gz <------- manually-corrected gray matter segmentation
│ ├── sub-ubc06_T2star_gmseg-manual.json
│ ├── sub-ubc06_T2star_lesion-manual.nii.gz <------- lesion segmentation
│ └── sub-ubc06_T2star_lesion-manual.json
│
└── dwi
├── sub-ubc06_dwi_moco_dwi_mean_seg-manual.nii.gz <-- manually-corrected spinal cord segmentation
└── sub-ubc06_dwi_moco_dwi_mean_seg-manual.json
The convention for suffix is the following:
_task-region
Fields:
- task = {seg, label}, default=seg
- region = {sc, gm, csf, brainstem, tumor, edema, cavity, lesion, disc, axon, myelin}, default=sc
The .json file includes the following information:
{
"Author": "Bob Dylan",
"Date": "2020-07-29 21:25:37"
}