Sets `metadata/subject/id` to the six-digit patient ID in the 164 in-vivo files in `data/Fundamental/` and `data/Harmonic/`. The files currently store `<subject_id>_<view_number>` (e.g. `118420_1`, `118420_2`), which is 82 distinct IDs for the README's 30 patients. Grouping by `subject/id` therefore treats every view as a different patient, and a subject-wise split leaks the same patient across folds.
Source: the README ("Each dataset follows the naming convention of `subject_id`_`view_number`_`sequence_type`. `subject_id` is a six-digit random number. `view_number` distinguishes between multiple views acquired on the same subject."). The script checked that every stored ID is exactly the `<six digits>_<view>` prefix of its filename; stripping the view leaves 30 IDs, the README's patient count. The view number stays in the filename. `CIRS_validation/CIRS_fundamental.hdf5` (`id = "CIRS"`) is unchanged.
Files changed: 164. Only `metadata/subject/id` was written, `raw_data` and all othe…
24 days ago
vanderbilt: add anatomy annotation, and mark the CIRS scan as a phantom (#17)
- vanderbilt: add anatomy annotation, and mark the CIRS scan as a phantom (775816c1a94cd5468d424b0d4552c81d01fa14c5)
Co-authored-by: Oisín Nolan <oisinn@users.noreply.huggingface.co>
