Datasets to add for the next version of the model

[naga-karthik]

Datasets included in v2.4 model:

• "basel-mp2rage"
• "canproco"
• "data-multi-subject"
• "dcm-zurich"
• "lumbar-epfl"
• "lumbar-vanderbilt"
• "sct-testing-large"

Datasets to be included next:

• Beijing MS (has several different contrasts)
• Karolinska (mainly T1w, T2w, T2star)
• lumbar-marseille
• dcm-brno
• philadelphia-paediatric (at the moment, T1w and T2w)
• sci-colorado
• sci-paris
• sci-zurich
• nih-ms-mp2rage

Optional (NOTE - this could destroy the model):

• stanford-epi (open-neuro epi spinalcord; good to use as testing data)

EDIT: updated with correct list of datasets in v2.4 (cross-ref'd from datasplits.md in the v2.4 release)

[naga-karthik]

Update: Here is the current list of datasets used for training the model (potentially v2.5). This function outputs the following info in a txt file gathering all the datasets statistics.

datasets stats

DATASETS USED FOR MODEL TRAINING (n=14):

	- datasplit_dcm-zurich-lesions-20231115_seed50.json
	- datasplit_sci-paris_seed50.json
	- datasplit_lumbar-epfl_seed50.json
	- datasplit_data-multi-subject_seed50.json
	- datasplit_dcm-brno_seed50.json
	- datasplit_basel-mp2rage_seed50.json
	- datasplit_sct-testing-large_seed50.json
	- datasplit_canproco_seed50.json
	- datasplit_sci-colorado_seed50.json
	- datasplit_dcm-zurich_seed50.json
	- datasplit_nih-ms-mp2rage_seed50.json
	- datasplit_lumbar-vanderbilt_seed50.json
	- datasplit_dcm-zurich-lesions_seed50.json
	- datasplit_sci-zurich_seed50.json


SPLITS ACROSS DIFFERENT CONTRASTS (n=11):

| Contrast   |   #train_images |   #validation_images |   #test_images |
|:-----------|----------------:|---------------------:|---------------:|
| dwi        |             176 |                   37 |             50 |
| mp2rage    |             142 |                   16 |             18 |
| mt-off     |             158 |                   36 |             49 |
| mt-on      |             167 |                   37 |             50 |
| psir       |             217 |                   49 |             67 |
| stir       |              56 |                   15 |             18 |
| t1map      |              71 |                    8 |              9 |
| t1w        |             248 |                   45 |             63 |
| t2star     |             653 |                  102 |            111 |
| t2w        |            1434 |                  226 |            254 |
| unit1      |             152 |                   19 |             20 |
| TOTAL      |            3474 |                  590 |            709 |


PATHOLOGY-WISE SPLIT:

| Pathology           |   Number of Subjects |
|:--------------------|---------------------:|
| MS                  |                  894 |
| HC                  |                  407 |
| MildCompression     |                   57 |
| DCM                 |                  239 |
| MildCompression/DCM |                   88 |
| SCI                 |                   19 |
| ALS                 |                   32 |
| NMO                 |                   19 |
| TOTAL               |                 1755 |

[naga-karthik]

Datasets that could be used for evaluation:

• whole-spine
• lumbar-vanderbilt sagittal T2w images
• dcm-oklahoma (maybe)
• EPI datasets
• ms-basel-2020 (Proton density PD images)
• qc_coil_7t_sc_seg (shared on Slack by Julien)